Progress, in the open.

Flagship run log

Each automated lab run designs or extends one flagship.

1. 2026-11-02 08:50 · Phase 2 R71 CricketOne

   Powder out of the slanted Hall holes: sim/holes-r71.mjs 6/6; bureau guides read; Ø2.5 chosen (1.14 mm to the lock hole, mouth in the pocket; Ø3 fails); no chamfer, clear with air, a Ø2 hand drill and a Ø1 wire; three tops re-made, maker files 45/45, R70 7/7; FIRST-STUMP.md section 7 re-upload head-top; R72 added.

2. 2026-11-01 11:05 · Phase 2 R27 CricketOne

   Live view on both boards: live.py through camera.py (halved preview, OpenCV JPEG, ball box, focus number, frame rate from stamps), sim/live-r27.mjs 20/20; fixed the Radxa cutting a 10 ms exposure to 8.30 ms (blanking now set first); sim ball true size; R28 added; no part change.

3. 2026-11-01 10:05 · Phase 2 R25 CricketOne

   A fresh Radxa OS step by step: sim/fresh-radxa-r25.mjs 23/23 against Radxa's headers package, image recipe, rsetup and Bookworm; fixed the driver build stopping on warnings (CONFIG_WERROR=y, gcc 10 vs 12), the missing python3-cryptography, and the overlay folder lookup; R26 added; no part change.

4. 2026-11-01 09:30 · Phase 2 R24 CricketOne

   Radxa camera overlay on the ZERO 3W's 5.10 kernel: sim/overlay-r24.mjs 20/20 against both real 5.10 boot files; all 12 labels present, no pin clash, route as Rockchip's board; CSI-2 host lane count matched; radxa_day1.sh sensor name fixed (m00_b_ov9281 2-0060); R25 added; no part change.

5. 2026-11-01 05:48 · Phase 2 R17 CricketOne

   Heat bar to the bore in the P1 CAD: the bar is the 30 mm rod with three flats (16 x 11.5 x 80, 37.2 g), the TP-3 strip under Kapton squeezed 17.8 % on the bore over 1341 mm², a 16.6 x 81 window in case-top (new STL), sled-top unchanged; fit, clash, bore and squeeze checks pass; finding: a ball hit loads the bar's seats about 35 % more (est.), R18 added; no cart change

   Folded none.

6. 2026-11-01 05:22 · Phase 2 R16 CricketOne

   Heat in the first stump: sim/thermal-revr.mjs 10/10 (Pi Zero 2 W under the heat bar in the TPU soft case and blue G-10 tube); as drawn the Pi fails the R8 soak (80.6 °C at 20 °C air); the bar's round face on a TP-3 strip through a sleeve window to the bore gives 54.1 °C at 25 °C (chosen, nothing to buy); cells limit ordinary blue to about 29 °C air in the sun, cool blue about 36 °C; R17 (the CAD change) added; no cart change

   Folded none.

7. 2026-11-01 04:58 · Phase 2 R15 CricketOne

   Lens calibration for the first camera: software/lens_cal.py (guided 30 poses, five-term fit, held-out check, lens.json), sim/lens-cal-r15.mjs 13/13; stand procedure 0.047 px in 95 % after re-aim, by hand 0.16 px; setup.sh adds python3-opencv; no order change

   Folded none

8. 2026-11-01 04:19 · Phase 2 R14 CricketOne

   Day-one set-up and preflight: setup.sh installs all 11 outside libraries and turns on I2C and SPI, gpio_edges.py for libgpiod v1/v2, preflight.py, sim/preflight-r14.mjs 24/24; no order change

   Folded none

9. 2026-10-31 23:25 · Phase 2 Q12 CricketOne

   B1 for the rev Q first stump: 8-bit mode (26 % frame-rate margin), raw camera test b1_raw.sh, camera hole traced; the camera moved forward in the P1 CAD so the Ø10 hole passes the whole view (it cut 89 % before); GUIDE.md B1 rewritten; order 1 unchanged, sled-top re-cut for order 2. Folds none.

10. 2026-10-31 22:55 · Phase 2 Q11

    CricketOne 3D game view prototype: sim/game-view.mjs, five views and the stitched end view, data sized (7.1 kB a ball, 9 MB end view), sample ball for the app, 29/29 checks; no order change

    Folded none

11. 2026-10-31 20:25 · Phase 2 rev R1 CricketOne

    Bench software for order 1: setup.sh, b1_bench.py (fps and drops from sensor time stamps, 720p clip from every 4th frame), live.py browser view, track2d.py ball finder; sim/track2d.mjs: 100 % found and 0.048 px rms in good light, 99 % and 0.457 px in dull light; no order changes

    Folded none

12. 2026-10-31 19:25 · Phase 2 Q9 CricketOne

    Production ash stump: bored English ash from two glued halves turned on the bore, the bore filled solid (a hollow wall splits across the grain), stainless window cup, 12 clear light inserts, radio fine through wet wood, 324 g shell; CAD section row Q9; first-stump parts list unchanged

    Folds none.

13. 2026-10-30 21:45 · Phase 2 P35 → CricketOne umpire band answer card and battery

    products/cricketone/sim/band.mjs (band-results.json), sources B1-B7; 200,000 simulated taps per hand (est. spreads): three live keys on the 16 mm pitch 3.114 % wrong with a cold hand in a thin glove, 4.959 % in a winter glove; design card FAIR left, asked call right (32 mm apart), OTHER middle: 0 wrong in any hand, only "tap again" (17.373 % cold thin glove, 24.566 % winter glove); own buzz per answer (FAIR one long, NO-BALL two, WIDE four, question three short); battery 88 mAh a 8.42 h 50-over day at the Seeed 10 mA BLE light sleep, 74 % left on an old cell on a cold morning (32 % at 27 mA); link to both middle stumps 99.95 % of questions in 1.5 s (brain's stump alone 97.685 %, 1.56 drops an hour); band out: scorer card "band not connected", stored taps sent later; one band is at the bowler's end only half the overs: 99.22 % before the next ball (99.52 % with one per umpire, owner decides); engineering.json "bandcard", review.mjs 68/68, DESIGN.md section and open item 35, integration.json band, CAD key names and power table, page FAQ; new B1.26 and T25; no part change; adds P36. Schedule on time (P1-P5 due 1 Nov); audit: last close-out row C11 OK (check-designs). Next: P36 band caps for gloves.

14. 2026-10-30 20:50 · Phase 2 P33 CricketOne

    Front foot wide of the crease on the line: sim/front-foot-wide-line.mjs adds the far fixed errors and paint wobble to the truth (P27 method 60.4 % sure, ± 62.3 mm, 1 wrong sure call); the outer camera places a grounded heel on the ground itself, the far stumps check the gap under it and a no-ball must be sure on the crossing alone: ± 12 mm at 95 %, sure on 92.1 % at 3.5 sd with 0 wrong sure calls; over count 0.67 answers an over (was 0.89); no CAD change; adds P34 (answering in the app)

    Folded none.

15. 2026-10-30 20:17 · Phase 2 P32 CricketOne

    Back foot on the line: sim/back-foot-line.mjs adds paint wobble to the truth; at 20 m the fixed errors rule, so the ground contact along the whole sole, the inner edge at the foot and a sole model learned from earlier landings (with staggered far shutters and the 360 fps strip) give ± 18 mm at 95 % (was ± 24.4), sure on 95.73 % of seen landings at 3.5 sd with 0 wrong sure calls; near stumps see no back feet; over count 0.89 answers an over (was 1.07); no CAD change; adds P33

    Folded none.

16. 2026-10-30 19:45 · Phase 2 P31 CricketOne

    Front foot on the line: sim/front-foot-line.mjs adds paint wobble and heel braking to the truth; staggered shutters, a 360 fps stride window, a heel-strike model and the crease edge at the heel give ± 2.6 mm at 95 % (was ± 9.9), sure on 98.21 % of near landings at 3 sd with 0 wrong sure calls on fresh landings; over count 1.07 answers an over (was 1.38); no CAD change; adds P32 (back foot on the line)

    Folded none.

17. 2026-10-30 19:20 · Phase 2 P30 CricketOne

    Leg-side wide by reach (MCC Law 22): sim/leg-reach.mjs, the leg-stump line called 81.1 % wrongly wide in a Law 22 match, so the app's match setup chooses the rule; pose reach from both positions sure on 86 % at 3.5 sd with no wrong sure calls; 1.36 answers an over in a Law 22 match; no CAD change; adds P31

    Folds none.

18. 2026-10-30 18:55 · Phase 2 P29 CricketOne

    Turn angle: sim/turn-angle.mjs weighs the front foot wide of the crease against run-outs and stumpings on one scale (answers a 40-over club match, est. mix): 12° fewest (112.2; straight 135.8, 10° 112.9), 9° ruled out by one wrong sure call, far stumps give no third view of the crossing (about 870 mm along the pitch per look); 12° kept, adds P30

    Folds none.

19. 2026-10-30 14:52 · Phase 2 P22 CricketOne

    Run-out and stumped check: sim/run-out.mjs shows the last frame before bail-off left the bat 35.1 mm out at 95 % (a sliding bat moves 17-44 mm a frame); a track fit from the three staggered stump cameras times crossing and touchdown against the bail: ± 6.5 mm for run-outs and ± 2.8 mm for stumpings at 95 %, no wrong sure calls with two or more cameras; one-camera views are unsure

    Folds none.

20. 2026-10-30 13:50 · Phase 2 P20 CricketOne

    Off-side wide check: sim/wide.mjs shows the ±50 mm rating was far too cautious; the post-bounce track from the six stump cameras, followed 1 m past the stumps, puts the ball at the popping crease within 1.6 mm at 95 % and 2.6 mm worst (simulated), no wrong sure calls; close wides now turn on the striker's movement and touches; folds none

21. 2026-10-30 13:30 · Phase 2 P19 CricketOne

    Front-foot no-ball check: sim/front-foot.mjs shows the ±15 mm rating read the stopped foot (9.1 % wrong calls, a fair landing slides over); a landing fit from two stump cameras gives ± 10.7 mm at 95 % with no wrong sure calls; wide-of-crease landings are a gap

    Folds none.

22. 2026-10-30 13:00 · Phase 2 P18 CricketOne

    Bail-off timing: sim/bail-off.mjs shows the Hall switch as drawn logs the wicket 10.75 ms late (median, 39.568 ms at 95 %); a linear Hall sensor with a window ADC per stump logs it within 0.278 ms; parts and CAD names updated

    Folds none.

23. 2026-10-30 10:12 · Phase 2 P13 CricketOne

    Lens and window view: sim/lens-view.mjs traces the 70° x 46° view through the lining hole, glass and pocket; the P5 lens position passed only about 23° across, so the lens is brought up to the lining (14 mm from the axis, needs 12.3) in the CAD with a check; product lens about 2.7 mm (stock 78.5° lens gives 14 % fewer pixels on the ball); open item 10 closed

    Folds none.

24. 2026-10-30 09:47 · Phase 2 P12 CricketOne

    Custom stump board cost study, sim/board-cost.mjs from public prices: the ZERO 3W with our long thin carrier and a bare camera saves about $105 a kit at 100 kits and keeps the radio grant; all custom saves less and needs new radio tests; open item 7 costed

    Folds none.

25. 2026-10-30 09:17 · Phase 2 P11 CricketOne

    Keeper's gloves view check in the CAD (Law 27.3.1, 2026 Laws): 6 % of glove positions seen from the striker's end, far-end depth ±580 mm, so the call is a gap the umpire keeps; open item 8 closed

    Folds none.

26. 2026-10-30 08:44 · Phase 2 P10 CricketOne

    Phase 2 review: sim/review.mjs re-checks 14 engineering.json numbers against the sims (one corrected: worst sure call 15 mm at 180 fps), radio slot in the CAD lining, heat re-run (+0.1 °C at most, rating 38 °C unchanged); phase 2 on paper complete

    Folds none.

27. 2026-10-30 08:40 · Phase 2 P9 → CricketOne phase 3 review kit

    Done: products/BUYER-REVIEW.md "## CricketOne phase 3" (sources R1-R8): eight topics from the engineering.json limits, ten conversations with five kinds of role, questions per group, a five-question survey and what each answer changes; DESIGN.md "### Phase 3 review kit"; product page tiers fixed to $1,990 and $1,490 (were the rev O $1,690 and $1,290) and a new FAQ; no review held, the owner decides when it starts. Folds none.

28. 2026-10-30 07:19 · P7 CricketOne

    Light and low sun: sim/light.mjs; red ball on grass needs about 13,600 lx at 0.1 ms unfiltered, 5,700 with a 600-680 nm filter (B1.11 decides); club floodlights leave LBW unsure; low sun blinds one end only on east-west pitches; exposure-based unsure rule; engineering.json "light" question. Folds none.

29. 2026-10-30 05:58 · P5 CricketOne

    Seals in the CAD: glass fibre tube Ø38 x 2.5 (bore 33), 0.5 mm 6061 lining z 400-688, AS568-122 O-ring gland on the spike plug (18 % squeeze, 1.6 % stretch, checked in the script), 1.5 mm toughened glass window 1 mm below the round face on the lining ledge, no light slot, pads and sounder on the sides, Gore VE80308 vent under the crown; Law 8, fit checks and assembly pass; new: lens too far behind the aperture for its 70° view, added to B1

    Folded none.

30. 2026-10-30 04:56 · P3 CricketOne

    Heat model sim/thermal.mjs: lumped network, 8 h of play, 36 cases; rev P as drawn reaches 88 °C SoC at 40 °C still air (wall spreads heat only about 8 mm), so a 0.5 mm aluminium lining is added (CAD in P5); every call to 38 °C still air and 45 °C in a breeze (est.); white kept; bench step B1.10 added

    Folded none.

31. 2026-10-30 04:45 · P2 CricketOne

    B1 camera and driver plan: OV9281 on the Radxa ZERO 3W needs our own driver patch and overlay (others run it at about 120 fps); 1280 x 720 at 180 fps from the mainline driver replaces the cropped 200 fps band after a re-run simulation (47 % sure, 6.4 / 8.5 mm); Pi Zero 2 W fallback about 7x slower pose (est.); nine bench steps with pass marks in DESIGN.md "### B1 plan"; engineering.json updated

    Folded none.

32. 2026-10-30 04:10 · P1 CricketOne

    Ball-tracking simulation sim/track-sim.mjs: 200 deliveries seen by the six stump cameras at their CAD positions, 11-number flight fit with its own uncertainty as the "sure" gate; sure calls within 13 mm at the stumps, 51 % sure, run-outs within 3 mm (95 %), camera aim must hold 0.01°; DESIGN.md results table and engineering.json simulated numbers

    Folded none.

33. 2026-10-30 03:17 · Backlog 4

    CareOne market and rules check (care homes and hospitals flagship, owner order F12): NHS England's 7 Aug 2026 response to a coroner's Regulation 28 report (bed and chair sensor alarms only an early warning, used after a multidisciplinary risk assessment written in the care plan; NICE NG249 notes limited evidence for movement sensors), so bed exit, not back and floor now switch on per bed from the recorded falls plan and its review date, and a new assist call flags a night pattern change for the falls plan review at handover (9 calls, 5 alone, 4 assist); REFINE trial (Age and Ageing 2014, 1,839 patients, rate ratio 0.90, not cost-effective) so no fall-prevention claim and an FAQ says so; CMS staffing minimums repealed (in effect 2 Feb 2026, facility assessment kept) so a signed night workload summary exports for 42 CFR 483.71; Sensio RoomMate added to competitors; no CAD change, price kept at $590 + $15/mo; audit C10 OK; on time; next FarmOne (backlog) until C11 at 31 Oct 18:00.

34. 2026-10-30 02:47 · Backlog 4

    ChillOne market and rules check (food, cold chain and pharma flagship, owner order F11): EPA AIM Act leak repair 40 CFR 84.106 in force 1 Jan 2026 for 15 lb or more of HFC (leak rate at every top-up, 20 % commercial line, repair within 30 days, three years of records), so a new assist call keeps a signed refrigerant top-up record with the leak rate, 30-day countdown and the coil and air curve from before (9 calls, 6 alone; a record aid, never leak detection); FSMA 204 not enforced before 20 Jul 2028 (out of scope, rule line); Kitchen Brains Modularm 75LC $438.49 added to competitors; no CAD change, price kept at $690 + $29/mo; audit C10 OK; on time; next CareOne (backlog) until C11 at 31 Oct 18:00.

35. 2026-10-30 02:25 · Backlog 4 → deeper market check (cellone)

    Done: DESIGN.md "## Market and rules check (30 Oct 2026)" with the EU Machinery Regulation 2023/1230 Annex III 1.1.9 and 1.2.1 (from 20 Jan 2027), Cyber Resilience Act Article 14 reporting (since 11 Sep 2026), the RED cybersecurity act and EN 18031, and the $107.70 Advantech WISE-S100 (sources M1-M5: UL, Jones Day, European Commission, BSI, Advantech); signed software version and five-year change record, 24-hour reporting plan, 2 new rule lines and the radio line updated, security entries, 2 FAQs and a records spec, lab-log entry; no CAD change; audit C10 OK; on time; next C11 from 31 Oct 18:00, backlog until then (ChillOne check next)

36. 2026-10-30 01:46 · Backlog 4

    DockOne market and rules check (logistics flagship, owner order F9): OSHA renewed its warehousing NEP (CPL 03-00-026) on 31 July 2026 for five years with loading docks and forklift-pedestrian interactions in focus, so a new assist call gives a signed monthly dock safety record per door from the calls already signed (gaps shown as gaps; a record aid, no compliance claim); detention billed after about two free hours (ATRI $63.71/h, no federal rule) so the dwell call's signed door times export per carrier (door time only; gate and yard stay with the yard log); ANSI MH30.3-2022 unchanged; no CAD change, 10 calls, price kept; audit C10 OK; schedule on time; next C11 not before 31 Oct 18:00, backlog until then (CellOne check next)

37. 2026-10-30 01:07 · Backlog 4

    RoomOne market and rules check (facilities and hospitality flagship, owner order F8): 2026 reports of sensor-only smoking fees reversed ($300-566; hair spray and heat tools, guests out of the room; The Points Guy Feb 2026, Checkbook June 2026), FTC fees rule in force 12 May 2025, Rest and Wynd room sensors; the fee assist now sets the air curve beside the radar presence log (room empty: do not charge; humidity rising: styling or steam likely; room check before any fee), guest notice on the e-paper and wall plate, 2 rule lines in equipment.json, 2 page FAQs and a spec, demo app line; no CAD change, 8 calls, price kept; audit C10 OK; schedule on time; next C11 not before 31 Oct 18:00, backlog until then (DockOne check next)

38. 2026-10-30 00:45 · C10

    Stage 2 product line page: src/pages/line.astro (/line) built from the plan, product pages, DESIGN.md reasons to buy and equipment.json calls for all 24 products, with the four-phase plan; header and home links; audit C9 OK; schedule on time; next C11 from 31 Oct 18:00 (backlog until then).

39. 2026-10-30 00:30 · C9

    First-product scoring: all 24 stage 2 flagships scored on six points in products/FIRST-PRODUCT.md (owner only, not on the site), with a recommended first product and a runner-up for the owner to decide; audit C8 OK; schedule on time; next C10, the /line page.

40. 2026-10-29 23:41 · Backlog 4

    AisleOne market and rules check (first industry flagship, owner order F7): NIST Handbook 130 Section V price verification (a store passes at 98 % of sampled prices matching the till; a guide, state law decides) so a new assist call gives a weekly price-accuracy figure and fix list per aisle from the label reads already taken (below 85 % read, list only; never sets a price; no CAD change, 9 calls); EU AI Act Article 5(1)(f) bans emotion reading at work since 2 Feb 2025, so staff are masked like shoppers and never timed or scored per person; Walmart's cameras now call low shelves and spills (Oct 2026 report) but shelf cameras stay chain contracts; price kept at $920 + $15/mo; phase 3 question 3 updated

41. 2026-10-29 23:19 · Backlog 4

    ClassOne market and rules check (second kids learning flagship, owner order K2): California PUC 1625 classroom CO2 rule (3 to 6 ft, 1,100 ppm alert, peak kept, ±75 ppm at 1,000 ppm) failed by the rev A SCD41 (±90 ppm, 2.0 m up), so the air sensor is now a Sensirion SCD30 (±60 ppm) in a pocket in the left side frame at 4.5 ft, 4.7 m from the door (CAD asserts both, fit and assembly clean), a California 1,100 ppm setting, parts $729 (+$9), price kept; BB101 unchanged; face-recognition attendance cameras are the opposite of ours; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00)

42. 2026-10-29 22:55 · Backlog 4

    StudyOne market and rules check (first kids learning flagship, owner order K1): the amended COPPA Rule in full force from 22 April 2026 (separate consent before disclosure to a third party, 16 CFR 312.10 written retention policy, 312.2 biometric identifiers), so the opt-in cloud homework help gets its own second consent, a written retention policy (unsure-answer photos 7 days, log and summaries 12 months, est.) and no voice enrolment or face template; EU AI Act Article 5(1)(f) emotion ban and Annex III 3(b) moved to 2 December 2027 by the 2026 Digital Omnibus deal, so home-use marking only and the focus timer never reads faces or feelings; 6AM Vicam camera lamps and LumioClaw added as competitors, none marks on the device, price kept at $279 + $4/mo; 2 rule lines in equipment.json, page FAQ and spec; no CAD change; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00)

43. 2026-10-29 22:14 · Backlog 4

    GarageOne market and rules check (fifth home flagship, owner order H5): California Health and Safety Code 19891 (SB 969, battery backup on every opener since 1 July 2019, $1,000 per opener) so the opener drive is bought with its backup battery and a new assist call reads that battery on a spare MCP3008 channel (no CAD change, 9 calls); 16 CFR 1211 took the 2023 UL 325 revisions on 13 May 2024 (eye height from the surface below the door, 30,000-cycle test) so G-T4 goes to 30,000 cycles; NHTSA keyless rule and PARK IT Act still not made, engine call kept; Chamberlain December 2025 line (B6753T about $740), controllers $70-90, none decides, price kept at $499 + $2/mo; page FAQ and spec; bom.csv opener row quoted; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00)

44. 2026-10-29 21:46 · Backlog 4

    PoolOne market and rules check (fourth home flagship, owner order H4): AS 1926.1:2024 called up by NCC 2025 (Victoria and the ACT 1 May 2026; NSW, Queensland, South Australia and Tasmania 1 May 2027; WA keeps 2012 for private pools; boundary barriers keep 500 mm), so the gate's own latch and hinges are bought marked to the 2024 edition and the climbable-strip assist now also notes plants in the 900 mm zone (no CAD change, 7 calls); ISPSC 2027 reorganises 305.3/305.4, 2024 gate rules kept and the final text re-checked before US sale; Trident Safe Signal latch (about $237) added as the closest alarm latch, price kept; 2 rule lines, page FAQ and spec; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00)

45. 2026-10-29 21:16 · Backlog 4

    DampOne market and rules check (third home flagship, owner order H3): Awaab's Law phase 2 from 30 Nov 2026 adds excess cold, falls and five more hazards for social landlords (10 working days to investigate), so a new assist call keeps the bathroom's hours below 18 °C (WHO guideline) in the signed record from the SHT45 and wall heat camera already inside (no CAD change, 8 calls); Approved Document F 2026 edition (in force 24 Mar 2027) makes continuous extract the new-home minimum, so the installer setting locks the 8 l/s trickle on; Decent Homes Standard damp rule from 2035 for private rentals; BS 7671 A4:2026 with no bathroom change found; Blauberg Smart Silent WiFi £220.82 and OREiN OL05S Matter fan (about $266-300) added as competitors; 2 rule lines in equipment.json; page FAQ and 2 spec lines; price kept; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00)

46. 2026-10-29 20:53 · Backlog 4

    BedsideOne market and rules check (second home flagship, owner order H2): 16 CFR 1261 unchanged since 1 Sep 2023 (ASTM F2057-23, open-shelf table stays outside it); FDA 21 CFR 890.5050 amended 4 Dec 2025 for the new Quality Management System Regulation (in force 2 Feb 2026), so a complaint file and records under 820.35 before US sale; NHS Highland home oxygen guidance (concentrators stop, cylinder 10-20 h, priority register, never candles) so in a power cut the lamp now stays on dim as a battery lamp (about 24 h, est.) and the alert says no candles; new assist call concentrator not back on after the power returns (mains sense + mic, no CAD change, 10 calls); Walabot HOME $149.99 + $4.99/mo, NtelCare and Creatrol facility radars, Hero unchanged; 2 rule lines in equipment.json; page FAQ and spec; price kept; repaired lost dollar signs in the KitchenOne lab-log entry and run log line; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00)

47. 2026-10-29 20:22 · Backlog 4

    KitchenOne market and rules check (first home flagship, owner order H1): India BIS licence for range hoods (Kitchen QCO 2023, IS 302-2-31) and the 2026 household order, UL 858 2026 oil-ignition test, LPG leak guidance (no switching fans or lights) so on a gas alarm with no flame the hood now holds its fan and lamps; Faber Hood Enigma switches its fan on with its own LPG sensor (we chose not to), Crompton AirIQ, iGuardStove radar ($399, closest), Xiaomi gas hob 3 Pro; new assist call kitchen not used by the usual time (14th radar view, 14/14 in view); 2 rule lines in equipment.json; page FAQ and spec; price kept; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00), else backlog 4 on BedsideOne

48. 2026-10-29 19:36 · Backlog 4

    AthleticsOne market and rules check: World Athletics 2026 rules and amendments, WMA 2026 and the four NFHS 2027 changes leave the finish and the start unchanged; high school states count a clock started by the gun's sound as hand timing (IHSA; WIAA +0.24 s), so every time is now labelled by its start source and never as FAT, with the state conversion as a separate column; ties under 0.01 s stay dead heats for the referee (NFHS 3-9-6); new phone photo-finish apps (SprintLens, Photo Finish: Automatic Timing) added as training competitors, FinishLynx Scholastic $5,375-5,699; price kept; page FAQ and spec; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00), backlog until then

49. 2026-10-29 19:20 · Backlog 4

    DiamondOne market and rules check for NFHS 2026-27 and Little League 2026: new double first base option for high schools (NFHS 1-2-9, mandatory 1 Jan 2027; 30 x 15 in tray, four more foot sensors under the coloured half, 16/16 fit and a 13th camera view in the CAD), force-out and fair-or-foul calls by half, new which-half assist (8-2-2), 18 in first base (July 2026) on the standard tray, NFHS 1-6-3 worn-device rule met; Musco Umpire Assist still has no price; price kept; page FAQ and spec; repaired lost dollar signs in the RacketOne and DockOne lab-log entries; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00)

50. 2026-10-29 18:50 · Backlog 4

    RacketOne market and rules check for USA Pickleball 2026 and ITF 2026: net post rules unchanged; ball-on-post call now checks a bounce in first (10.C.5), out calls timed before the next hit (8.F.3), 2026 rally scoring and 7-point weather games, score fixes after the rally (6.F), doubles alley serve and tie-break change-ends reminder (ITF 18, 27); PlayReplay and Owl AI are pro-event systems with no club price, price kept; page FAQ and spec; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00)

51. 2026-10-29 18:15 · Backlog 4

    CourtOne market and rules check for FIBA 2026 (in force 1 Oct 2026) and NFHS 2026-27: equipment rules unchanged, goaltending and interference clips kept after a foul for the new replay rule, 3-point line placed from court geometry (line colours now free), new 0.3 s late throw-in assist (NFHS 5-9-5), no new competitor that makes calls, price kept; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00)

52. 2026-10-29 17:34 · Backlog 2

    Drawn idea plan for airmate-ir (a flat with a split AC in the bedroom, living room and home office, one AirMate pin under each and the phone app) replaces the last fallback concept card; product page and make-concepts now read products/<page>/idea.json when no design folder names the page; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00), backlog 4 until then

53. 2026-10-29 17:17 · Backlog 4

    FootballOne market and rules check for the IFAB 2026/27 Laws: DESIGN.md section with every change and what it changes, new goal kick 5-second countdown call (alone with no referee, 4 new CAD view targets, 14 pass), page FAQ and calls spec, no new grassroots competitor that makes calls, price kept; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00), backlog until then (airmate-ir needs a design folder before its idea plan)

54. 2026-10-29 16:35 · Backlog 4

    CricketOne market and rules check: MCC 2026 Laws (in force 1 Oct 2026, Law 8 unchanged) mapped to every call; run-out and stumping now check a direct hit or the ball held with complete control (unsure when not seen), hit wicket follows the batter until balance is regained; StumpEye (club stump camera, video only) and V6 Sports DRS added as competitors, price kept; two page FAQs; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00), until then the keeper-gloves view for CricketOne or a market check on the next flagship

55. 2026-10-29 16:40 · Backlog 2

    WasteSort page brought in line with its rev A clip-on coach design (name, specs, $299 + $19/station/mo, tiers, FAQs) and a drawn idea plan of an office floor replaces the fallback concept card; airmate-ir is the last catalogue product without a plan (no design folder yet); audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00), backlog until then

56. 2026-10-29 15:55 · Backlog 2

    Drawn idea plans for wardpole and waterleak-ai replace the fallback concept cards (wastesort-ai waits until its page matches the clip-on coach design); audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00), backlog until then

57. 2026-10-29 15:10 · Backlog 2

    Drawn idea plans for switchroom and traptop replace the fallback concept cards; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00), backlog until then

58. 2026-10-29 14:45 · Backlog 2

    Drawn idea plans for smokesentinel-ai and storesense replace the fallback concept cards; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00), backlog until then

59. 2026-10-29 14:25 · Backlog 2

    Drawn idea plans for safeyard and smartline-ai replace the fallback concept cards; audit C8 OK; schedule on time; next C9 first-product recommendation (not before 30 Oct 00:00), backlog until then

60. 2026-10-29 13:55 · C8 Buyer review prep: industries

    products/BUYER-REVIEW.md "## Industries": extra industry review rules, then aisleone, roomone, dockone, cellone, chillone, careone, farmone, pumpone, siteone, bayone, roadone each with who to ask (with sources), three open questions, five-question survey and what would change

    audit C7 OK

61. 2026-10-29 13:25 · C7 Buyer review prep: kids learning

    products/BUYER-REVIEW.md "## Kids learning": extra kids review rules, four child-safety and privacy questions with sources, then studyone and classone each with who to ask (8), three open questions, five-question survey and what each answer would change; audit C6 OK; schedule on time; next C8 buyer review prep: industries

62. 2026-10-29 12:45 · C6 Buyer review prep: home

    products/BUYER-REVIEW.md "## Home": extra rules for home reviews, then kitchenone, bedsideone, dampone, poolone, garageone each with who to ask (8: payer, the person who lives with it, installer or fitter, with sources), the three open questions from DESIGN.md "## Stage 2", a five-question survey and what the answers would change; shared themes: replace or add a kit, who hears first, price. Audit C5 OK; schedule on time. Next: C7 buyer review prep: kids learning.

63. 2026-10-29 12:22 · C5 Buyer review prep: sports

    products/BUYER-REVIEW.md "## Sports": review rules for every group, then cricketone, footballone, courtone, racketone, diamondone, athleticsone each with who to ask (8 kinds of club, league, school and official, never people, with sources), the three open questions from DESIGN.md "## Stage 2", a five-question survey and what the answers would change; shared themes: league approval, price, the real day. Audit C4 OK; schedule on time. Next: C6 buyer review prep: home.

64. 2026-10-29 11:55 · Backlog 2

    Drawn idea plans for raidpost and rubdock replace the fallback concept cards; audit C4 OK; schedule on time; next C5 buyer review prep: sports (not before 12:00)

65. 2026-10-29 11:20 · Backlog 2

    Drawn idea plans for puretrack and qualitycheck-ai replace the fallback concept cards; audit C4 OK; schedule on time; next C5 buyer review prep: sports (not before 12:00)

66. 2026-10-29 07:22 · C4 Line quality pass

    products/LINE-REVIEW.md "## Quality pass": check-designs over 127 designs and 1,071 parts now 0 errors and 0 warnings, check-assembly 127/127 and 24/24 flagships clean; six floating parts on older designs fixed (fieldside band on the forearm, heatflag two clamp sets and bay mast pad drawn, linecell and qualitycheck-ai ring-light standoffs drawn, schoolmeet TapGrip and vapesense-ai test hood marked hand-held with reasons); REQUIRE_VERIFIED = true in scripts/check-designs.mjs; audit C3 OK; schedule on time; next C5 buyer review prep: sports (not before 29 Oct 12:00), backlog until then

67. 2026-10-29 06:47 · C3 One pricing table

    products/LINE-REVIEW.md "## Pricing": price, plan, cost qty 1 and qty 100, margin and today's price for all 24; look-alike pairs side by side; no price changed; cricketone, footballone, racketone, careone costs and margins updated for the $6.99 cell; cricketone, footballone, careone site.json parts totals fixed; next C4 line quality pass

68. 2026-10-29 06:25 · C2 Overlaps and gaps

    products/LINE-REVIEW.md "## Overlaps and gaps": 24 flagships compared in pairs on buyer, place, item and calls; no overlap, none folded; FAQs on studyone, classone, bedsideone, careone, dampone, roomone; athleticsone flagship label fixed; places without a product listed (rugby and hockey, volleyball and badminton, swimming, golf, children's rooms, offices, airfields, forklifts, vehicles, gyms); next C3 pricing

69. 2026-10-29 05:55 · C1 Shared parts across the line

    products/LINE-REVIEW.md "## Shared parts": all 24 BOMs, buy.json and integration.json grouped by job; one part per job kept wherever the jobs allow (Pi Zero 2 W, Pi 5 2 GB, XIAO ESP32-S3, Camera Module 3 Wide, AI Camera, MLX90640, LD2450, Walter, 18650-35E, PowerBoost 1000C, ATECC608, D2F-01, one IEMI app), the few exceptions kept with a job reason; 35E price unified at $6.99 in 4 BOMs and 3 buy.json files, DockOne door camera part number fixed. Audit S24 OK; schedule on time. Next: C2 overlaps and gaps.

70. 2026-10-29 05:06 · S24 Public sector, roads and airfields

    RoadOne stage 2 refinement: buyer check (City of Tulsa head bid $295, distributor list $285, Florida DOT planning $800, Washington State Miovision contract licences $3,000-6,500 per intersection and $195-750 a year per service, Iteris Corona contract), price kept at $2,990 + $59/mo; brain the same-size Pi 5 2 GB ($90); Camera Module 3 Wide $38.50, ATECC608 $4.95, piezo $1.50 checked, IRM-20-5 and G5V-1 aligned: $749 parts qty 1 ($405 ours), cost about $989 / $730 at qty 100, margin 67-76 % (est.); 21/21 fits, 15 views, check-assembly clean; power cut about 12.6 h; page top-three FAQs (why not existing detection, cost, fits my poles and controller) and power and size spec; 3 phase 3 questions. Stage 2 line ready (S1–S24 all done), said on /build and the home page. Audit S23 OK; schedule on time. Next: C1 shared parts across the line.

71. 2026-10-29 04:50 · S23 Waste and recycling

    BayOne stage 2 refinement: buyer check (NWRA/RRS 2026 report, over 5,000 US recycling facility fires a year and catastrophic losses up 41 % in five years; Fire Rover March 2026 report, 448 reported fires in 2025, about $2.5 billion, batteries the leading cause; NWRA, ReMA and SWANA MRF battery guide Jan 2026; Deegee IPN/018 bay light £261.60 ex VAT and an 8 in dock light $315.90; Fire Rover $235,500 + $3,650/mo in Kent County minutes and about $200,000 + about $2,000/mo in Bourne; Dahua Eureka and Greyparrot on quote), price kept at $3,990 + $49/mo; Raspberry Pi 5 4 GB now $130 (BOM said $60), so the brain is the same-size 2 GB ($90, Adafruit 6007); Camera Module 3 Wide $38.50, ATECC608 $4.95, piezo $1.50, MLX90640 $74.95 checked, IRM-20-5 and G5V-1 aligned: $1,138 parts qty 1 ($464 ours), cost about $1,348 / $960 at qty 100, margin 66-76 % (est.); 21/21 fits, 15 views, check-assembly clean; power cut about 12.6 h; page top-three FAQs (why not a roof camera or suppression, cost, fit my light) and power and size spec; 3 phase 3 questions. Audit S22 OK; schedule on time. Next: S24 RoadOne.

72. 2026-10-29 04:15 · S22 Construction and mining

    SiteOne stage 2 refinement: buyer check (Coram 2026 guide: full-height turnstiles $4,600-14,350 per lane installed; hire £215 a week at Portable Space, €178.20 a week at Boels, $1,875 per 4 weeks at the Whiting-Turner hub; PPE detection now a feature of site camera services such as TrueLook, about $199-750 a month where priced (est.); OSHA heat NEP renewed 10 April 2026 to 2031, supplemental heat proposal planned Dec 2026, so rev A's "due to expire" was corrected), price kept at $8,990 + $99/mo; Raspberry Pi 5 4 GB now $130, so the brain is the 2 GB ($90, same outline, 12 views unchanged); Camera Module 3 Wide $38.50, ATECC608 $4.95, piezo $1.50, SPS30 $60.50 checked, PN532 and IRM-20-5 aligned to buy.json: $5,739 parts qty 1 ($511 ours), cost about $6,439 / $5,210 at qty 100, margin 28-42 % (est.); 23/23 fits, check-assembly clean; power cut about 19 h; page top-three FAQs (why not a turnstile and a PPE camera service, cost, hire or retrofit) and power and size spec; 3 phase 3 questions. Audit S21 OK; schedule on time. Next: S23 BayOne.

73. 2026-10-29 03:55 · S21 Water, energy and utilities

    PumpOne stage 2 refinement: buyer check (SJE Rhombus panels $1,000-1,244 at R.C. Worst, Mission service $575/yr per station in Kittitas 2026, Novi install about $3,000 a station, Sensaphone Sentinel $1,561.50 + plan, pump-health software), price kept at $2,690 + $45/mo; parts $1,446 qty 1 (Adafruit prices checked), margin 34-53 % (est.); 16/16 fits, 13 views, CAD review clean; page FAQs and power and size specs; 3 phase 3 questions; schedule on time; S20 audit OK.

74. 2026-10-29 03:30 · S20 Farms

    FarmOne stage 2 refinement: buyer check (JFC DT600FF trough £318 ex VAT at Tanks Direct, so about $420 not $300; Farmbot Trough Direct A$749 and cellular / satellite monitors A$1,011 / A$1,290 on the NSW DPI catalogue; Ranchbot Trough Direct $489 at its US launch plus a plan; Moocall $329 + $150/yr; no camera trough found), price $1,290 → $1,490 + $9/mo (qty-1 margin was about zero); Pi Zero 2 W $19.05, Camera Module 3 Wide $38.50, VL53L1X $22.95, Voltaic 6 W $35, ATECC608 $4.95, piezo $1.50 checked, D36V28F5 $23.95 and 35E $6.99 aligned: $1,107 parts qty 1 (page said $990), cost about $1,287 / $840 at qty 100, margin 14-44 % (est.); 23/23 fits, 21 views, check-assembly clean; power about 4 days with no sun holds; page top-three FAQs (why not a trough monitor and calving sensor, cost, replace my trough) and power spec; 3 phase 3 questions (incl. a retrofit cover). Repaired the garbled S18 log line. Audit S19 OK; schedule on time. Next: S21 PumpOne.

75. 2026-10-29 02:55 · S19 Care homes and hospitals

    CareOne stage 2 refinement: buyer check (AliMed pads and alarms, Ally Cares acoustic monitoring, Nobi and Vayyar Care, Medi-Plinth and Medguard lockers; new competitor class: acoustic monitoring), $221 parts qty 1 (page said $262), bom.csv locker row quoted, cost about $331 / $245 at qty 100, margin 44-58 % (est.), price kept at $590 + $15/mo; 7 radar views pass; page FAQs and power spec; 3 phase 3 questions. Audit S18 OK; schedule on time.

76. 2026-10-29 02:05 · S18 Food, cold chain and pharma

    ChillOne stage 2 refinement: buyer check (ThermoWorks ThermaData Wi-Fi walk-in kit $449, ComplianceMate $650 / $875 with one year, LoneStar $7.95 per sensor a month, VWR and Corepoint CDC-style data loggers $311.78-336.98; new competitor class door-ajar sensors, Modularm and Kelsius), price kept at $690 + $29/mo; Pi Zero 2 W $19.05, Camera Module 3 Wide $38.50, ATECC608 $4.95, piezo $1.50 checked: $429 parts qty 1 (site said about $420), cost about $539 / $405 at qty 100 with a per-unit calibration certificate, margin 22-41 % (est.); 11 views pass, power cut about 25 h on the 1200 mAh cell; page top-three FAQs (why not a kit or logger, cost, vaccine rules) and power spec; 3 phase 3 questions; check-assembly clean. Audit S17 OK; schedule on time. Next: S19 CareOne.

77. 2026-10-29 01:27 · S16 Logistics and warehouses

    DockOne stage 2 refinement: buyer check (Rite-Hite Rite-Vu dealer kits, GMR POWER CHOCK Auto, IDEALVu restraint camera, Roboflow dock safety and pallet counting on existing cameras, Zimark proof of load, WORKD March 2026 on dock vision spend; no public prices), price kept at $1,290 + $49/door/month; Raspberry Pi 5 4 GB now $130, so the brain is the 2 GB ($90, same outline, 16 views unchanged); Camera Module 3 Wide $38.50, piezo $1.50, ATECC608 $4.95 checked: $796 parts per door qty 1 (site said $760), cost about $896 / $665 at qty 100, margin 31-48 % (est.); bom.csv guard row had an unquoted comma, fixed; freezer-dock call 11.5 W with the heater pad, inside the 20 W supply; page top-three FAQs (why not light kits or camera AI, cost, does it stop the trailer) and power spec; 3 phase 3 questions; check-assembly clean. Audit S15 OK; schedule on time. Next: S17 CellOne.

78. 2026-10-29 00:46 · S15 Facilities and hospitality

    RoomOne stage 2 refinement: buyer check (Verdant VX $165 offered to IHG owners, networked occupancy thermostats expected in IHG rooms by end 2026 with a conflicting Nov 2025 date on another vendor page; Telkonet EcoTouch+ PIR on quote; Minut about $129-150 + subscription), price kept at $349 + $8/mo, separate sum $673 + $13/mo; prices checked on shop pages ($146 parts qty 1; Pololu D36V28F5 now $23.95), cost about $163 / $110 at qty 100, margin 53-68 % (est.); 7/7 radar views, check-assembly clean; top-three FAQs; 3 phase 3 questions. Audit S14 OK; schedule on time. Stage 2 ready.

79. 2026-10-29 00:26 · S14 Retail

    AisleOne stage 2 refinement: buyer check (gondola sections $150-350 per 4 ft in a March 2026 guide; Vusion Captana rail cameras with Carrefour Feb 2026 after Walmart EdgeSense; Simbe Tally 4.0 mid-2026, UL 3300, B&R Stores June 2026; icetana spill alerts on existing CCTV; no public price for any of them), price $690 → $920 + $15/mo (parts were above the old price); Raspberry Pi 5 4 GB now $130, so both canopies use the 2 GB ($90, same outline, 28 camera views unchanged); Camera Module 3 Wide $38.50, ATECC608 $4.95, piezo $1.50 checked: $704 parts qty 1, cost about $804 / $590 at qty 100, margin 13-36 % (est.); page top-three FAQs (why not shelf cameras or a robot, faces, cost) and power spec; 3 phase 3 questions; check-assembly clean. Audit S13 OK; schedule on time. Next: S15 RoomOne.

80. 2026-10-28 23:50 · S13 Kids learning: classroom

    ClassOne stage 2 refinement: buyer check (Time Timer 12 in $41.95, Yacker Tracker $156.50 at Schoolmasters, Aranet4 about $190-250, S+S classroom CO2 light from EUR 187, HALO 3C $713.99 at Staples, Promethean ActivPanel 10 75 in $2,695-3,638, Milesight AM107 classroom platforms), separate sum now about $1,150 + $5/mo; Raspberry Pi 5 4 GB found at $130 (Adafruit) / $110 (PiShop), so the brain is the 2 GB ($90, same outline, fit and views unchanged); Camera Module 3 Wide $38.50, ATECC608 $4.95, 27 W supply $14.04 checked: $720 parts qty 1; price $990 → $1,090 + $4/mo, margin 23-43 % (est.); page top-three FAQs; 3 phase 3 questions; check-assembly clean. Audit S12 OK; schedule on time. Next: S14 AisleOne.

81. 2026-10-28 23:20 · S12 Kids learning: study desk

    StudyOne stage 2 refinement: buyer check (BenQ MindDuo 2 NZD 529 at PB Tech, MindDuo 2 Plus NT$7,990 at momo, Photomath Plus $9.99 a month, Ello $14.99 a month reading coach as a new competitor, Youdao Smart Lamp CNY 1,999 and its device revenue down 22.1 % in Q3 2025), price $229 → $279 + $4/mo (qty-1 margin was about 6 %); Pi Zero 2 W $19.05, Camera Module 3 Wide $38.50, VL53L1X $22.95, ATECC608 $4.95 checked: $188 parts qty 1, cost about $215 / $154 at qty 100, margin 23-45 % (est.); the two RGB reading sticks became Adafruit 2868 RGBW natural white on the same 10.2 x 51.1 x 3.2 board; page top-three FAQs; 3 phase 3 questions. Audit S11 OK; schedule on time. Next: S13 ClassOne.

82. 2026-10-28 22:58 · S11 Home: door and garage

    GarageOne stage 2 refinement: buyer check (Chamberlain B4643T camera opener about $369 at Home Depot, Secure View $509.99 at Ace, myQ storage $3.99-9.99 a month; new competitor Tailwind iQ3 $59.99-93.99 with phone-distance auto-close; NHTSA 2011 keyless ignition proposal never made final), price $449 → $499 + $2/mo (qty-1 margin was about 10 %), fitting extra in one wording; Pi Zero 2 W $19.05, Camera Module 3 Wide $38.50, piezo $1.50, ATECC608 $4.95, 35E $6.99: $356 parts qty 1; IRM-05-5 → IRM-10-5 (same 45.7 x 25.4 x 21.5 case per both Mean Well sheets); power cut about 8 h, not 3 h; check-assembly clean, 12/12 fit. Audit S10 OK; schedule on time. Next: S12 StudyOne.

83. 2026-10-28 22:19 · S10 Home: pool

    PoolOne stage 2 refinement: buyer check (CPSC June 2026: about 376 child pool and spa drownings a year, nearly 80 % under five, over 70 % at home; new competitor Poolguard GAPT gate alarm $61.99-83, MagnaLatch ALERT $149.50 at D&D's store, Coral Manta and MYLO in-pool cameras), price kept at $699 + $3/mo with "fitting extra" made consistent; Pi Zero 2 W $19.05, Camera Module 3 Wide $38.50, D36V28F5 $23.95, ATECC608 $4.95, piezo $1.50, 35E $6.99 checked: $424 parts qty 1, cost about $494 / $369 at qty 100, margin 29-47 % (est.); watching power recounted from 2.1 W to 3.1 W, so a second 35E cell went into the latch post (fitCheck passes) for about 10 h in a power cut; page top-three FAQs (drowning, why not a gate alarm, replace my gate); 3 phase 3 questions (incl. a kit for existing gates). Audit S9 OK. Schedule on time. Stage 2 ready.

84. 2026-10-28 21:55 · S9 Home: bathroom and damp

    DampOne stage 2 refinement: buyer check (Vent-Axia Svara £117.31 unchanged; new competitors Kair Smart Fan Plus £129.99 at B&Q and Vent-Axia Revive 7 with Switchee, landlord-visible fan data for Awaab's Law; private-sector Awaab's Law consultation 9 Oct-18 Dec 2026, 2027 earliest), price kept at $349 + $2/mo, BOM and buy.json prices checked ($250 qty 1: ATECC608 $4.95, piezo $1.50), margin 21-46 % (est.), supply changed from IRM-05-5 (5 W, at its limit on boost) to IRM-10-5 (10 W, same case), page top-three FAQs, power and housing size, 3 phase 3 questions. Audit S8 OK; schedule on time.

85. 2026-10-28 21:24 · S8 Home: parent's bedside

    BedsideOne stage 2 refinement: buyer check (Hero pricing page now $29.99 prepaid / $44.99 / $59.99 a month, no device fee listed; new competitor class wall fall radars, Walabot HOME about $179-249 (secondary source) and Akuvox MR01 A$731.50, bed-exit alarms $38-94), price kept at $349 + $5/mo; ATECC608 $4.95, piezo $1.50, 35E $6.99, Pololu 3415 $22.95 checked: $205 parts qty 1, cost about $260 / $187 at qty 100, margin 25-46 % (est.); page problem line and top-three FAQs; 3 phase 3 questions. Audit S7 OK; schedule on time.

86. 2026-10-28 21:05 · S7 Home: kitchen

    KitchenOne stage 2 refinement: buyer check (Faber and Elica 60 cm chimneys Rs 10,690-15,990 in 2026; new competitor class EN 50615 stove guards, Firechief FSW1 £399.98 and Innohome £554.23 plus an electrician, electric hobs only; smart hoods KOBE and GE $1,300-2,800 set the fan only), price raised from $329 to $369 + $3/mo because the qty 1 margin was about zero; ATECC608 $4.95, piezo $1.50, 35E $6.99 checked, LRS-50-5 aligned to buy.json ($17): $272 parts qty 1, cost about $332 / $227 at qty 100, margin 10-38 % (est.); backup cells now only charge below 43 °C (AO3401A gate, cells rated 0-45 °C); page top-three FAQs (gas off, cost vs stove guard, camera) and power-cut run time; 3 phase 3 questions (new hood vs retrofit strip, price with a shut-off valve, who gets the first call). Audit S6 OK. Schedule on time. Stage 2 ready.

87. 2026-10-28 20:20 · S6 School athletics

    AthleticsOne stage 2 refinement: buyer check (UCS posts from $795.50; FinishLynx Scholastic $5,699 and Bronze $9,099 on the Dec 2025 sheet, was "under $5,000"; new competitor SprintTimer phone photo-finish $4.99; NFHS officials +23 % but ageing; Ohio 2026 state tournament five divisions), price kept at $2,090 + $299/yr; Pi Zero 2 W $19.05, Camera Module 3 $29.25, 35E $6.99, ATECC608 $4.95 checked, AS3935 and PN532 aligned with buy.json: $648 parts qty 1, cost about $984 / $670 at qty 100, margin 53-68 % (est.); home post race power corrected (Walter awake for the gun mic: 2.8 W, about 15.3 h); page FAQs (no officials, cost vs phone app, sanctioned meets) and run time spec; 3 phase 3 questions (incl. whether posts stay out all season). Audit S5 OK. Schedule on time. Stage 2 ready.

88. 2026-10-28 20:05 · S5 Baseball and softball

    DiamondOne stage 2 refinement: buyer check (Hollywood Impact bases $399.95-499.95 + in-ground plate $79.95, youth umpires $15 to high school $65 a game, Richmond games with one umpire or none, July 2026; new competitor Musco Umpire Assist, one stereo camera mount, no price yet), price kept at $1,990 + $29/mo; Pi Zero 2 W $19.05, ATECC608 $4.95, AI Camera $77, 35E $6.99 checked: $1,018 parts qty 1, cost about $1,430 / $1,000 at qty 100, margin 28-50 % (est.); plate front camera power corrected (3.6 W, about 12 h); site.json prices had lost their dollar signs, fixed; page top-three FAQs and run time; 3 phase 3 questions (incl. whether sets stay out all season). Audit S4 OK. Schedule on time. Stage 2 ready.

89. 2026-10-28 19:10 · S4 Tennis and pickleball

    RacketOne stage 2 refinement: buyer check (Putterman 3 in post pair $559, SwingVision Plus/Pro/Max $95.99-299.99 a year, pickleball referees about $10-20 a match, PlayReplay four post-mounted cameras at PPA and MLP 2026 with a reported $3,500-4,500 per court a year), price kept at $1,590 + $29/mo; Pi Zero 2 W $19.05 and ATECC608 $4.95 checked, $648 parts qty 1, cost about $1,080 / $730 at qty 100, margin 32-54 % (est.); full-match power 3.1 W per post, about 13.8 h per charge (was 19 h camera time), now in the CAD power table, page, site.json and integration.json; site.json rev A hardware, architecture and GateKey use case replaced with the rev O posts; net winder missing from the CAD (costed, next CAD step); check-assembly clean. Stage 2 ready.

90. 2026-10-28 18:49 · S3 Basketball

    CourtOne stage 2 refinement: buyer check (Huupe Arena $2,995 / Portable $995 / $699 Upgrade Kit track shots but make no calls; Noah $4,800 + $100/mo for two hoops; varsity referee fees about $92-131 a game), BOM prices checked today (Pi Zero 2 W $19.05, piezo $1.50, ATECC608 $4.95): $3,328 qty 1, cost about $3,830 / $2,860 at qty 100 (est.); price raised $4,690 → $4,990 (margin 23-43 %, est.), still below about $6,350 for plain boards and a shot clock set; page: shot clock reset wording now matches rev O, power 9.5 W, corner cameras and secure element in the sensors, rev A alerts (RimPod, cornhole) replaced, top-three FAQs; check-assembly clean. Stage 2 ready.

91. 2026-10-28 18:45 · S2 Football

    FootballOne stage 2 refinement: buyer check (plain goal pairs £1,615-2,775, youth referee fees £10-45 capped by county FAs, Veo from $1,299, SmartGoals training goals, no grassroots calling goal found), price kept at $4,690; BOM prices checked today (Pi Zero 2 W $19.05, piezo $1.50, ATECC608 $4.95, e-paper SKU 12955 $9.99): $2,770 qty 1, cost about $3,100 / $2,270 at qty 100 (est.), margin 34-52 % (est.); page power corrected to 15.3 h per post, specs list calls alone vs assists, top FAQs; new finding: youth sizes (21 x 7, 16 x 7 ft) to phase 3; check-assembly clean. Stage 2 ready.

92. 2026-10-28 18:10 · S1 Cricket

    CricketOne stage 2 refinement: buyer check (leagues invoice clubs for panel umpires; Wicktron calls front-foot no-balls only; Zing lights only), price kept at $1,690; bom.csv now lists every rev O part (4 XIAO ESP32-S3 Sense, 24 NeoPixel Sticks, 12 piezos, 12 D2F-01, 2 ATECC608) with prices checked today: $976 qty 1, about $720 qty 100 (est.), margin 37-52 % (est.); buy.json band quantities and the e-paper SKU 12955 fixed; side-stump run time corrected to about 19 h of play; page summary, specs and top-five FAQs; check-assembly clean. Stage 2 ready.

93. 2026-10-28 15:25 · Backlog 2 (drawn idea plans)

    Wrote idea.json for DampHome (an Indian home kitchen: hub beside the RO purifier, flow sensor on the tap line, leak foot in the sink cabinet, wall tag on the damp wall) and DripCrown (a six-bed ward bay at night: a weighing crown on two IV stands, a drip camera on one line, the ward view at the nurses station), so their concept images are drawn plans instead of the fallback card. 36 catalogue products still use the fallback card.

94. 2026-10-28 14:55 · Backlog 2 (drawn idea plans)

    Wrote idea.json for ColdChain Guard AI (a reefer box truck backed onto a receiver's dock: logger on the front wall, product probe in its simulant bottle, air probe at the return air, door tag on the rear door) and CycleTap (an apartment laundry room: gateway by the outlet, pucks on two washers and a dryer, free-now QR tag on the door), so their concept images are drawn plans instead of the fallback card. 38 catalogue products still use the fallback card.

95. 2026-10-28 14:25 · Backlog 2 (drawn idea plans)

    Wrote idea.json for AlarmEcho (a two-bedroom rental unit: a disc beside each of the three alarms and the Echo Link hub in a living-room outlet) and BagPost (a care-home corridor at night: stands in two rooms and a RailHook on a bed rail), so their concept images are drawn plans instead of the fallback card. 40 catalogue products still use the fallback card.

96. 2026-10-28 14:20 · Backlog 2 (drawn idea plans)

    All flagships have a first design and no [UNVERIFIED] sizes; wrote idea.json for AirfieldWatch AI (airport plan: pond, both runway ends, creek fence, ditch, review queue) and AirLeak AI (compressor room: wall box, two clamps, header sensor, probe), so their concept images are drawn plans instead of the fallback card. Also fixed the floating compute backplate in the AirfieldWatch assembly (bought box back wall shown on the mast). 42 catalogue products still use the fallback card.

97. 2026-10-28 13:45 · F18 Public sector, roads and airfields

    Built RoadOne (roadone): the signalised crossing's own near-side signal head (356 x 254 x 1,067 mm, bottom 2,650 mm above the footway, on its pole at the stop line) with every part behind its own LED modules: Pi 5 brain, approach camera in a floor window 25 degrees down the approach (first placed between the green and amber sections; the green visor hid its near view), crossing camera in the road-side wall, LD2450 kerb radar, lamp clamps, controller contacts, call lights, tone, ATECC608, D2F-01, LTE-M and backup; 8 calls (6 alone); 21/21 fit items from datasheets, 15 views checked, no collisions, no floating parts

    Folded roadsense-ai. Every row in the plan now has a first design; next runs work the backlog (CAD review, drawn idea plans).

98. 2026-10-28 13:05 · F17 Waste and recycling

    Built BayOne (bayone): the tipping bay's own traffic light (300 x 260 x 860 mm, three 200 mm aspects, on its base plate on the push wall top at 3,212-4,072 mm) as the one product; bay-side wall: Camera Module 3 Wide bay camera 22 degrees down, MLX90640 heat camera 12 degrees down, LD2450 radar 40 degrees down; behind the door: approach camera and call lights behind the aspects; floor: Pi 5 brain, Walter LTE-M, ATECC608, tone and horn; back: IRM-20-5, PowerBoost + LiPo, G5V-1 supervisory relay; D2F-01 pressed by the door; CAD asserts the size, nothing over the bay mouth and the visor reach, inView checks 15 views with a new push-wall line-of-sight test; equipment.json with zone, 8 calls, 4 impact cases and every security key; bay plan, 6 integration parts, 5 screens; $3,990 + $49/mo vs about $6,630 + $158/mo

    Folded recycleyard, emberwatch-ai. CAD review clean, all sizes from datasheets. Next: F18 Public sector, roads and airfields.

99. 2026-10-28 12:30 · F16 Construction and mining sites

    Built SiteOne (siteone): the site's own full-height entry turnstile (1,400 x 1,400 x 2,403 mm, 787 mm passage, the same baseplate, rotor, lock and controller) as the one product; canopy: Pi 5 brain, entry camera 35 degrees down (hard hat, hi-vis, boots), site camera 20 degrees down (smoke, muster point), LD2450 radar down into the rotor, SPS30, SHT45, PT1000 sun load, MAX9814, call lights, voice prompt, ATECC608, D2F-01 pressed by the lid, Walter LTE-M, IRM-20-5, LiPo; reader head: PN532 tap reader, e-paper heat flag, PPE light; CAD asserts the size, headroom, protrusion and passage, inView checks 12 views; equipment.json with zone, 8 calls, 4 impact cases and every security key; gate plan, 6 integration parts, 5 screens; $8,990 + $99/mo vs about $9,098 + $198/mo

    Folded ppegate-ai, heatguard-ai. CAD review clean, all sizes from datasheets. Next: F17 Waste and recycling.

100. 2026-10-28 11:50 · F15 Water, energy and utilities

     Built PumpOne (pumpone): the lift station's own duplex pump control panel (508 x 610 x 203 mm NEMA 4X stainless, door and back plate) with every part inside: brain, level and pump current inputs, two pump clamps, LTE-M, backup cell, secure element and tamper switch on the back plate; station camera and radar behind side windows; heat camera, e-paper, call light and tone in the door; 16/16 fit items from datasheets, 13 views, no collisions, no floating parts; equipment.json (8 rules, 8 calls, 4 impact cases, security), idea.json, integration 6 parts 5 screens; $2,690 + $45/mo vs about $3,911 + $87/mo

     Folded pumpstation, liftstation-ai, gaugeread-ai, panelheat-ai. Next: F16 Construction and mining sites.

101. 2026-10-28 11:27 · F14 Farms: crops and livestock

     Built FarmOne (farmone): the 600 L fence-line stock trough, with every part in its own valve cover (cameras and heat cameras for both paddocks, meter, float sensor, probe, air sensor, mic, LTE-M, solar, call lights, secure element, tamper switch); 9 calls (7 alone), 23/23 fit items from datasheets, 21 views checked, no collisions, no floating parts; $1,290 + $9/mo vs about $1,524 + $135/yr

     Folded troughtally, calvingpen, calveclock. Next: F15 Water, energy and utilities.

102. 2026-10-28 11:40 · F12 Care homes and hospitals

     Built CareOne (careone): the bedside locker made smart, every part in its sealed 70 mm top (XIAO ESP32-S3, LD2450 radar turned 40 degrees to the bed and 25 down, 5 kg jug and 1 kg hand-rub plates, 16 mm call key + MAX9814 voice note, call light, tone, ATECC608, D2F-01, backup); no camera; 8 calls (5 alone, 3 assist); 16/16 parts from datasheets, 7 radar views, no collisions, no floating parts; $590 + $15/mo per bed vs $897 + $27/mo

     Folded bedside, carewatch-ai, turncup, callnote.

103. 2026-10-28 11:05 · F11 Food, cold chain and pharma

     Built ChillOne (chillone): the walk-in's own door frame (1,064 x 2,030 x 101.6 mm, 34 x 76 in opening) with every part inside the header and jambs; 17/17 fit items from datasheets, 11 camera and radar views, no collisions, no floating parts; equipment.json (10 rules, 8 calls, 3 impact cases, security), idea.json, integration 6 parts 5 screens; $690 + $29/mo vs about $797 + $90/mo

     Folded coldroom. Next: F12 Care homes and hospitals.

104. 2026-10-28 10:25 · F10 Manufacturing plants

     Built CellOne (cellone): the machine's own Ø70 x 446 mm signal tower as the one product (80 mm base, 66 mm sensor tier, blue, green, yellow and red 66 mm lamp tiers, 36 mm cap, the WERMA KombiSIGN 71 size, same bracket and 8-pin 24 V lead); base: Pi Zero 2 W brain standing upright, ADS1115 on the controller's four tier lines, LSM6DSOX cycle sensor, ATECC608, D2F-01 pressed by the sensor tier, D36V28F5; sensor tier: Camera Module 3 Wide 45 degrees down, MLX90640 15 degrees down along the barrel, MAX9814; NeoPixel Stick 8 per lamp tier, piezo in the cap; CAD asserts Ø70, 446 mm and red on top, inView checks 10 views; equipment.json with zone, 7 calls, 4 impact cases and every security key; cell plan, 6 integration parts, 5 screens; $790 + $29/mo vs about $1,703 + $47/mo

     Folded machinestate-ai. CAD review clean, all sizes from datasheets. Next: F11 Food, cold chain and pharma.

105. 2026-10-28 09:45 · F9 Logistics and warehouses

     Built DockOne (dockone): the dock door's own pair of track guards (273 x 203 x 1,232 mm, 4.8 mm steel, same anchors, the door track in the same notch) as the one product; sealed top bays hold the brain, two door cameras in stereo across the doorway, a dock camera, a radar, air and strike sensors, call lights, tones, secure elements and tamper switches; CAD asserts the 2,743 mm opening, height and notch, inView checks 16 views; equipment.json with zone, 9 calls, 4 impact cases and every security key; dock plan, 6 integration parts, 5 screens; $1,290 + $49/door/mo vs about $2,374 + $278

     Folded loadingdock-ai, doorpost. CAD review clean, all sizes from datasheets. Next: F10 Manufacturing plants.

106. 2026-10-28 09:05 · F8 Facilities and hospitality

     Built RoomOne (roomone): the guest-room thermostat itself, rev S and rev O from the start; breathing radar, SEN66 air module, e-paper, keys, call light, tone, secure element, tamper switch, HVAC relays and 24 VAC power inside the plain 140 x 110 x 32 outline; 8 calls (6 alone, smoking fee and welfare checks assist); 15/15 parts from datasheets, 7 radar views, no collisions, no floating parts; 5 integration parts, 5 app screens with night mode; $349 + $8/room/month vs about $658 + $13

     Folded guestroom, vapesense-ai. Next: F9 Logistics and warehouses.

107. 2026-10-28 08:30 · F7 Retail and grocery stores

     Built AisleOne (aisleone): the aisle's own gondola pair as the smart equipment, cameras, call lights, tone, secure element and tamper switch inside both top canopies; 22/22 parts fit from datasheets, 28 camera views, ADA width and protrusion checks, no collisions or floating parts; 5 integration parts, 5 app screens with No staff in the aisle mode; $690 + $15/mo per aisle vs about $938 + $91/mo

     Folded storeeye, shelfscan-ai, spillguard-ai.

108. 2026-10-28 07:41 · K2 Kids learning: the classroom

     Built ClassOne (classone): the 2,400 x 1,200 front whiteboard as the one product; header rail 70 x 80: Raspberry Pi 5 brain, 2 Camera Module 3 Wide room cameras at the centre turned 30 degrees out and 20 down (heads only), SCD41 CO2 over vents, SPH0645 sound-level mic, 12 NeoPixel Sticks as the time bar behind an opal cover, MAX98357A + speaker, ATECC608, D2F-01 pressed by the cover; pen tray 70 x 45: PN532 tap-in reader under a lid; CAD asserts the outline and 70 mm from the wall (ADA 100 limit), inView checks 12 views; equipment.json with zone, 7 calls, 3 impact cases and every security key; classroom plan, 6 integration parts, 5 screens; $990 + $4/mo vs about $1,094 + $5/mo

     No product folded (iemi-gecko stays listed). CAD review clean, all sizes from datasheets.

109. 2026-10-28 07:25 · K1 Kids learning: the study desk

     Built StudyOne (studyone): the child's bar desk lamp as the one product (200 x 160 base with 1.2 kg ballast, 30 mm column, 60 x 400 head arm at 450-480 mm); head arm: Pi Zero 2 W brain, Camera Module 3 Wide straight down at the page, SPH0645 mic, VL53L1X tilted 45 degrees at the child, VEML7700, 3 NeoPixel Sticks, ATECC608, D2F-01 pressed by the cover; base: MAX98357A + 28 mm speaker; CAD asserts the outline, nothing below the head and no tip on a 10 degree tilt, inView checks 10 views; equipment.json with zone, 8 calls, 3 impact cases and every security key; desk plan, 6 integration parts, 5 screens; $229 + $4/mo vs about $344 + $22/mo

     No product folded (iemi-gecko stays listed). CAD review clean, all sizes from datasheets.

110. 2026-10-28 06:48 · H5 Home: the front door and garage

     Built GarageOne (garageone): the garage door opener head as the one product (300 x 460 x 180, underside 2,220 mm over a 2,134 mm opening, on its own rail and straps, eyes at 150 mm kept); door-end bay: Pi Zero 2 W brain, Camera Module 3 Wide tilted 25 degrees down, piezo, ATECC608, 2 NeoPixel Sticks, D2F-01 pressed by the cover; back-end bay: MLX90640 tilted 40 degrees at the water heater, MAX9814, SHT45, IRM-05-5, 18650; CAD asserts the head size, door travel clearance and eye height, inView checks 12 views; equipment.json with zone, 8 calls, 4 impact cases and every security key; garage plan, 6 integration parts, 5 screens; $449 + $2/mo vs about $528

     Folded garagehome, garageeye-ai. CAD review clean, all sizes from datasheets.

111. 2026-10-28 06:40 · H4 Home: the pool

     Built PoolOne (poolone): the pool gate itself is the product; latch post holds the brain, pool camera, radar, Hall sensor, call lights, tone, secure element and tamper switch, hinge post the outside camera and power; 3 parts, 16/16 fit items from datasheets, 15 sensor views, no collisions, no floating parts; 6 integration parts, 5 app screens; $699 + $3/mo vs about $1,039

     Folded poolguard-ai, pooldeck.

112. 2026-10-28 05:55 · H3 Home: bathroom, laundry and damp

     Built DampOne (dampone): the 100 mm ceiling extractor fan as the one product (200 x 200 x 30 grille, 170 x 170 x 110 housing, Ø100 spigot); grille rim around a walled 110 x 110 air path: XIAO ESP32-S3 brain, MLX90640 wall heat camera behind a side window tilted 30 degrees at the outside wall, MLX90640 floor heat camera, SHT45 behind PTFE vents, LD2410C, MAX9814, 2 NeoPixel Sticks + piezo, ATECC608, D2F-01 pressed by the ceiling; Mean Well IRM-05-5 in the housing so only 5 V is below the ceiling; CAD asserts the rim parts out of the air path, nothing below the grille and mains above the ceiling, inView checks 15 views; equipment.json with zone, 7 calls, 4 impact cases and every security key; bathroom plan, 7 integration parts, 5 screens; $349 + $2/mo vs about $443

     Folded mouldguard. CAD review clean, all sizes from datasheets.

113. 2026-10-28 05:15 · H2 Home: an older parent's bedside

     Built BedsideOne (bedsideone): the bedside table and lamp as the one product (450 x 400, 600 mm top, open shelves, lamp head to 1,000 mm); lamp head: XIAO ESP32-S3 Sense brain (camera 10 degrees out over the pill box, PDM mic), 2 NeoPixel Sticks, piezo, ATECC608, D2F-01 pressed by the lamp cover; top box: 2 Adafruit 4540 load cells + 2 NAU7802 under the pill-box and when-required plates, LD2450 turned 15 degrees to the bed, VL53L1X at the walker spot, PowerBoost + 2 x 18650; floor glow in the kick board; CAD asserts the footprint, floor contact and flush plates, inView checks 11 views; equipment.json with zone, 9 calls, 3 impact cases and every security key; bedroom plan, 6 integration parts, 5 screens; $349 + $5/mo vs $447 + $14/mo plus a table and lamp

     Folded parenthome, pillperch. CAD review clean, all sizes from datasheets.

114. 2026-10-28 05:01 · H1 Home: the kitchen

     Built KitchenOne (kitchenone): the 60 cm chimney hood as the one product, everything in a sealed front strip (XIAO ESP32-S3 brain, MLX90640 110 degree heat camera through a Ø14 window, LD2410C behind a 1.5 mm fascia skin, MAX9814 mic, 2 NeoPixel Sticks + piezo, ATECC608, D2F-01 pressed by the fascia, LRS-50-5, PowerBoost + 2 x 18650); CAD asserts 700 mm over the gas hob and nothing below the underside, inView checks 13 views; equipment.json with zone, 7 calls, 4 impact cases and every security key; kitchen plan, 6 integration parts, 5 screens; $329 + $3/mo vs about $342 + $6/mo

     Folded kitchenguard, stovesentry-ai, cookercount. CAD review clean, all sizes from datasheets.

115. 2026-10-28 04:40 · O6 Athletics

     AthleticsOne rev O: the finish posts are their own finish judge. Re-checked the item: the posts stay (a smart block set would make one call and cannot see the finish). equipment.json gets zone (nothing added), 6 calls (5 alone in school and club play: places and dead heats, times, finishers, the bell lap, the lightning start hold; lanes at the line assist; unsure below a stated confidence), 4 impact cases (athlete into the post, topple onto the track, spike or baton on a window, weather) and every security key; CAD adds an Adafruit 160 piezo, an ATECC608 and an Omron D2F-01 tamper switch in each post and a MAX9814 gun mic in the home post (times with no starter unit), mounts each finish camera 10 degrees down, and inView() checks 26 views; new app screens (No official, Finish calls); $2,090 vs $3,592 + about $795; folds kept. CAD review clean, all sizes from datasheets. O1–O6 done: REQUIRE_OFFICIAL set to true. Also repaired the O5 log line.

116. 2026-10-28 03:57 · O5 Baseball and softball

     DiamondOne rev O: re-checked the equipment; the L-screen cannot officiate, so the product is now a smart home plate and three smart bases at OBR 2.02 and 2.03 size (CAD asserts 17 in front, 8.5 in sides, 12.02 in back edges, top flush and nothing above ground, bases 15 in square and 100 mm thick); plate: Pi Zero 2 W brain, AI camera under a flush Ø34 window looking up the zone, XIAO ESP32-S3 Sense front camera and OPS243-A radar facing the pitcher, NeoPixel light, piezo, ATECC608, D2F-01, 4 x 18650; each base: XIAO ESP32-S3 Sense camera and mic, 4 Adafruit 1075 force sensors, light, piezo, ATECC608, D2F-01, 2 x 18650; equipment.json zone/decisions (6 calls, 5 alone)/impact/security, new diamond plan, 6 integration parts, 5 screens; inView checks 11 call targets; $1,990 a set + $29/month vs $4,178 + $138/month; folds kept. CAD review clean, all sizes from datasheets.

117. 2026-10-28 03:15 · O4 Tennis and pickleball

     RacketOne rev O: the net posts are their own line and net judge. equipment.json gets zone (nothing added), 8 calls (7 alone in club play, far baseline and foot faults assist; unsure below a stated confidence), 4 impact cases (90 J serve on the windows, player into the post, cord tension, weather) and every security key; CAD adds a XIAO ESP32-S3 Sense kitchen camera behind a flush Ø12 window low on each post, a NeoPixel Stick 8 call light, a piezo tone, an Omron D2F-01 tamper switch and an ATECC608 in each post, turns the AI camera 30 degrees in, and inView() checks 14 call targets; new app screens (No official, Line calls, Kitchen and net calls); $1,590 vs $2,374 + $44/month; folds kept. CAD review clean, all sizes from datasheets.

118. 2026-10-28 02:45 · O3 Basketball

     CourtOne rev O: the backboard is its own official. equipment.json gets zone (nothing added), 8 calls (7 alone in club play, fouls near the basket assist; unsure below a stated confidence), 4 impact cases (EN 1270 ring load, 45 J ball on the windows and clock, player into the padded edge) and every security key; CAD adds a XIAO ESP32-S3 Sense corner camera behind a flush Ø12 window in each side rail (55 degrees out, 30 down), tilts the AI camera 48 degrees, and adds a piezo tone, an Omron D2F-01 tamper switch and an ATECC608 in each top rail; inView() checks 10 call targets; new app screens (No official, Basket calls, Shot clock, Goaltending); $4,690 vs about $6,350; folds kept. CAD review clean, all sizes from datasheets.

119. 2026-10-28 02:22 · O2 Football

     FootballOne rev O: the goal is its own referee. equipment.json gets zone (nothing added; corner flags stay plain), 8 calls (6 alone in club play, encroachment and keeper handling assist; unsure below a stated confidence), 4 impact cases (200 J shot, EN 748 crossbar 1,800 N and tip-over 1,100 N, player collision) and every security key; CAD adds a XIAO ESP32-S3 Sense goal-line camera behind a flush Ø12 window in each post's inside face, 4 NeoPixel Sticks behind flush clear strips + piezo horn + Omron D2F-01 tamper switch in all 4 posts, an ATECC608 in each left post, and inView() checks 10 call targets; new app screens (No official, Goal calls, Restarts and penalties); $4,690 vs about $7,214; folds kept. CAD review clean, all sizes from datasheets.

120. 2026-10-28 01:55 · O1 Cricket

     CricketOne rev O: the wicket is its own umpire. equipment.json gets zone (nothing added), 9 calls (8 alone in club play, LBW assists; unsure below a stated confidence), 4 impact cases (150 km/h ball 136 J, stump flying, bails free with no magnetic hold, bat strike) and every security key; CAD adds a XIAO ESP32-S3 Sense stereo camera in each off and leg stump, 2 NeoPixel Sticks + piezo + Omron D2F-01 tamper switch in all 12 stumps, an ATECC608 in each middle stump, and inView() checks that each camera sees its calls; new app screens (No official, Run-out call, No-ball call, LBW evidence, Over-rate); $1,690 vs $4,154; folds kept. CAD review clean, all sizes from datasheets.

121. 2026-10-28 01:00 · R6 School athletics

     AthleticsOne rev S: the smart finish posts. 2 posts at the 80 x 20 mm winning-post profile, 1,400 mm on 400 x 250 x 40 bases, 1 m outside lane 1 and lane 8 on the finish line's extension (CAD asserts the profile above the base and the height), Camera Module 3 (720p100) + Pi Zero 2 W + NeoPixel Stick 8 in each post, Walter brain + AS3935 + SHT45 + MAX31865/PT1000 on the black top + PN532 in the home post, 4 x 18650 + PowerBoost in each base, starter unit (XIAO ESP32-S3, MAX9814, BPW34) docked on the home base as the one detachable part; station, digits, MeetBox case, TapGrips, poles, pucks, tags and rain gauge removed; equipment.json, new finish-line plan, 7 integration parts, 5 screens; $1,990 a pair vs $3,592 + about $795 plain posts; folds kept. CAD review clean, all sizes from datasheets. R1–R6 done: REQUIRE_EQUIPMENT set to true. Also fixed the DiamondOne log line above.

122. 2026-10-28 00:25 · R5 Baseball and softball

     DiamondOne rev S: the smart L-screen. 7 x 7 ft frame with the 3 x 3 ft cut-out (CAD asserts 2,134 x 2,134, 914 x 914 clear, nothing in front of the hit face above the foot plates), welded 6061 tube 101.6 x 50.8 main post and 50.8 x 50.8 top bar; Pi 5 + OPS243-A + LD2450 in the main post, 2 x Arducam B0385 (500 mm stereo baseline, also the replay) + NeoPixel Stick 8 lane lamp in the top bar; lane node, back-net radar and ReplayCam, kiosk, guard pod, reed clips, listed plug and DialDrive removed; no detachable part; equipment.json, new lane plan, 5 integration parts, 5 screens; $2,490 + $69/month vs $4,178 + $138/month plus a screen; folds kept. CAD review clean, all sizes from datasheets.

123. 2026-10-27 23:50 · R4 Tennis and pickleball

     RacketOne rev S: the smart net posts. 2 posts at Ø76.2 x 1,090 mm (CAD asserts ≤ 76.2 USA Pickleball, ≤ 150 ITF, top 20 mm above the 1,070 mm cord, centres 0.914 m outside), AI camera + Pi Zero 2 W + LD2450 + XIAO nRF52840 Sense + 2.13 in e-paper + 2 keys + 4 x 18650 + PowerBoost in each post, Walter in the home post; score head, solar cap, base unit, RallyPads, LineCams, GateKey and LightLink removed; no detachable part; equipment.json, new court plan, 7 integration parts, 4 screens; $1,490 per court vs $2,374 + $44/month; folds kept. CAD review clean.

124. 2026-10-27 23:35 · R3 Basketball

     CourtOne rev S: the smart backboard. 2 ends at FIBA 2026 size (CAD asserts board 1,829 x 1,067, ring Ø457 Ø18 rod, top 3,050, 151 out, no ring-glass contact, clock 320 above / 100 behind, strips 99 %), AI camera + Pi Zero 2 W + Matrix Bonnet in each top rail, NeoPixel red/yellow edge strips, XIAO nRF52840 Sense in a printed ring box, 2 x Adafruit 2277 panels + Mean Well LRS-50-5 in each clock box; ceiling head, far-half head, CourtKey, RimPods, wall faces and ShotKey removed; no detachable part; equipment.json, new court plan, 6 integration parts, 5 screens; $4,490 pair vs about $6,350 plain (est.); folds kept. CAD review clean.

125. 2026-10-27 23:05 · R2 Football

     FootballOne rev S: the smart goal. 2 goals at Law 1 size (CAD asserts 7,320 inside, 2,440 under the crossbar, 120 x 100 posts and crossbar), AI camera + Pi Zero 2 W at each post top, 4 x 18650 + PowerBoost in each post foot, XIAO nRF52840 Sense + cell in each crossbar, referee wrist unit (CricketOne band) docked in a flush tray in the home left post; hub, AP stand, saddles, HeadCount pod, both boards, flags and RefCam removed; keeper count now automatic; equipment.json, new pins, 6 integration parts, 5 screens; $4,490 pair vs about $7,214; folds kept. CAD review clean.

126. 2026-10-27 22:35 · R1 Cricket

     CricketOne rev S: the smart wicket. 6 stumps + 6 spikes at Law 8 size (CAD asserts Ø38.0, 711.2 top, 228.6 wide, bails +7.5), AI camera + Pi Zero 2 W + PowerBoost + 2 x 18650 in each middle stump, XIAO nRF52840 Sense + 18650 in each side stump, DRV5032 under 8 grooves, band reused as the one detachable part; mast, clock, ReviewPad, CloseEye, BallCheck, DialDrive removed; equipment.json, new idea plan, 6 integration parts, 5 screens; $1,490 vs $4,154; folds kept. CAD review clean.

127. 2026-10-27 22:20 · F6 School athletics and swimming

     AthleticsOne CAD review third pass: 10 of 15 bought-part sizes from maker/seller pages; brain board is 110 x 32 x 20 (moved, standoffs 100 x 24); cover tag holder corrected from Keystone 2482 (4 x AAA) to 2468 (2 x AAA, 53 x 24.6 x 13), lid posts out 1 mm; wet-spot pocket 59.5 x 32.9 for the Keystone 2462; 143/143 fit items pass, no collisions, no floating parts. Still [~]: 5 sizes.

128. 2026-10-27 21:55 · F6 School athletics and swimming

     AthleticsOne rev B: GroundsKit GroundCheck joins as a module (4 wet-spot pucks, 2 cover tags, rain gauge on post B) heard by a RAK3272S LoRa receiver inside the station (25.4 x 32.3 from the RAK datasheet); new morning ground page (4/4 checks); 17 reused prints, no collisions, no floating parts; striper decided: stays SmartLine (different buyer, every field, own RTK base); $2,739 + $299/yr vs $3,592 + $299/yr

     Folded groundskit. [~] left: 15 bought-part sizes.

129. 2026-10-27 21:25 · F6 School athletics and swimming

     Built AthleticsOne (athleticsone): FieldSide station as the one brain + SchoolMeet MeetBox, 8 TapGrips and 2 SplitPole finish poles as modules; new storm hold (no start during the lightning clock, 4/4 checks); 13 reused printed parts assembled at a school track by the practice-field fence; 7 integration parts, 5 app screens, $2,490 + $299/yr vs $3,243 + $299/yr

     Folded schoolmeet, splitblock, splitpole, fieldside, fieldflag. [~] left: GroundCheck module + fold groundskit, 13 sizes. F5 DiamondOne closed for desk work (8 caliper items wait for parts).

130. 2026-10-27 20:50 · F5 Baseball and softball

     DiamondOne CAD review fifth pass: strobe pair named as 2 x Adafruit 3111 22 AWG (1.5 OD, fits the 3.6 grommet), both USB-C leads as Adafruit 4474 (Ø4.5, fits the pod and kiosk lead spaces), OPS243 product brief OPS-PB-004-F confirms 75 x 90 x 12 and the USB socket edge; no collisions, no floating parts. Still [~]: 8 board details that makers do not publish; next step is calipers on delivered parts.

131. 2026-10-27 20:20 · F5 Baseball and softball

     DiamondOne CAD review fourth pass: Waveshare ESP32-S3-Touch-LCD-7 is 13.5 mm deep (5.00 glass + 8.50 board, Waveshare drawing), still seats in the 30 mm kiosk; Pi 5 port stack (about 13.4 mm) and underside (1.5 mm) measured on the 1:1 Raspberry Pi drawing; USB-C plug keep-out 12.4 wide from the Type-C spec; no collisions, no floating parts. Still [~]: 9 small board details (listed in DESIGN.md).

132. 2026-10-27 20:05 · F5 Baseball and softball

     DiamondOne CAD review third pass: reed clip rev B (pocket 29.6 x 15.8 x 9.4) for the real Adafruit 375 gate contact, PAUSE key cap from the Adafruit 1441 page, OPS243-A and HLK-LD2450 outlines confirmed on maker pages; BOM names both parts; no collisions, no floating parts. Still [~]: 13 sizes left (listed in DESIGN.md).

133. 2026-10-27 19:35 · F5 Baseball and softball

     DiamondOne CAD review second pass: 16 inherited bought-part sizes replaced with datasheet values (SFH 4715AS LEDs, B0385 holder/lens/header, Pololu D36V28F6 and D24V5F5, Adafruit 1441 PAUSE key); the D36V28F6 is 20.3 x 17.8 x 8.8, so it is turned across the DialDrive tub; no collisions, no floating parts. Still [~]: 20 sizes left (listed in DESIGN.md).

134. 2026-10-27 18:55 · F5 Baseball and softball

     Built DiamondOne (diamondone) from CageLane + new ReplayCam module (same cam-pod prints, Arducam B0385 colour USB camera on the back-net upright) + live-pitcher pitch count on the existing radar: 17 reused printed parts + 2 ReplayCam prints, assembled lane scene, 6 integration parts, 5 app screens, $2,990 + $89/mo vs $4,178 + $138/mo

     Folded cagelane, cagetrax, coachvision-ai (its football SidelineReplay kit is a gap for FootballOne). [~] left: datasheet sizes for 25 inherited [UNVERIFIED] items.

135. 2026-10-27 18:25 · F4 Tennis and pickleball

     Finished RacketOne: CAD review second pass, all 26 reused [UNVERIFIED] bought-part sizes replaced with datasheet values (sources in fitCheck); GS camera mount ring is Ø36, so cam-pod and baseclip-box got a Ø37.2 bore; solar-cap rim +0.6 mm for the Voltaic P103C (210 x 113 x 5); exact parts named (Adafruit 3423 keys, Pololu U3V70F5, Adafruit 2011/1441/1192/3968, Keystone 2462); BOM ~$845 full court; no collisions, no floating parts. Marked [x].

136. 2026-10-27 17:55 · F4 Tennis and pickleball

     Built RacketOne (racketone) from CourtPost + ScorePost RallyPads and BaseClips (LineCam) + new GateKey module (QuietLight case reused on the gate post, BLE phone pass, relay on the existing 12 V lock): 19 reused printed parts, assembled court scene with net post, side fence and gate, 7 integration parts, 5 app screens, $1,749 + $29/mo full court vs $2,374 + $44/mo plus a gate system

     Folded courtpost, scorepost, picklequiet. smartline-ai not folded (F6). [~] left: datasheet sizes for reused [UNVERIFIED] items. Fixed courtone page self-fold.

137. 2026-10-27 17:12 · F3 Basketball and indoor courts

     Built CourtOne (courtone) from CourtCam + SmartHoop's ShotClock as a module and a far-half head module: 10 reused printed parts, assembled gym scene with shot clock on the end wall and ShotKey on the scorer's table, 7 integration parts, 5 app screens, $2,970 + $69/mo full court vs $9,174 + $257/mo

     Folded courtcam, courtcall-ai, smarthoop, cornholecam. coachvision-ai not folded (cages, F5).

138. 2026-10-27 16:52 · F2 Football

     Built FootballOne (footballone) from GoalSide + OffsideFlag referee kit (flag grips, wrist unit with KeeperCount, RefCam): 24 reused printed parts, assembled pitch scene with referee, 5 app screens, $3,990 + $150/mo vs $4,814 + $150/mo

     Folded goalside, smartgoal, offsideflag. coachvision-ai not folded (cage replay, F5); UmpCam waits for F5.

139. 2026-10-27 16:40 · F1 Cricket

     Built CricketOne (cricketone) from MatchKit + CreaseCam ReviewPad, CloseEye pods and BallCheck dock: 16 reused printed parts, assembled boundary scene, 5 app screens, $2,999 vs $4,154

     Folded creasecam, smartstump-pro, lengthlock, pitchtally, matchkit.

140. 2026-11-01 13:40 · Phase 2 R33 CricketOne

     Calls survive a stump-service restart: the outbox kept on the card (fsync before send, answers appended, head line with start id and last q), sim/outbox-r33.mjs 25/25; a club day with stump restarts and power cuts lost 2 of 624 calls the old way, 0 now on both boards; R34 added; no part change.

141. 2026-11-01 14:21 · Phase 2 R34 CricketOne

     Bails read at start: stump.py reads each bail and the wicket state on the card when it starts; a bail still off 200 ms later is an unsure call with a clip; sim/bails-r34.mjs 29/29; missed wickets on a hard simulated day 2.0 → 0.2 (Pi) and 0.14 (Radxa), no false calls; R35 added; no part change.

142. 2026-11-01 16:30 · Phase 2 R38 CricketOne

     One pair check on day one: pair_test.py --watch passes or fails the R36 hold and R37 takeover from a laptop on the pair Wi-Fi; sim/pair-r38.mjs 35/35; takeover mark corrected to 16 s; R39 (stump clocks with no internet) added; no part change. Folds none.

143. 2026-11-01 16:55 · Phase 2 R39 CricketOne

     Stump clocks with no internet: chrony on both stumps with one keeper (call_link.py class Clock), marked unsynced lines and signed clock steps; sim/clock-r39.mjs 29/29: pair times agree within about 1 ms (were 28-57 min apart); R40 (clock floor at boot) added; no part change. Folds none.

144. 2026-11-01 17:09 · Phase 2 R40 CricketOne

     Clock floor at boot: clock_floor.py and cricketone-clockfloor.service set the clock no earlier than the last alive mark or signed call, signed as a clock-floor note; sim/clock-r40.mjs 26/26: signed times never run backwards after a double power cut (were 98.7-99.8 %), within 62 s of the real time after a short cut; R41 (phone time for the pair) added; no part change. Folds none.

145. 2026-11-01 17:55 · Phase 2 R41 CricketOne

     Phone time for the pair: POST /time, only the keeper steps forward between balls, the follower resyncs; sim/clock-r41.mjs 24/24; long switch-off 1,199 s → 0.1 s behind (median); rows R42-R46 added for the rev S to-dos.

146. 2026-11-01 18:20 · Phase 2 R42 CricketOne

     Camera view at the rev S height: sim/view-r42.mjs 18/20; both views clear the head window; z 662 sees the pitch about 7 % finer, popping crease in view, finder reach 28.6 m; the Ø18 disc stood 0.77 mm proud on a 0.2 mm ledge, so a boss, a 1.6 mm ledge and a flush Ø16 disc (head CAD and FIRST-STUMP.md section 7). Folds none.

147. 2026-11-01 18:52 · Phase 2 R43 CricketOne

     Spring button sizing: sim/button-r43.mjs 12/12; 1/4 in × 3/16 in snap button on a 0.6 mm seat pad (tip 0.29 mm inside); hole moved from 5.55 to 9.55 mm below the tube top in an 18 mm spigot (hold 383 → 492 N est.), heat bar 3 mm lower; 5 mm hex key opens the head; FIRST-STUMP.md section 7 updated (tube hole Ø6.5 at 9.5 mm). Folds none.

148. 2026-11-01 19:55 · Phase 2 R45 CricketOne

     Maker files for rev S: build/cad-s/maker-files-s.mjs 30/30; head key notch fixed (non-manifold edge); tube.step and tube.dxf (stock 590.55 mm tube, one Ø6.5 hole), head, head-side, base and bail STEP, clear-parts.dxf; finding: the vinyl wrap would cover the button hole, so a Ø8 cut-out (FIRST-STUMP.md section 7); no part change. Folds none.

149. 2026-11-01 20:20 · Phase 2 R46 CricketOne

     Drop and ball test plan for rev S: sim/tube-r46.mjs 13/13; t1_test.py and drop_test.py --rev s (98 hits, head last, looks by hand); finding: a ball on the head breaks the printed spigot (411 vs 111 N·m at 40 m/s, est.), so R47 (spigot sleeve) added and one spare head in order 2; R3: test pultruded first. Folds none.

150. 2026-11-01 21:25 · Phase 2 R48 CricketOne

     Head collar parts and a side button hole: collar (6061-T6, 64.5 g) and printed top (37.1 g) drawn as two parts with STEP files; hole at +x; sim/side-r48.mjs 12/12 (ball from the bowler ×1.22, was ×0.59; shielded ±28.8°; button 1,240 N; bond ×1.29 est.; Wi-Fi clear); maker files 33/33; R49 added (off-stump mirrored collar, bond lap test); FIRST-STUMP.md section 7: side hole, collar job

     Folded none.

151. 2026-11-01 21:55 · Phase 2 R49 CricketOne

     Mirrored off-stump collar (collar-off.step, button at -x) and the core key moved to the front so one core fits every stump; bond test designed (sim/lap-r49.mjs 10/10): pass mark 6.98 MPa, published PA12 glue 5.0 MPa gives ×1.08, so coupons, wet/hot and a 15.8 kg joint check before T1 (BRINGUP.md 4b). Next: R50 camera view at z 662.

152. 2026-11-01 22:20 · Phase 2 R50 CricketOne

     Head camera view at z 662 with the wicket in the picture: sim/view-r50.mjs 19/19; bails, other stumps and own collar hide 0 %; far wicket hides nothing called; finding: rev S heads drawn straight see 74.4 % of the popping crease (P28 12° turn: 100 %) and 77.8 % of the front-foot zone (98.9 %); tilt band 6.9-17.3°; first stump unchanged; R51 added (draw the turned outer heads). Folds none.

153. 2026-11-01 23:15 · Phase 2 R52 CricketOne

     Lens numbers in one place: build/cad-s/lens.json read by the rev S CAD, the R51 trace, view-r42 and view-r50 through lens.mjs (same results with est. numbers); new R52 head fit in the CAD and lens-check.mjs (PASS or what to change; --grid table in BRINGUP.md section 2, now the rev S caliper form). Finding: tilted 12°, the est. lens front meets the window disc (top edge y 15.71 vs ledge 15.6), needs the bracket 0.45 mm back; only a 7.0 mm lens with a Ø6-7 barrel fits as drawn. First stump order 1 unchanged; R53 added (bracket from the measured lens). Folds none.

154. 2026-11-01 22:55 · Phase 2 R51 CricketOne

     Outer heads' cameras turned out 12° in rev S: head-side-top (leg) and head-off-top (off, mirror) drawn; window trace clear for both lenses, nearest fixed feature 2.75 mm; head-off-top.step; view-r50 21/21 from the CAD (crease 100 %); first stump unchanged; R52 added (lens numbers in one place for the 2 Nov calipers). Folds none.

155. 2026-11-01 23:55 · Phase 2 R53 CricketOne

     Camera bracket set back 0.45 mm for the est. lens in all three rev S head tops (lens.json bracket.backMm): lens 0.3 mm off the window disc, lens-check PASS, window traces and views unchanged, head STEP files re-made; lens-check now prints the one set-back a measured lens needs (a 7.0 mm lens needs 0); FIRST-STUMP.md section 7: hold head-top until PASS with the real lens; R54 added (board numbers into the head). Folds none.

156. 2026-11-02 00:40 · Phase 2 R54 CricketOne

     Camera board numbers into the head: lens.json board (est. 24 x 25 x 6, square) sizes the board, bracket and corner pockets in all three rev S tops; new board fit (tilted 12° with the lens, 0.3 mm to the printed parts, 1.5 mm skin); finding: the corner pockets left 0.95 mm of PA12 outside them, now 1.85 mm; boards up to 25 mm wide fit; lens-check PASS, maker files 39/39, head STEP files re-made (FIRST-STUMP.md section 7, hold unchanged); R55 added (ball on the head's camera side). Folds none.

157. 2026-11-02 00:55 · Phase 2 R55 CricketOne

     Ball on the head's camera side: sim/head-hit-r55.mjs 10/10; the resin/PC window (×0.19) and the 1.85 mm pocket skin (×0.46) break first, then the glue line to the collar (118 N·m for a ball at the window, best adhesive ×0.77); glass window ×1.54; proposal R56 (collar to z 690 in 6061-T6, glass window, +27 g est.); BRINGUP.md 4c head-hit marks and looks; FIRST-STUMP.md section 7; row R55 [x], R56 added

158. 2026-11-02 01:55 · Phase 2 R57 CricketOne

     Collar making and cost: sim/collar-r57.mjs 10/10; the board-corner pockets inside the bore cannot be cut with standard tools and the tilted bracket needs a 5-axis mill (est. $851-1,710 one-off); a 24 mm board needs no pockets; proposal R58: turned collar with a drop-in bracket ($427-775 one-off, $94-204 at 30, est.); R48 $40-90 was too low; FIRST-STUMP.md section 7; row R57 [x], R58 added

159. 2026-11-02 02:25 · Phase 2 R58 CricketOne

     Turned collar with a drop-in camera bracket: the three collars are a plain turned bore with a lip (z 647-649) and a key slot open to the top; the bracket is its own 6061 part (7.8 g, one for all three stumps) held by the cap plug; corner pockets removed (24 mm board 0.35 mm to the plain bore); lens-check PASS, view-r50 21/21, head-r56 12/12, collar-r57 10/10, maker files 45/45 (new bracket.step), check-assembly clean. R1 still waits for the Pi (due 6 Nov). Audit: R57 output present. Schedule on time. Next: R59 camera aim with the drop-in bracket.

160. 2026-11-02 02:50 · Phase 2 R59 CricketOne

     Camera aim with the drop-in bracket: sim/aim-r59.mjs 10/10; loose play ±0.90° turn, +1.53/-0.51° tilt; offsets inside the view margins and measured by the re-aim, but on its own weight the bracket rocks at 0.031 g and a loose bracket makes 0 % of calls sure (+577 % error); silicone 50A half-ring pad under the cap plug drawn (47-94 N, crown top 680.5), bracket.step re-made, BRINGUP.md 4d, FIRST-STUMP.md section 7; R60 added.

161. 2026-11-02 03:25 · Phase 2 R60 CricketOne

     PA12 or 6061 camera bracket: sim/bracket-r60.mjs 10/10; a printed bracket's feet on the 0.5 mm lip flatten under the pad when warm (86.2 MPa, 0.598 mm in a season, no preload left) and it lets the camera tilt 0.01° at 4 g; 6061 holds (38 MPa, 0.01° at 299 g). PA12 or no bracket for B1, 6061 for bail-off, soak, drop and T1 (FIRST-STUMP.md section 7, BRINGUP.md 4d). Audit: R59 output present. Schedule on time. Next: R61 silicone pad or a wave spring.

162. 2026-11-02 03:55 · Phase 2 R61 CricketOne

     Silicone pad or a wave spring under the cap plug: sim/pad-r61.mjs 10/10; the 50A pad keeps 36.5 N after a season at 50 °C and 12.2 N in the worst case (needs 3.2 N); a wave spring would dent the printed plug, rock on the half-ring crown and is not a catalogue part. Silicone stays, sheet with set ≤ 25 % (FIRST-STUMP.md section 7, BRINGUP.md 4d). Audit: R60 output present. Schedule on time. Next: R62 buyable silicone sheet and a clean half-ring pad.

163. 2026-11-02 04:40 · Phase 2 R62 CricketOne

     Buyable silicone sheet and a clean half-ring pad: sim/pad-r62.mjs 10/10; three sheets leave two holes in the gap range, closed by 0.1/0.2 mm shims; a sheet's ±0.30 mm tolerance means picking by the measured sheet; stocked sheets do not state the 22 h / 175 °C set, so order 2 asks for the test report (S50FDA 19.6 % the published example); compass cutter, not punches (FIRST-STUMP.md section 7, BRINGUP.md 4d step 1). Audit: R61 output present. Schedule on time. Next: R63 the squeezed pad and what sits inside it.

164. 2026-11-02 04:55 · Phase 2 R63 CricketOne

     The squeezed pad and what sits inside it: sim/pad-r63.mjs 9/9; nothing presses on the camera cable (ribbon 13.5 mm below, Hall wires 2.34 mm); finding: the centre-cut pad overhung the crown's front ends and at 30 % squeeze a third of it rolled over a milled edge; crown in to r 12.8 and the pad halved 3 mm behind the centre (bracket.step re-made, FIRST-STUMP.md section 7); 250 N clamp to seat the cap flat (BRINGUP.md 4d). Audit: R62 output present. Schedule on time. Next: R64 a press cup for the 250 N cap clamp.

165. 2026-11-02 05:30 · Phase 2 R64 CricketOne

     A press cup for the cap clamp: sim/clamp-r64.mjs 10/10; printed press cup bears only on the two lands beside the bail groove (0.204 mm smallest gap, 1.6 mm clear of the groove edge, 5.6 mm from the Hall pockets) and a foot ring lets the camera ribbon out (build/cad-s/press, STL, PETG, no supports); all parts take a 6 in one-hand bar clamp's full 1,334 N; finding: with the jaw 1.7 mm off the axis the cap needs 303 N, not 241, so the bench squeezes firmly and the feeler proves it (BRINGUP.md 4d step 2, FIRST-STUMP.md section 7: one clamp for order 2). Audit: R63 output present. Schedule on time. Next: R65 the adhesive for the cap bond and the clamp time.

166. 2026-11-02 05:55 · Phase 2 R65 CricketOne

     The adhesive for the cap bond: sim/bond-r65.mjs 10/10; four adhesives from their data sheets on the PA12 and anodised sides, at 23 °C and 60 °C; first pick 3M DP8910NS, second DP8010 Blue (L1 coupons decide), the epoxy drops out; finding: R49's DP8910NS number was too high (6.35 MPa, not 8.1) and both acrylics lose much of their strength at 60 °C (×0.63-0.78 hot, est.); new coupon pass marks for the R56 joint, prep, clamp 24 h (BRINGUP.md 4b and 4d, FIRST-STUMP.md section 7). Audit: R64 output present. Schedule on time. Next: R66 how hot the cap's glue line gets.

167. 2026-11-02 06:20 · Phase 2 R66 CricketOne

     How hot the cap's glue line gets, and the screw lock: sim/heat-r66.mjs 10/10; the glue line runs at the collar's temperature, 21 K over the air on a blue collar in the sun and 12 K on a white one; finding: by the data sheets the bond drops under ×1.0 at 40 m/s above about 27 °C air even with a white collar; R49's M3 lock does not fit, so four countersunk M2 go through the collar wall into the cap's plug (drawn, collar and top files re-made); they keep the cap on but do not reach ×1.0 alone; collar white or clear, not blue (BRINGUP.md 4b step 4, FIRST-STUMP.md section 7). Audit: R65 output present. Schedule on time. Next: R67 a cap joint that holds on a hot day.

168. 2026-11-02 06:50 · Phase 2 R67 CricketOne

     A cap joint that holds a 40 m/s ball on a hot day: sim/joint-r67.mjs 10/10; a longer plug, a neck thread and more screws compared in the rev S CAD; the cap's plug now runs 11 mm, not 8, with the bracket's crown 3 mm lower (drawn, top and bracket files re-made); finding: with a white collar the bond holds ×1.0 up to about 41 °C air in full sun (was 27 °C); the thread (the thin skirt bursts first) and more screws (they carry nothing until the glue breaks) drop out; R66's four screws kept; new coupon pass marks (BRINGUP.md 4b, FIRST-STUMP.md section 7: re-upload two files, no new parts). Audit: R66 output present. Schedule on time. Next: R68 the pad and the press cup on the R67 head.

169. 2026-11-02 08:30 · Phase 2 R70 CricketOne

     Slanted Hall holes for a gentle wire bend: sim/wires-r70.mjs 7/7; each Hall hole now slants back from its pocket so the wires leave the cap behind the camera bracket's plate and bend at 5 x OD or more (was 1.9 mm); finding: R69's corridor point put the hole 0.68 mm from a lock screw's tapped hole, so the holes are pulled in (1.39 mm); finding: R69's 170 mm wire cut was short, now 190; three tops re-made, maker files 45/45, R68 10/10 (FIRST-STUMP.md section 7: re-upload head-top, no new parts). Audit: R69 output present. Schedule on time. Next: R71 powder out of the slanted holes.

170. 2026-11-02 09:20 · Phase 2 R72 CricketOne

     Loose wires or a 3-core cable through the Hall hole: sim/cable-r72.mjs 6/6; four options from the makers' pages; round 3-core cables (3.91 mm) and silicone ribbon (2.53) do not pass the Ø2.5 hole; a 3M 3756 flat strip passes (2.01) but its 0.64 mm conductors are under the GH crimp's 0.8-1.0 range; three loose 30 AWG silicone wires kept (0.78 mm room, no part change); new bench steps: tack the Hall, tin the ends into one tip, draw the wires by hand (0.78 N at the pocket, est.); R70 re-run 7/7 (FIRST-STUMP.md section 7: no change). Audit: R71 output present. Schedule on time. Next: R73 the Hall's tack and its leads under a ball.

171. 2026-11-02 09:50 · Phase 2 R73 CricketOne

     The Hall's tack and its leads under a ball: sim/tack-r73.mjs 8/8; LOCTITE 454 gel chosen (stays 1.12 mm above the hole mouth, creeps 0.66, est.), DP8910NS fallback, light-cure acrylic out; two steps: tack before the cap bond, pot after the clamp; ball on the cap 1.57 N on a joint (x3.2), landing x6.9; finding: whole-wire worst bound x0.54 bare, so the pocket is potted (x1.2 copper, x1.66 pot); FIRST-STUMP.md section 7 adds LOCTITE 454 to order 2. Audit: R72 output present. Schedule on time. Next: R74 rain in the bail groove.

172. 2026-11-02 10:30 · Phase 2 R74 CricketOne

     Rain in the bail groove: sim/rain-r74.mjs 8/8; the level groove sheds water over both open ends and holds at most a 1.1 ml puddle (est.); a wet bail still falls; every path to the hollow followed (pot, cap seam, lock screws, vent); finding: the pot sits under standing water, so it moves to the cap's DP8910NS (which our L2 coupons soak in water); finding: R66's dry lock screws were a wet path into the cap, now sealed with the same shot; IP target IP65 and IP67; no drain, no CAD or part change (FIRST-STUMP.md section 7). Audit: R73 output present. Schedule on time. Next: R75 the breather vent under a hose.

173. 2026-11-02 11:00 · Phase 2 R75 CricketOne

     The breather vent under a hose: sim/vent-r75.mjs 9/9; the vent is a Gore VE90308 (3.3 × 7.6 × 0.34 mm, silicone adhesive, IP68 inside) from Gore's own data sheet; finding: the round bore bows 0.46 mm under it, so each collar gets a small flat at the back of its bore (drawn, three collar files re-made); the vent hole tilts 15° down to the outside so a stuck drop runs out; it breathes at about 5 mbar on a cold shower (est.); finding: no data sheet covers a hose jet on an inside vent, so T5 aims its IPX5 jet at it and the care note says never hose (FIRST-STUMP.md section 7: re-upload the collar, add the vent to order 2). Audit: R74 output present. Schedule on time. Next: R76 the thinner back wall under a ball.

174. 2026-11-02 11:35 · Phase 2 R76 CricketOne

     The thinner back wall under a ball: sim/backwall-r76.mjs 9/9; finding: R75's one pass of a 1 in keyseat cutter would bulge past the flat and leave 0.79 mm of wall, so the flat is cut with a Ø6.35 keyseat cutter in three passes (wall 2.41 mm; a long-reach special, or EDM), drawn with its round ends, three collar files re-made; a 40 m/s ball on the cap ×2.09 to yield; a throw on the back still breaks the tube first; finding: the vent hole on the back is the wall's fatigue limit, about 6 seasons at the bound (est.), so R77 moves it toward the side (FIRST-STUMP.md section 7: re-upload the collar, cutter note). Audit: R75 output present. Schedule on time. Next: R77 the vent hole off the back.

175. 2026-11-02 11:55 · Phase 2 R77 CricketOne

     The vent hole off the back: sim/venthole-r77.mjs 8/8; any-angle bending reproduces R76 on the back (132 MPa, 6 seasons); a sweep round the collar picks 40° toward the button side, the smallest angle that lasts twice our 10-season aim at the bound (about 21 seasons, est.), keeps the camera bracket's arms 4.1 mm away and stays 16.6 mm from the groove's run-off; finding: the button side, because the outer stumps' arms turn out with their cameras; three collar files re-made (FIRST-STUMP.md section 7: re-upload the collar, flat at 40°). Audit: R76 output present. Schedule on time. Next: R78 the lock-screw holes under a season of hits.

176. 2026-11-02 12:30 · Phase 2 R78 CricketOne

     The lock-screw holes under a season of hits: sim/lockholes-r78.mjs 7/7; the four M2 holes sit 16.5 mm under the hit, so every one lasts millions of seasons at the bound (est.) and static ×6.1; the 1.15 mm neck behind the key slot outlasts the stump; finding: each countersink ran into the neck's shoulder and the cap skirt's glued foot, so the screws drop 0.6 mm to z 683.5, six files re-made (FIRST-STUMP.md section 7: re-upload collar and top, CNC note); finding: R66 check 8 reads out of date since R67 (R79). Audit: R77 output present. Schedule on time. Next: R79 R66's screw-lock check against the R67 head.

177. 2026-11-02 13:05 · Phase 2 R79 CricketOne

     R66's screw lock against the R67 head: sim/heat-r66.mjs 12/12; the lock's real job is to keep the cap captive once the glue lets go, and on any day each screw crushes the plastic before it tears out (×2.13) or shears (×2.0 or more, est.); with no glue the 11 mm plug and four screws take a full 40 m/s ball only on a cool day (×1.03 at 23 °C, ×0.67 white at 40 °C), so the glue stays the first line; M2.5 left out (countersink back at the shoulder); no part change; BRINGUP.md 4b step 4 adds a thumb rock check (FIRST-STUMP.md section 7: no change). Audit: R78 output present. Schedule on time. Next: R80 can the stump tell a loose cap by itself.

178. 2026-11-02 13:20 · Phase 2 R80 CricketOne

     Can the stump tell a loose cap by itself: sim/loosecap-r80.mjs 9/9; no sensor sees the rock itself (the Hall bail switches ride on the cap, the camera bracket stays held on the collar under a full 1.04° rock, the motion sensor saturates on every ball); the aim check is the backstop if a lifting cap frees the bracket; new software/capcheck.py asks for the umpire's thumb check after a tracked ball on the cap at 30 m/s or more (est.) or a knock with no tracked ball, with camera calls unsure until answered; no part change; BRINGUP.md 4d step 5 (FIRST-STUMP.md section 7: no change). Audit: R79 output present. Schedule on time. Next: R81 join the cap check to the match program.

179. 2026-11-02 13:55 · Phase 2 R81 CricketOne

     The cap check joined to the match program: sim/stump-r13.mjs 28/28 (part F, 7 new checks); the tracker's ball message carries the ball's height and speed at the stumps (track2d.at_stumps), the heartbeat carries the re-aim offset, stump.py's router feeds them and every knock into CapCheck, asks are signed "ask" calls, ball calls unsure while a cap-hit or aim-jump is open, the umpire answers over POST /cap on call_link.py; no part change; BRINGUP.md 4d step 6 (FIRST-STUMP.md section 7: no change). Audit: R80 output present. Schedule on time. Next: R82 the re-aim offset in the eyes.

180. 2026-11-02 14:08 · Phase 2 R82 CricketOne

     The re-aim offset in the eyes: sim/reaim-r82.mjs 10/10; clip_keeper.py finds the far wicket's three stump tops (15 px patches over ±16 px, sub-pixel refined, 0.047 px a frame on rendered 1.6 px tops) after a knock or a camera start, fits turn and tilt over 30 frames and sends "aim" to the cap check; a held camera reads 0.0011° (95 %) and never asks, a 0.3° jump is sent at once and stored after a second steady window, a loose bracket is never taken as steady; --set-aim stores the aim in lens.json; one stump alone has the aim-jump and knock asks, cap-hit waits for the pair; no part change; BRINGUP.md 4d step 7 (FIRST-STUMP.md section 7: no change). Audit: R81 output present. Schedule on time. Next: R83 find the far stump tops by themselves.

181. 2026-11-02 14:58 · Phase 2 R83 CricketOne

     Find the far stump tops by themselves: sim/stumptops-r83.mjs 7/7; clip_keeper.py --set-aim auto searches the band the rev S CAD and Law 8 give (±4°), scores rows for three 1.6 px bars 4.4 px apart, takes the run's top and keeps the wicket only when 3 looks a second apart agree within 1 px; with a sight-screen (sun, shade), an umpire behind, a fielder beside or a 1° lean every kept mark is within 1.3 px; finding: our white heads vanish on a white sight-screen, so marks sit at the heads' lower edge (5 px down, still a fixed point); finding: a stand's shadow can pull marks 13 px down the stumps; a crowd and no sight-screen asks for the pixels; never a wicket where there is none; no part change; BRINGUP.md 4d step 7 (FIRST-STUMP.md section 7: no change). Audit: R82 output present. Schedule on time. Next: R84 our heads on a white sight-screen.

182. 2026-11-02 15:20 · Phase 2 R84 CricketOne

     Our heads on a white sight-screen: sim/heads-r84.mjs 8/8; four paints in R83's scenes with blue block bails: white as drawn, a carbon-black ring above the glue line, the same in cyan, the whole head cyan; only the whole head shows its top on every screen (spread 1.17 px), but its lower 99 mm is the white collar that sets the glue line's heat (R66/R67), so it is out; a 0.67 px ring is too short for the finder's 20-row run (still 127 mm low with an umpire behind) and warms the glue about 1.3 K (est.); Law 8 sets no colour; the heads stay white, no part, print or paint change; BRINGUP.md 4d step 7 adds a grey-level look (FIRST-STUMP.md section 7: no change). Audit: R83 output present. Schedule on time. Next: R85 tell the head's edge from the top with the bail line.

183. 2026-11-02 16:07 · Phase 2 R86 CricketOne

     A weak bail line in deep shade, at dusk and under floodlights: sim/bailline-r86.mjs 9/9; clip_keeper.py bail_line_weak: when the 30-frame line is unsure, 60 then 90 frames, then a 3-level line only when three 90-frame looks agree; 0 wrong in 103 wickets found (22 scenes); more frames alone never helped (the bail's contrast is the limit, not the noise); three agreeing looks found wooden bails at dusk 3 in 5 (R85 rule 0); floodlights as in sun; our stumps on a black screen at dusk still unsure (set the aim in daylight or under lights); the bails' ends at the outer stumps hold no line (Law 8), dropped; set-up about 7 s at worst on the Pi Zero 2 W (est.); no part change; BRINGUP.md 4d step 7 dimmed-light look (FIRST-STUMP.md section 7: no change). Audit: R85 output present. Schedule on time. Next: R87 our stumps on a black screen at dusk.

184. 2026-11-02 16:54 · Phase 2 R87 CricketOne

     Our stumps on a black screen at dusk: sim/bailline-r87.mjs 12/12; clip_keeper.py measure_light reads the grass under the far wicket and head_step_for sets the white-top step to 20 x the light, never under 6 sd of the noise (sun 20, shade 9, floodlights 12, dusk 4.9); ours on black at dusk top 5 in 5 within 1.19 px (was unsure 5); dark-screen unsure 12 -> 1; 0 wrong in 137 found (31 scenes, shadows, bails off, fielders, umpire); the 5-level bail line kept (already 6.1 sd at dusk); finding: R82's daylight patches stopped matching at 0.6x light (0/12), so match_mark is now a gain-and-offset match (12/12 from sun to 0.1x on our stumps, worst 0.015°; a fielder in whites in front refused); late dusk 0.1x is now the finder's limit (1 in 5); no part change; BRINGUP.md 4d step 7 black-card dimmed step (FIRST-STUMP.md section 7: no change). Audit: R86 output present. Schedule on time. Next: R88 when the light goes.

185. 2026-11-02 17:20 · Phase 2 R88 CricketOne

     When the light goes: sim/light-r88.mjs 10/10; light swept from sun to 0.05x (est.): set-up finds our far wicket to 0.2x (wooden 0.3x), the stored aim re-aims after a knock to 0.07x, the ball tracker keeps a white ball to 0.05x and a red ball to 0.2x (96 %, 0 wrong); never a wrong far wicket or offset; finding: track2d's median-of-whole-grey-levels threshold let noise pass as a ball at about 0.45x (28 wrong fixes in 99 frames), now 5.6 x the mean difference (0 wrong; R1 sim re-run 100 % in sun); new light rule: clip_keeper.py sends the light each second, stump.py makes ball calls unsure with why "light" under the limit for the ball in use (red 0.2, white 0.05, pink as red, est.), notes light-low, light-back, light-falling, light-set-aim, bail calls untouched; no part change; BRINGUP.md 4d step 7 dimmer step (FIRST-STUMP.md section 7: no change). Audit: R87 output present. Schedule on time. Next: R89 the light from the camera itself.

186. 2026-11-02 18:40 · Phase 2 R90 CricketOne

     The red ball in poor light: sim/redball-r90.mjs 8/8; finding: more gain only lifts a faint ball over the tracker's 8-level floor, past about 8x it scales the ball and the noise alike (8x 92 %, 15.5x 90 % at 0.2x), so a smaller gain step cannot help; a longer exposure first (100 to 800 µs), then gain, follows the red ball to 0.05x (was 0.3x) and the white to 0.05x (was 0.07x); a 45 m/s ball smears 1.4 px at 800 µs, found 100 %, 0.14 px rms; the fix sits at the exposure's middle, so camera.py adds STAMP_EDGE and Frame.t_mid_ns and the pair need not share the darker light (end stamps would put them up to 350 µs, 15.8 mm apart); a two-frame average in the tracker gains one light step at best, left out; camera.py set_exposure(), clip_keeper.py bright_plan() and heartbeat "exposure", stump.py limits red 0.05 white 0.05; no part change; BRINGUP.md 4d step 8 (FIRST-STUMP.md section 7: no cart change). Audit: R89 output present. Schedule on time. Next: R91 the longer exposure for the far wicket's set-up and re-aim.

187. 2026-11-02 19:05 · Phase 2 R91 CricketOne

     Setting the aim in poor light: sim/aim-r91.mjs 10/10; R88's set-up and re-aim re-run on R90's real picture (exposure 100 to 800 µs first, then gain, est. noise); finding: --set-aim opened the camera at 100 µs and 4x with no loop, so it found our far wicket only to 0.1x (wooden 0.3x); clip_keeper.py settle_bright() now walks the brightness on the grass first, and our tops and wooden ones are found to 0.03x (the lowest tried; R88 0.2x), the re-aim holds to 0.03x (R88 0.07x), the light read back follows the field within 10 %; a knock ringing up to 0.6° at 40 Hz during an 800 µs exposure smears the far tops 1.9 px, 0 marks refused, every stored aim within 0.012° of rest, a 0.29° jump still flagged in the first window; 0 wrong wickets in 84 runs; stump.py aim_warn 0.3 -> 0.045 (after light-low now, the ball calls stop first); no part change; BRINGUP.md 4d step 9 (FIRST-STUMP.md section 7: no cart change). Audit: R90 output present. Schedule on time. Next: R92 the set-up brightness against a bright sky.

188. 2026-11-02 19:20 · Phase 2 R92 CricketOne

     The set-up brightness against a bright sky: sim/sky-r92.mjs 8/8; the sky above the sight-screen is 20 % of the picture at 12° down; finding: R91's whole-picture stand-in let the sky hold the brightness back, so at 8x the grass the grass settles at 29 of 105 and at 16x at 14 (0.05x light), and wooden bails on a white sight-screen fall under the bail line's 5-level floor (sure tops only to 0.3x at 8x, not even 0.3x at 16x; our stumps hold to 0.03x on the heads' step); clip_keeper.py finder_window() and sky_rows(): the stand-in is read on R83's finder window less rows clipped or over 2.5 x the grass, both kinds to 0.03x with any sky 1.9x to 16x; finding: skipping rows over 230 stalled on the screen's blurred edge (grass 27 at 16x), so rows are judged against the grass; in sun our heads stay under the clip (172); 0 wrong in 768 wickets kept; no part change; BRINGUP.md 4d step 10 (FIRST-STUMP.md section 7: no cart change). Audit: R91 output present. Schedule on time. Next: R93 the bail line's floor in grey levels.

189. 2026-11-02 19:55 · Phase 2 R93 CricketOne

     The bail line's floor in grey levels: sim/bailfloor-r93.mjs 8/8; picture held darker (grass 105 to 14 of 105) and out of gain (0.03x to 0.01x) on 13 scenes from R83 to R92; finding: R85's fixed 5-level floor, not the picture, lost wooden bails on a white screen at grass 14 (0 and 1 of 6 sure), where the line still stood 10-20 sd over the noise; clip_keeper.py noise_sd() reads the noise on R87's grass rows (0.95 to 1.10 of the true value) and line_floor() asks max(1, 8 sd), third step max(1, 5 sd): sure at every darker level, ours and wooden; pure noise made at most 4.1 sd; 503 sure, 0 wrong where the finder was on the stumps (fixed 471); out of gain both to 0.01x; finding: a crowd with no sight-screen let the finder keep marks 2.3-2.8 px to the side in 2 of 26 (fixed 3), so R94; no part change; BRINGUP.md 4d step 11 (FIRST-STUMP.md section 7: no cart change). Audit: R92 output present. Schedule on time. Next: R94 a crowd behind the far wicket with no sight-screen.

190. 2026-11-02 20:20 · Phase 2 R94 CricketOne

     A crowd behind the far wicket with no sight-screen: sim/crowd-r94.mjs 7/7; 18 scenes (R93's plus 8 crowds) at 6 lights, 444 set-ups, 400 wickets found, every finding counted; finding: a still crowd carried the finder's marks half or a whole spacing to the side in 24 of 236 crowd set-ups (2-4.7 px), and a shadowed wooden wicket once a whole spacing, with 23 wrong tops (up to 9.6 px); the three stumps' ridges share a sign row by row (on the stumps at least 95 % of rows, off them 5-65 %): at 80 % the sign test refused 25 of 25 off and kept 163 of 163 on; crowd rows could still pass for the bails, so the background beside the wicket is measured over the picture's own noise (no crowd at most 1.7, crowds at least 2.5) and over 2.0 the set-up says "cannot tell"; crowd wrong tops 23 -> 0, screens, shadows, umpire and fielder 139 sure before and after; Law 8 spacing is no check (the finder lays it itself); clip_keeper.py on_stumps(), busy_beside(), FIND col_agree 0.8 col_busy 2.0; no part change; BRINGUP.md 4d step 12 (FIRST-STUMP.md section 7: no cart change). Audit: R93 output present. Schedule on time. Next: R95 tapping the far tops in the app.

191. 2026-11-02 21:00 · Phase 2 R95 CricketOne

     Tapping the far tops in the app when the set-up cannot tell: sim/tap-r95.mjs 8/8; 14 scenes at 4 lights, 168 scorer taps, 1,176 wrong taps; finding: one tap on the finder's whole window (0.44 mm a px) lands past half a stump spacing 144 of 168 times, so the app zooms to 32 px round the first tap (95 % within 1.87 px); the stump refines only 2 px across and 8 px up or down, then the top comes from R94's sign test and R85's bail line on a quiet background or our three white heads in front of a crowd, and the tap must sit within 2.5 px of it; 0 wrong tops from scorers' taps (quiet 71 of 84 stored, bails off on a white screen refused; crowds 22 of 84); wrong taps: 0 stored wrong on quiet backgrounds, 2 of 588 on a still crowd passed a whole spacing off (10 mm on the glass), so crowd taps are stored only after the scorer confirms the drawn marks; dark crowd 12 of 12, light crowd 1 of 12, crowd with an umpire in white 0 of 12, wooden in crowds refused (R96); about 8 s at worst (est.); clip_keeper.py TAP, finder_at(), on_heads(), tap_marks(), --set-aim tap:U,V, --confirm; no part change; BRINGUP.md 4d step 13 (FIRST-STUMP.md section 7: no cart change). Audit: R94 output present. Schedule on time. Next: R96 a light crowd or an umpire in white behind the far wicket.

192. 2026-11-02 21:30 · Phase 2 R96 CricketOne

     A light crowd or an umpire in white behind the far wicket: sim/lightcrowd-r96.mjs 8/8; 17 scenes (R95's plus a very light crowd, a light crowd with an umpire in white, wooden on a light crowd) at 4 lights, R95's rule and the new one on the same pictures and taps (204 scorer taps, 1,428 wrong taps each); the blue tube under each head is darker than a light background in every row and meets the white head in a sharp edge one head-length (5.45 px) under the top: clip_keeper.py on_tubes() finds that edge a head-length under the tap (±3 px), all three tubes darker than both sides in 80 % of 20 rows, no second edge a head-length lower, top = edge less a head-length; light crowd 1 -> 12 of 12, very light 0 -> 12, crowd with an umpire in white 0 -> 12, light crowd with an umpire 1 -> 11, crowd taps stored 21 -> 72 of 120, 0 wrong tops (worst 1.4 px); wrong taps 1 of 1,428 a whole spacing off with either rule (the scorer confirms); findings: a tap a head-length high found the bail's top edge, 20 of 588 stored 7 px high, until the lowest-edge test; read on the column alone the light crowd's foot a head-length lower looked like a second edge, so both steps are read against the stump's own sides; mixed crowd 6 of 12 (R97); no time added; no part change; BRINGUP.md 4d step 14 (FIRST-STUMP.md section 7: no cart change). Audit: R95 output present. Schedule on time. Next: R97 a mixed crowd behind the far wicket.

193. 2026-11-02 21:55 · Phase 2 R97 CricketOne

     A mixed crowd behind the far wicket: sim/mixedcrowd-r97.mjs 8/8; 18 scenes (R96's plus a mixed crowd that never moves) at 4 lights, the two mixed crowds 40 set-ups each; finding: most refusals were the finder at the tap settling half a spacing to the side, so the heads and tubes were read in the gaps; clip_keeper.py SCAN, across_tap(): both cues at nine places across the tap, set only when every passing place agrees (1 px across, 1.5 px in the top); mixed crowd 32 -> 40 of 40, still 27 -> 38 of 40, crowd taps 123 -> 147 of 188, 0 wrong tops; tried and not kept: both cues row by row (24 of 40) and five looks 0.5 s apart (+2 s, little gain); wrong taps stored wrong 4 -> 13 of 1,904, the 7 from across the tap a whole spacing off; one 2.25 px wrong top by R95's heads on wooden stumps (R98); no time added; no part change; BRINGUP.md 4d step 15 (FIRST-STUMP.md section 7: no cart change). Audit: R96 output present. Schedule on time. Next: R98 one agreement test for every crowd cue.

194. 2026-11-02 22:32 · Phase 2 R98 CricketOne

     One agreement test for every crowd cue: sim/agree-r98.mjs 8/8; R97's 18 scenes at 4 lights, R97's rule and R98's on the same pictures and noise (272 scorer taps, 1,904 wrong taps each); clip_keeper.py AGREE, agree_tap(): in a crowd both cues at the finder's place and nine places across the tap, every passing reading kept, set only when all lie within 2 px across and 1.5 px in the top and no cue passes a whole spacing to either side; wrong taps stored wrong 3 -> 0, crowd taps 142 -> 136 of 188, 0 wrong tops; finding: R97's 1 px spread refused real wickets (they pass over about 1.5 px); two readings needed and a same-height side test tried, not kept; 16 of 152 taps on our stumps in crowds still refused (R99); no time added; no part change; BRINGUP.md 4d step 16 (FIRST-STUMP.md section 7: no cart change). Audit: R97 output present. Schedule on time. Next: R99 the taps still refused on our stumps in a crowd.

195. 2026-11-02 22:50 · Phase 2 R99 CricketOne

     The taps still refused on our stumps in a crowd: sim/retap-r99.mjs 8/8; R98's 18 scenes at 4 lights, R98's rule and two new rules on the same pictures (272 scorer taps, 1,904 wrong taps each), the second look a second later with the crowd moved and its own noise; clip_keeper.py RETAP, retap_look(): after a refusal in a crowd, one more 30-frame look read alone with R98's agreement test; our stumps in crowds 136 -> 141 of 152, 0 wrong tops, 0 wrong taps stored wrong; the two looks pooled set 138 (a false pass in the first look spoils the pool), not kept; a crowd that never moves gains nothing; 11 of 152 left (R100 the scorer's second tap); worst crowd tap 5.2 s (est.); no part change; BRINGUP.md 4d step 17 (FIRST-STUMP.md section 7: no cart change). Audit: R98 output present. Schedule on time. Next: R100 the scorer's second tap.

196. 2026-11-03 05:52 · Phase 2 R113 CricketOne

     The whole fit in the brain's own code: track2d.fit_ball(points, cams, p0) (Levenberg-Marquardt with the 2 ms step and 0.01 % stop, the 12 columns as one fit_flight(), ONE ending flight of the fit, its 12 shifts and the drift shift, call_sigma() and the rms gate) and nominal_cams(); sim/fitball-r113.mjs 8/8: track-sim.mjs SIM_FITBALL gives R112's calls, sigma and sure gate on all 100 (0 mm, 0 flips, 48 % sure, worst 7.7 / 18.5 mm); 29,213 C track steps (was 33,366), 16 flights, 3,484 numpy calls and 10 solves a ball; Pi 70 ms median / 167 ms worst (est.), inside 0.8 s; _fitball_selftest(), BRINGUP.md 4d step 31 (FIRST-STUMP.md section 7: no cart change). Audit: R112 output present (10/10). Schedule on time. Next: R114 the first guess on the stump.

197. 2026-11-02 23:38 · Phase 2 R100 CricketOne

     The scorer's second tap: sim/secondtap-r100.mjs 8/8; R99's 18 scenes at 4 lights, R99's first tap repeats (141 of 152), then up to three taps a scorer (a new finger draw about 2 s later, the crowd moved again, est.) and every refused wrong tap repeated at the same place (about 3,150 later wrong taps); the second tap sets 9 of R99's 11 (141 / 9 / 0), 0 wrong tops; finding: R99's rule stored 2 repeated wrong taps about 3 px off, so clip_keeper.py later_tap() needs a later crowd tap's two looks to agree (0 stored wrong, the same 9); TAPS, tap_count(), after_refusals(): after two refusals the app says what to change; a guide dot tried, no gain; fix: the confirm for crowd taps had stopped firing since R98, now fixed; 2 of 152 left (a still crowd at 0.05x, R101); slowest set aim 7.9 s (est.); no part change; BRINGUP.md 4d step 18 (FIRST-STUMP.md section 7: no cart change). Audit: R99 output present. Schedule on time. Next: R101 a still crowd in poor light.

198. 2026-11-02 23:58 · Phase 2 R101 CricketOne

     A still crowd in poor light: sim/stillcrowd-r101.mjs 8/8; R100's 18 scenes at 4 lights, R100's totals repeat (141 / 9 / 0, 2 never); four later-tap rules on the same fingers, crowds and noise: 30 frames (kept), 60 and 90 frames with the floor read for them, 90 frames with the 30-frame floor; finding: none sets the 2 left (a mixed crowd that never moves at 0.05x: one fits a wicket a stump to the side in every look, one hides the heads), the still crowd's pattern is the limit, not the noise; 0 wrong tops, 0 of about 3,157 repeated wrong taps stored wrong under every rule; up to 1 s more; clip_keeper.py LATER (30 frames kept), keep_last_aim(): after two refusals the last aim stored that day stays in use, else ball calls unsure until an aim is set; no part change; BRINGUP.md 4d step 19 (FIRST-STUMP.md section 7: no cart change). Audit: R100 output present. Schedule on time. Next: R102 how long a stored aim holds.

199. 2026-11-03 01:10 · Phase 2 R103 CricketOne

     The roll the re-aim cannot see: sim/roll-r103.mjs 8/8; 7 rolls (to 0.48°) x 4 lights x 30 frames; the far stumps' slant (ours bend in the same sun; wooden too thin, pixel-locked: 0.38° to 1.6°) rejected; the near popping crease kept (worst 0.016° with the batter on it, flag-equivalent 0.061°), turn taken out; clip_keeper.py ROLL, crease_roll(), heartbeat roll; BRINGUP.md 4d step 21; FIRST-STUMP.md section 7 no cart change; audit OK, on time; next R104 the tracker uses the roll.

200. 2026-11-03 01:40 · Phase 2 R104 CricketOne

     The tracker uses the roll: sim/trackroll-r104.mjs 8/8; six cases on the P1 track sim, the same 100 deliveries; read once at set-up the afternoon sun's 0.48° roll puts every sure call 36 mm off in line, all outside their own 2-sigma, unseen by the fit's residual (0.32 px under 0.35); read at every re-aim line 3.1 -> 4.2 mm rms, as many sure calls; a 0.061° old roll keeps sure calls within 13.5 mm; track2d.unroll(), stump.py roll_age 34 min (why "roll"); no part change; BRINGUP.md 4d step 22 (FIRST-STUMP.md section 7: no cart change). Audit: R103 output present. Schedule on time. Next: R105 the far end's roll over the pair link.

201. 2026-11-03 02:00 · Phase 2 R105 CricketOne

     The far end's roll over the pair link: sim/pairroll-r105.mjs 8/8; five cases on the P1 track sim, the same 100 deliveries as R104; a lost partner roll leaves sure calls 35 mm off (19 of 19 outside 2-sigma, residual 0.38 px), the wrong end's roll 71 mm, both rolls sent 4.2 mm rms, the partner's roll at 34 min within 14.1 mm, the striker's end alone 16 % sure and none over 25 mm; the link's delay moves the roll under 0.0001°; call_link.py beacon roll/roll_s, stump.py roll.json and why "partner-roll", track2d.pair_points()/partner_points(); no part change; BRINGUP.md 4d step 23 (FIRST-STUMP.md section 7: no cart change). Audit: R104 output present. Schedule on time. Next: R106 the ball points over the pair Wi-Fi.

202. 2026-11-03 02:20 · Phase 2 R106 CricketOne

     The ball points over the pair Wi-Fi: sim/pairpoints-r106.mjs 9/9; one delivery's far points are 203 (246 most), 5.1-6.2 kB, 5 IP fragments; over 20 simulated club days one message per ball never arrives whole on 4 % of balls, and Wi-Fi power save leaves only 75 % in by a 0.1 s wait; parts of 0.05 s of flight, each sent twice 20 ms apart with power save off, 99.8 % in by 0.1 s (p99 32 ms, 0.22 % lost in Wi-Fi gaps), sure calls 51 % and 4.2 mm rms as both ends every ball; a 1 s wait gains nothing; no mix over 25 mm; track2d.pair_parts()/PartnerBall (UDP 5607, 0.1 s wait then "ends": 1), pair_wifi.sh powersave 2, track-sim.mjs SIM_PER; no part change; BRINGUP.md 4d step 24 (FIRST-STUMP.md section 7: no cart change). Audit: R105 output present (8/8). Schedule on time. Next: R107 a hole in the far track.

203. 2026-11-03 02:40 · Phase 2 R107 CricketOne

     A hole in the far track: sim/hole-r107.mjs 9/9; the P1 track sim on the same 100 deliveries with one 0.05 s part of the far points taken out (track-sim.mjs SIM_HOLE): a middle part 50 % sure, worst 19.1 mm line / 24.8 mm height; the bounce part 47 %, 10.9 / 19.7 mm; the first part 48 % but 24.6 / 31.8 mm (refused); the last part 40 %, safe but unusable (it carries "last"); over R106's 20 club days 0.08 % of balls have exactly one part missing, sure calls 50.96 -> 50.98 %; track2d.PartnerBall.points() fits with a middle or bounce hole and marks it, ball_message "hole"; no part change; BRINGUP.md 4d step 25 (FIRST-STUMP.md section 7: no cart change). Audit: R106 output present (9/9). Schedule on time. Next: R108 sure calls outside their own 2 sigma.

204. 2026-11-03 03:30 · Phase 2 R108 CricketOne

     Sure calls outside their own 2 sigma: sim/sigma-r108.mjs 10/10; the P1 track sim on the same 100 deliveries with each camera's leftover roll (0.013°) and clock (0.05 ms) carried through the fit camera by camera into the call's sigma (track-sim.mjs SIM_SIGMA): sure calls 51 -> 48 %, worst 20.3/26.4 -> 7.6/19.8 mm, every sure call inside 25 mm; outside their own 2 sigma 45 -> 31 % (28 % with twice the roll), median sigma +1 %: the rest is the changing swing and seam kick the fit does not model; track2d.call_sigma()/is_sure() (SURE rule); no part change; BRINGUP.md 4d step 26 (FIRST-STUMP.md section 7: no cart change). Audit: R107 output present (9/9). Schedule on time. Next: R109 the swing that changes in flight.

205. 2026-11-03 03:55 · Phase 2 R109 CricketOne

     The swing that changes in flight: sim/swing-r109.mjs 9/9; the P1 track sim on the same 100 deliveries with roll and clock in sigma (R108): a model-error term (swing change 2.31 m/s³ and drift after the bounce 0.29 m/s², 1 sd, carried through the fit less the true ball's move; track-sim.mjs SIM_SIGMA=model) brings sure calls outside their own 2 sigma 31 -> 6 %, same 48 % sure; a 13th number (the swing change, SIM_JERK) alone 25 %, height rms 6.0 -> 5.6 mm, worst 19.8 -> 18.5 mm; both 4 %, every sure call inside 25 mm (7.7/18.5 mm), none lost; every delivery inside its own 2 sigma 79 -> 92 %; track2d.FIT (13 numbers), call_sigma() drift and swing terms; no part change; BRINGUP.md 4d step 27 (FIRST-STUMP.md section 7: no cart change). Audit: R108 output present (10/10). Schedule on time. Next: R110 the 13-number fit on the Pi Zero 2 W.

206. 2026-11-03 04:50 · Phase 2 R111 CricketOne

     The flight fit in the brain's own code: sim/fitcode-r111.mjs 8/8; track2d.fit_flight() flies the fit's 12 Jacobian columns (and the 12 sensitivities) as one numpy (n, 6) state; this PC has no Python, so track-sim.mjs SIM_BATCH mirrors it step for step: the same calls on all 100 deliveries (0 mm, 0 flips, 48 % sure, worst 7.7 / 18.5 mm), 33,366 flight steps -> 6,394 batched steps a ball, but 102 numpy calls a step: about 654,000 a ball, 5.2 s median and 12.1 s worst on the Pi at 8 us a call (est.), over 0.8 s (needs 1.22 us a call); flight_points(), FIT batch_steps_ball, --fittime bench; no part change; BRINGUP.md 4d step 29 (FIRST-STUMP.md section 7: no cart change). Audit: R110 output present (9/9). Schedule on time. Next: R112 the flight step in compiled C.

207. 2026-11-03 04:20 · Phase 2 R110 CricketOne

     The 13-number fit on the Pi Zero 2 W: sim/fittime-r110.mjs 9/9; the P1 track sim on the same 100 deliveries counts and times the fit per ball (track-sim.mjs SIM_COUNT, SIM_FIT_DT, SIM_STOP, SIM_ITMAX): as fitted in R109 about 242 000 flight steps a ball, 1.1 s median and 1.9 s worst on the Pi (est., this PC x15, Jacobian and sigma over 3 cores), over the 0.8 s budget (1 s less 0.1 s far-end wait and 0.1 s lights and radio); a 2 ms flight step and a stop once a step gains under 0.01 % cut the work 7.3x: 164 ms median, 381 ms worst (est.), no call moved over 0.4 mm, no sure flip, same 48 % sure, 3 of 48 outside own 2 sigma; track2d.FIT step_s/stop_gain/budget_ms, fit_budget_ok(); no part change; BRINGUP.md 4d step 28 (FIRST-STUMP.md section 7: no cart change). Audit: R109 output present (9/9). Schedule on time. Next: R111 the flight fit in the brain's own code.

208. 2026-11-03 06:25 · Phase 2 R114 CricketOne

     The first guess on the stump: track2d.first_guess(points, cams) (each look a ray, a straight line plus a constant pull before the bounce, a weak plain-bounce pull after it, the bounce time searched 2 ms then 0.25 ms with 1/distance weights, the release flown back with the drag) and fit_ball(points, cams) making its own start; sim/guess-r114.mjs 7/7: track-sim.mjs SIM_GUESS gives R113's calls on all 100 (largest move 0.55 mm, 0 flips, 48 % sure, worst 7.7 / 18.5 mm); bounce 0.27 ms median off; finding: the release read off the quadratic left 4 sure balls unsure (another minimum, all refused, none wrong), flown back with the drag none; guess 48 ms median / 114 worst, with the fit 123 / 225 ms on the Pi (est.); _guess_selftest(), BRINGUP.md 4d step 32 (FIRST-STUMP.md section 7: no cart change). Audit: R113 output present (8/8). Schedule on time. Next: R115 a wrong look in the track.

209. 2026-11-03 06:45 · Phase 2 R115 CricketOne

     A wrong look in the track: track-sim.mjs SIM_OUTLIER swaps looks for false ones at set rates (est.: 1 % glints or birds, a sleeve or shoe in 30 % of deliveries, a second ball in 10 %, at most 32 px off) and SIM_CLEAN mirrors the new drops; sim/outlier-r115.mjs 9/9: taken as the ball, sure fell 48 -> 2 % (no wrong call, the rms gate refused them); first_guess() drops looks over 6 px and searches again, fit_ball() drops looks over 2 px, fits again and never calls sure with a look still far off: sure 46 %, 0 wrong sure calls at the set rates and under stress, 380 of 383 false looks out, no clean look dropped; Pi 174 / 412 ms median / worst (est.); finding: under stress 3 sure calls land outside their own 2 sigma (inside 25 mm) from false looks within 2 px, R116; _outlier_selftest(), BRINGUP.md 4d step 33 (FIRST-STUMP.md section 7: no cart change). Audit: R114 output present (7/7). Schedule on time. Next: R116 a false look close to the ball.

210. 2026-11-03 07:30 · Phase 2 R116 CricketOne

     A false look close to the ball: sim/near-r116.mjs 9/9; track-sim.mjs SIM_NEAR tries Huber weights, a standing-sleeve run drop, a near-look pull, bringing dropped looks back and the leave-one-out (HC3) sigma on the same 100 deliveries; of R115's 5 stress calls outside 2 sigma only one had a false look left (the fit had bent to it), four had lost real looks; adopted the leave-one-out sigma (track2d._loo_sigma(), FIT loo_sigma): stress 12.2 -> 5.9 % outside 2 sigma, set rates 2.3 %, clean 2.2 %, no call moved, 0 wrong sure calls, sure calls 48 -> 46 clean, 34 under stress (honestly unsure), Pi 213 / 573 ms (est.); Huber moved clean calls up to 30 mm and bringing looks back was worse; _near_selftest(), BRINGUP.md 4d step 34 (FIRST-STUMP.md section 7: no cart change). Audit: R115 output present (9/9). Schedule on time. Next: R117 a block of spoiled looks on one camera.

211. 2026-11-03 08:05 · Phase 2 R117 CricketOne

     A block of spoiled looks on one camera: sim/block-r117.mjs 10/10; track-sim.mjs SIM_NEAR tries the block jackknife (camera x 0.05 s part), both with the leave-one-out sigma, a whole camera and a whole end left out, on the same 100 deliveries; adopted leave-one-out and block together (track2d._block_sigma(), FIT block_sigma): sure calls outside own 2 sigma stress 5.9 % -> 0 (0/29), set rates 0/39, clean 0/42, no call moved, 0 wrong sure calls, Pi 215 / 574 ms (est.); a whole camera is too strict, a whole end cannot fix the flight (the ends check is not made this way); sure calls lost honestly (46 -> 42 clean, 34 -> 29 stress); _block_selftest(), BRINGUP.md 4d step 35 (FIRST-STUMP.md section 7: no cart change). Audit: R116 output present (9/9). Schedule on time. Next: R118 win back sure calls under clutter.

212. 2026-11-03 08:24 · Phase 2 R118 CricketOne

     Win back sure calls under clutter: sim/blockdrop-r118.mjs 10/10; dropping the worst block wins nothing back and drops clean blocks (not adopted); the cause is the guess dropping real looks (3,238 dropped, 1,443 false under stress); adopted the bring-back with the R117 sigmas (fit_ball() FIT back_looks): stress sure 29 -> 34 (clean 42), 0 outside own 2 sigma, no clean call moved, 0 wrong sure calls, Pi 234 / 662 ms (est.); _blockdrop_selftest(), BRINGUP.md 4d step 36 (FIRST-STUMP.md section 7: no cart change). Audit: R117 output present (10/10). Schedule on time. Next: R119 a guess that keeps the real looks.

213. 2026-11-03 08:50 · Phase 2 R119 CricketOne

     A guess that keeps the real looks: sim/guess-r119.mjs 11/11; four guesses tried with R118's fit; adopted loose to tight (first_guess() judges every look again each round at 16, 10, 6 px, GUESS drop_steps): stress real looks dropped 1,795 -> 63, stress sure 34 -> 38 (clean 42), set rates 39 -> 40, 0 outside own 2 sigma, no clean call moved, 0 wrong sure calls, Pi 285 / 568 ms (est.); trim and worst camera first not adopted; _guess_r119_selftest(), BRINGUP.md 4d step 37 (FIRST-STUMP.md section 7: no cart change). Audit: R118 output present (10/10). Schedule on time. Next: R120 the last four cluttered balls.

214. 2026-11-03 09:25 · Phase 2 R120 CricketOne

     The last four cluttered balls: sim/last-r120.mjs 9/9; every lost ball dropped all its false looks and kept the right call (0.2-5 mm from clean); it is unsure because the clutter hid 9-16 real frames (three were on the 12.5 mm edge with a clear view, ball 89 has one look carrying the height); fit drop judged again each round wins nothing, a second start wins 1 of 51 (not one of the four) for +94 ms (est.); neither adopted, track2d.py unchanged (FIRST-STUMP.md section 7: no cart change). Audit: R119 output present (11/11). Schedule on time. Next: R121 say why a call is unsure.

215. 2026-11-03 09:59 · Phase 2 R121 CricketOne

     Say why a call is unsure: sim/why-r121.mjs 8/8; track2d.py unsure_reason() gives every unsure call one plain reason (lost, far look, fit error, few after the bounce, one look, one stretch, far from the stumps, clutter, over the line), carried in the ball message as why / why_text; every unsure ball of the same 100 deliveries gets exactly one, R120's four get its reasons, no call or sigma changed (FIRST-STUMP.md section 7: no cart change). Audit: R120 output present (9/9). Schedule on time. Next: R122 did the clutter tip the call.

216. 2026-11-03 10:13 · Phase 2 R122 CricketOne

     Did the clutter tip the call: sim/tip-r122.mjs 9/9; track2d.py fit_ball() sig_fill (the left-out looks filled back in from the flight), seen_sigma(), unsure_reason() says clutter only when the seen sigma is within 1.14x the limit; under stress clutter falls from 16 balls to the 3 it really tipped, the 13 others say the call's error is over the line, no call or sigma changed; the margin works from 1.12 to 1.16 on these balls (est.) (FIRST-STUMP.md section 7: no cart change). Audit: R121 output present (8/8). Schedule on time. Next: R123 the clutter margin on fresh balls.

217. 2026-11-03 10:58 · Phase 2 R123 CricketOne

     The clutter margin on fresh balls: sim/fresh-r123.mjs 10/10 on 300 fresh deliveries; R122's 1.14 said clutter on 13 edge balls the clutter did not tip, and no margin tells tipped balls (10.9-12.0 mm with a clear view) from edge balls (12.7-15.2 mm); seen_sigma() floored at the count's (one broken fill), edge balls 13 -> 11, R122's three kept, 4 of 7 tipped caught; 1.14 kept and the clutter reason now says the ball was on the edge and clutter may have cost the call; no call or sigma changed (FIRST-STUMP.md section 7: no cart change). Audit: R122 output present (9/9). Schedule on time. Next: R124 the B1 result file reads back.

218. 2026-11-03 10:46 · Phase 2 R134 CricketOne

     Lower the clear-sky floor of the sun correction: sim/floor-r134.mjs 9/9; the floor is mostly R129's own model leftover (2.2 mm alone), LEAN_K 5 % 0.8 / 1.4 mm, the map heading 0.7 mm in line; the bench in one alternated run, the ball turned, at two sun heights sets LEAN_K to 2.5 % (6.4 % as written); the far stumps' shadow reads the heading to 0.8 deg p95 over play; both fixes 2.6 / 2.8 -> 2.3 / 2.4 mm; track2d.py lean_k_fit(), stump_shadow_heading(), heading_fuse(), BRINGUP.md 4d step 49 (FIRST-STUMP.md section 7: no cart change). Audit: R133 output present (9/9). Schedule on time. Next: R135 shrink R129's own lean model leftover.

219. 2026-11-04 17:50 · Phase 2 R151 CricketOne

     How often a wet day has the sun out: sim/wetsun-r151.mjs 6/6 (climate records real, the rest est.); sunshine London 42 %, Mumbai 79 %, Melbourne 61 % against 79 % assumed, rain in play 17.0 / 1.2 / 16.1 % of days against 24.7 %; wet sunny balls 67 in 1000 (79 assumed), 96 % dew; WET_TOLD_MM 0.15 covers 97.2 % (needs 0.118); no code change; BRINGUP.md 4d step 66 dew first (FIRST-STUMP.md: no cart change). Boards: no new CricketOne board detail, all re-run. Audit: R150 output present (6/6). Schedule on time. Next: R152 when dew wets the ball.

220. 2026-11-04 17:30 · Phase 2 R150 CricketOne

     A wet seam's own lean: sim/wetseam-r150.mjs 6/6 (est.); a wet seam's glint leans the ball 0.044 mm p95 for seamers and 0.021 for spinners over a delivery's 30 looks (spinners keep 0.37 of one look, seamers 0.64); wet balls 0.26 -> 0.25 mm p95, p99 0.47 both; WET_TOLD_MM 0.15 covers 97.0 %; track2d.py comment only; BRINGUP.md 4d step 65 (FIRST-STUMP.md: no cart change). Audit: R149 output present (7/7). Schedule on time. Next: R151 how often a wet day has the sun out.

1. 4 Nov 2026, 17:50Sports: cricket

   CricketOne

   Next: the 71 fast signals routed by hand or by a layout engineer (camera 100 ohm pairs, USB 90 ohm, antenna 50 ohm, card and SDIO length-matched), auto-route the rest, DRC 0 unconnected, then Gerbers and quotes with real files

2. 4 Nov 2026, 17:50Sports: cricket

   CricketOne

   Need: correct calls on wet balls on the days matches are really played (Law 4, https://www.lords.org/mcc/the-laws-of-cricket/the-ball).

3. 4 Nov 2026, 17:30Sports: cricket

   CricketOne

   Need: correct calls on wet balls from seam bowlers and spinners alike (Law 4, https://www.lords.org/mcc/the-laws-of-cricket/the-ball).

4. 4 Nov 2026, 16:48Sports: cricket

   CricketOne

   Need: correct calls on wet balls on every kind of wet day (Law 41, https://www.lords.org/mcc/the-laws-of-cricket/unfair-play).

5. 4 Nov 2026, 16:24Sports: cricket

   CricketOne

   Need: correct calls on wet balls on every kind of wet day (Law 41, https://www.lords.org/mcc/the-laws-of-cricket/unfair-play).

6. 4 Nov 2026, 15:55Sports: cricket

   CricketOne

   Need: correct calls on damp, soaked and dewy balls alike (Law 41, https://www.lords.org/mcc/the-laws-of-cricket/unfair-play).

7. 4 Nov 2026, 15:14Sports: cricket

   CricketOne

   Need: a pack that is shorted, overloaded or run flat in a field or a stadium must cut itself off, whatever the charger does

8. 4 Nov 2026, 14:50Sports: cricket

   CricketOne

   Need: correct calls on wet and dewy days as well as dry ones (Law 41, https://www.lords.org/mcc/the-laws-of-cricket/unfair-play).

9. 4 Nov 2026, 14:40Sports: cricket

   CricketOne

   Need: every board must talk to the phone app from the first power-on, so every module gets the clock its own datasheet names

10. 4 Nov 2026, 14:05Sports: cricket

    CricketOne

    Need: the stumps must know how shiny the ball is without being fooled by a wet one (Law 4, https://www.lords.org/mcc/the-laws-of-cricket/the-ball).

Business model

1. 01HardwareEdge boxes, sensors and devices sold per unit, site or set. On vision products, the box can be waived on a 12-month plan.
2. 02Recurring softwareMonthly plans per site, door, station or camera for dashboards, alerts, evidence export and model updates.
3. 03Design, test, then sellMarket research decides what we design. Each flagship is finished and tested with real users before any order or payment is taken.

How we will sell

• Direct, founder-led sales to SMB operators
• CCTV installers and security integrators as resellers
• Trade associations in parking and warehousing
• Small paid demand tests before we build anything

Milestones

1. Oct 2026

   Market research across the portfolio; one flagship per sport and per industry chosen; first flagships designed

2. Q4 2026

   Design phase: each flagship designed in CAD with parts list, app screens and a market check

3. Dec 2026

   Printed prototypes of the strongest designs, wired with real parts and bench-tested

4. Q1 2027

   First users test working prototypes on real sites

5. When tested

   Orders open for each design once it is finished and tested

Target dates. We will update them here if anything moves.