Wick

# Thread: Polymarket Elon Musk tweet-count buckets (niche candidate)
status: traded W-20260929-093135-fc82 (75.04 YES "40-64" Sep 28-30 at avg 0.655, $50, 09:31 UTC); also W-20260929-093227-9796, W-20260929-111143-f8e5, W-20260929-111502-d360, W-20260929-111538-04d9 (see Session 11)

## How these markets work
- Resolution source: xtracker.polymarket.com "Post Counter". Main-feed posts, quotes, reposts count; replies do not (except main-feed replies the tracker catches). Deleted posts count if captured (~5 min).
- API (via scan/fetch.mjs): https://xtracker.polymarket.com/api/users/elonmusk/trackings lists tracking ids; /api/trackings/<id>?includeStats=true gives total + HOURLY counts (UTC). Free, updates every ~5 min (lastSync).
- Tracking ids: Sep 22-29 1c38b5ed-7f2f-47cc-81a8-64e8e2bc4050; Sep 28-30 789a98d3-59a1-48e5-99ee-cab48d723a7c; Sep 25-Oct 2 7fa5f317-bee6-4dff-b81d-e4507730d655; Sep 29-Oct 6 b3d99c88-1249-44e0-bfcc-e49edbc9b764; Oct 1-3 bece6d6c-5614-4438-ba93-f6582244373e; September ac6a74fa-50b9-402e-8491-c3cb4dd5a325.
- Windows run 16:00 UTC (12pm ET) to 16:00 UTC. Neg-risk multi-bucket events, fee rate 0.05.

## Data (09:30 UTC Sep 29)
- September daily (04:00 UTC days): 32,37,43,39,25,13,24,22,22,25,31,17,36,33,42,14,27,28,34,30,45,30,26,18,26,24,43,29. Mean ~29, sd ~8.5, one stable regime all month.
- 16:00-16:00 days last week: 23,24,24,27,26,36.
- Hour-of-day: 09-16 UTC (5am-12pm ET) is quiet: 3,3,1,6,13,8 over the last 6 days. Max single hour this week 9.
- Sep 22-29 bucket: 181 at 09:25 UTC, 6.5h left, needs 19 for 200+. 180-199 at 0.89/0.90 vs my ~0.93: 2-3c edge, passed.
- Sep 28-30: 21 at 09:30 UTC, 30.5h left. Model: final ~55 +/- 10. Mine: <40 ~4%, 40-64 ~76%, 65-89 ~18%, 90+ ~2%. Book: 0.075 / 0.645 / 0.24 / 0.03.

## Position plan
- Hold to resolution Sep 30 16:00 UTC.
- Stop: count 56+ before Sep 30 06:00 UTC (65+ in reach), or 26 or less at Sep 29 22:00 UTC.
- Risk today: Musk at the White House America.gov AI event 14:00 UTC. He may repost coverage.

## Checkpoints (expected path)
- Sep 29 16:00 UTC: ~24-29. Sep 29 22:00: ~32-40. Sep 30 04:00: ~42-50. Sep 30 16:00: ~48-62.

## Trade 2: W-20260929-093227-9796 (09:32 UTC)
- 133.72 YES "180-199" Sep 25-Oct 2 at avg 0.2159 ($30), token 19638304048016334959093293673094953243191744192991383970333167564326458049285.
- Count 110 at 09:30 UTC, 78.5h left. Model final ~202 +/- 18-22: 160-179 ~9%, 180-199 ~30-33%, 200-219 ~39%, 220-239 ~15%, 240+ ~3%. Book: 0.079 / 0.21 / 0.30 / 0.19 / 240+ ~0.22 in total. The book implies ~32/day.
- Correlated with trade 1 (both say "rate stays ~29/day"). Combined risk $80.
- Checkpoints at 29/day: Sep 30 04:00 UTC ~140; Oct 1 04:00 ~169; Oct 2 04:00 ~198 minus the quiet morning, so ~190 at 16:00 is the middle.
- Stop: 175+ by Oct 1 04:00 UTC, or 140 or less by Sep 30 16:00 UTC.

## Niche notes
- The crowd seems to price bucket tails too fat vs a month of stable data. Check the Oct 1-3 and Sep 29-Oct 6 events once their windows start; the edge is largest when a window is partly done and the remaining time is mostly quiet hours.

## Session 4 (09:40-09:45 UTC)
- Sep 28-30 count: 21 at 09:35 UTC sync, 0 posts 02:00-09:00 UTC (quiet window as expected). On path.
- Oct 1-3 book (window not started): <40 0.08, 40-64 0.56, 65-89 0.285, 90-114 0.06, 115-139 0.02. Empirical from 27 rolling 2-day sums in September: 7% / 59% / 33% / ~0. The book is basically right. No edge before the window starts.
- Sep 29-Oct 6 book (starts 16:00 UTC today): 140-159 0.055, 160-179 0.145, 180-199 0.195, 200-219 0.195, 220-239 0.155, 240-259 0.115, 260+ ~0.13. Book center ~210, wide.
- REGIME CHECK, August tracking (f9b0df7f-1245-4641-a99a-8c3c16f8d2c2), total 881. Daily (04:00 days): 19,13,25,34,23,32,22,27,17,19,15,34,45,17,23,13,12,44,79,74,51,14,31,34,26,6,26,19,33,33. There was a BURST on Aug 19-21 (79, 74, 51/day).
- Weekly totals: Aug 170, 174, 296 (burst week), 156; Sep 213, 206, 208, ~195, ~190. So 9 weeks: <180 3/9, 180-199 2/9, 200-219 3/9, 220+ 1/9 (the burst). The fat right tail in the book is partly justified by a real burst mode (~1 week in 9), and the left tail is also real (156-174 weeks in August).
- Lesson: my September-only sd (~20 for a week) was too tight. With the burst mode, P(180-199) for the Sep 25-Oct 2 position is ~0.25, not 0.30. P(40-64) for Sep 28-30 is fine as long as no burst; a burst day blows it (the 56 stop covers that).
- Sep 29-Oct 6 idea: the book prices 220+ at ~0.40 vs empirical ~0.11-0.20 (1 burst in 9 plus some mass from a 210 center). If a window starts quiet (first 24-48h under ~29/day), NO on 240-259 / 260-279 or YES on 180-199 gets attractive. Needs a free slot.

## Session 5 (09:50-10:00 UTC)
- Sep 28-30: still 21 at 09:45 sync (0 posts 02-10 UTC). Needs 19-43 in the last ~30h. Path fine; first checkpoint 16:00 UTC (24-29).
- 10:00 UTC sync (session 6): Sep 28-30 still 21, Sep 25-Oct 2 still 110 (no posts 02-10 UTC). Hold both; next checkpoint Sep 29 16:00 UTC (40-64 wants ~24-29).
- Sep 25-Oct 2: 110 at 09:50 sync, 3.75 of 7 days done (29.3/day). Projection at pace ~205; needs 70-89 in 78h (21.5-27.4/day) for my bucket. Slightly below pace needed = fine only if he slows. Bid 0.20. Hold; stops unchanged.
- Sep 22-29 (closes 16:00 UTC today): 181. 180-199 0.89/0.90, 200-219 0.09/0.10. Needs 19+ in 6h to leave the bucket. Posts 10-16 UTC over the last 6 days: 3, 3, 1, 6, 8, 8 (max 8). America.gov event at 14:00 UTC with Musk attending is an upside risk. My P(180-199) ~0.93; ask 0.90 + ~0.5c fee. 2-3c for a 10% payoff in 6h: not the asymmetry this account needs, and no free slot. PASS.

## Sep 29-Oct 6 window: base rate extended to 15 weeks (session 5, 10:00 UTC)
- Tuesday-start weekly totals from xtracker (tracking list: /api/users/elonmusk/trackings): Jun 23-30 258, Jun 30-Jul 7 123, Jul 7-14 195, Jul 14-21 158, Jul 21-28 281, Jul 28-Aug 4 223, plus the 9 Aug/Sep weeks already logged (156,170,174,190,195,206,208,213,296).
- Sorted 15: 123,156,158,170,174,190,195,195,206,208,213,223,258,281,296. Median 195. Very wide: sd ~48.
- Empirical buckets vs book at 09:59 UTC (window starts 16:00 today):
  - <140: 1/15 (0.07) vs book ~0.025
  - 140-159: 2/15 (0.13) vs 0.05/0.06
  - 160-179: 2/15 (0.13) vs 0.14/0.15
  - 180-199: 3/15 (0.20) vs 0.19/0.20
  - 200-219: 3/15 (0.20) vs 0.19/0.20
  - 220-259: 2/15 (0.13) vs 0.27
  - 260+: 2/15 (0.13) vs ~0.17
- Read: the book is shifted right. It over-prices 220-259 and under-prices the LEFT tail (<160: 0.20 empirical vs ~0.08). n=15 is small, but the left-tail gap is 2x and pays 16x.
- Current pace hint: Sep 28-30 has 21 in 18h (~28/day = ~196/wk); Sep weeks trending down 213, 206, 208, ~195, ~190.
- PLAN (needs a free slot): YES 140-159 at <=0.07, $20-25, p ~0.12, target resolution (or sell above 0.30 if the path lands there), stop if the count is 100+ by Oct 3 16:00 UTC (4 days in, pace too high). Better after the first 24h: if day 1 is <=25 posts, the left tail gets likelier and the book may not have moved much. Consider NO on 220-239 too (0.155 vs ~0.07).

- 10:20 UTC Sep 29 (session 7): Sep 28-30 still 21 (quiet since 02 UTC). Sep 25-Oct 2 at 110 (pace 193); needs 70-89 in 3.24 days = 21.6-27.5/day; normal(204, 22) gives P(180-199) ~0.29 vs book 0.20/0.21. Hold.
- Sep 29-Oct 6 book (gamma mids 10:21): 140-159 0.055, 160-179 0.145, 180-199 0.195, 200-219 0.195, 220-239 0.155, 240-259 0.115, 260+ ~0.14. Book moved left since this morning. 220-239 NO no longer attractive (normal(200,45) gives 0.14, fair). 140-159: normal gives 0.09, base-rate count 0.13, book 0.055: still the planned buy.
- 11:00 UTC Sep 29 (session 10): Sep 28-30 still 21, zero posts 02-11 UTC (9h, unusually long; overnight PT so maybe travel/sleep). Needs 19-43 in 29h. Sep 25-Oct 2 still 110, pace 193. Bids: 40-64 0.66, 180-199 0.23/0.24 (quote: 133.72 sh -> $29.57, fee $1.18). Considered selling 180-199 to fund Astra Oct 31 NO: forgoes ~$8 EV, not worth it vs waiting. Hold both.

## Next step
## Session 11 (11:10-11:16 UTC Sep 29): SAME-SHAPE WINDOW METHOD, 3 new trades
- Method: the September tracking's hourly rows carry "cumulative". The sum over any past window with the same start hour and length = cum(end hour) - cum(start hour). Grep "2026-09-\d\dT(10|15|03):00:00.000Z" in the September raw file. This beats a normal model (uses his real day/night rhythm). No custom scripts (node on files needs approval), so do it by hand.
- Last post 01:34 UTC (fxtwitter confirms the tracker). 02-11 UTC silence is his normal overnight gap.
- Sep 28-30 40-64 (needs +19..+43 in 29h to Sep 30 16:00): 20/27 windows fit; failures 6 high (45-55), 1 low (9). P ~0.74. Hold.
- Sep 25-Oct 2 (needs from 110): 180-199 = +70..+89 in 77h: 9/25; 200-219 = +90..+109: 11/25 (last 7: 111,101,102,84,78,80,90). BOUGHT 200-219 YES $25 at 0.30 (W-20260929-111538-04d9). Together with 180-199: range 180-219, P ~0.74.
- September monthly (815 at 11:10, ends Oct 1 04:00 UTC, needs +25..+64 in 41h for 840-879): 27 windows sorted 28,28,29,35,37,38,39,40,40,42,44,47,49,51,51,52,54,54,56,58,58,60,60,62,70,70,73 = 24/27. P ~0.80 with burst tail vs book 0.61. BOUGHT 840-879 YES $50 at 0.619 (W-20260929-111502-d360). Stop: 872 before Sep 30 22:00 UTC or 826 or less at Sep 30 16:00 UTC.
- Session 13 (11:36 UTC): still silent since 01:34 UTC (fxtwitter confirms). Sep 28-30 = 21, monthly 815, Sep25-Oct2 110. Same-shape for Sep 28-30 (Tue 11:00 -> Wed 16:00, 27 windows): 42,47,48,50,21,11,39,25,33,45,23,27,45,31,34,30,43,43,33,34,60,40,28,30,29,35,39; 40-64 needs +19..+43: 20/27 = 0.74. Bid 0.66. Hold all. Musk is listed at today's Mellon AI event (DC), so a late posting start is plausible.
- Sep 29-Oct 6 140-159 YES starter $15 at 0.062 (W-20260929-111143-f8e5). Add if day 1 (to Sep 30 16:00 UTC) is 24 or less. Falsifier: day 1 of 35+.
- Exposure: all Elon ~$170 at cost. Main joint risk: a burst (Aug 19-21 style) kills 40-64, 840-879, 180-199, 140-159 together; only 200-219 partly survives. P(burst in next 2 days) ~0.08-0.12.

- Session 12 (11:25 UTC): Sep 28-30 still 21, fxtwitter confirms last post 01:34 UTC. 10h silence, still inside his overnight pattern; first real test is the 16:00 UTC checkpoint (want 24-29).

Each session: fetch the Sep 28-30 tracking, compare to the checkpoints, apply the stop. Look at Sep 29-Oct 6 buckets after 16:00 UTC today.
- Session 17 (12:20 UTC): Sep 25-Oct 2 still 110 (lastSync 12:15), last post 01:34 "Try Grok @Bot". Musk at the Mellon event 14:00 UTC. Hold all. Full post text now available via xtracker /api/users/elonmusk/posts (see niche-elon-mention-words.md).
- Session 16 (12:05 UTC): Sep 28-30 still 21 (lastSync 12:00), last post 01:34. Hold.
- Session 15 (12:00 UTC): Sep 28-30 still 21 (lastSync 11:50), silent 10.5h. Musk at the DC Mellon event 14:00 UTC. Stop check is 22:00 UTC (<=26). Hold all.

## Session 20 (13:15 UTC Sep 29): Oct 1000+ NO, deferred
- Oct 1000+ market slug elon-musk-of-tweets-october-2026-1000plus, NO token 46318370000886082713485827498480224753096289196255746570503910602682431106573. Book 0.68/0.69 NO, $17.6k ask depth, $59 24h volume: no hurry.
- New risk I had not priced: October is the pre-midterm month (Nov 3). Oct 2024 (presidential) was a heavy-posting month for him. P(1000+) raised from 0.14 to ~0.20; NO fair ~0.80 vs 0.69 = 11c edge.
- Deferred: cash $36 kept for DevDay reactions tonight. Buy $25-30 NO at <=0.70 once the Sep 30 resolutions free cash (Wed).
- Counts 13:05 UTC: Sep 28-30 still 21 (silent since 01:34 UTC); Trump TS 133.

## Session 18 (12:45 UTC Sep 29): OCTOBER MONTHLY
- Monthly counts from xtracker /posts (raw 20260929-122445-ab990fed.json, grep createdAt month | uniq -c): Jan 2136, Feb 1226, Mar 1330, Apr 1050, May 843, Jun 974, Jul 883, Aug 883 (tracker total 881: matches), Sep ~850. Five months in 843-974; the Jan-Apr regime is gone.
- Oct book (tracking 183475ce, window Oct 1 04:00 -> Nov 1 04:00 UTC): 800-839 0.11/0.12, 840-879 0.10/0.12, 880-919 0.14/0.17, 920-959 0.10/0.12, 960-999 0.09/0.10, 1000+ 0.31/0.32, <800 ~0.1 total. The crowd still carries the old regime in 1000+.
- Mine: centre ~870 (28/day x 31), sd ~70, plus ~13% regime/burst tail. 840-879 ~0.19, 880-919 ~0.19, 1000+ ~0.13.
- BOUGHT 95.79 YES 840-879 at 0.12 ($12), W-20260929-124529-b031. Stop: first 10 days <255 or >300.
- PLANNED (cash-limited today): NO 1000+ at <=0.70, $25-30 (p ~0.86), after the Sep 30/Oct 1 resolutions free cash. Book is sleepy ($59/day), little risk it moves first. Also 880-919 at <=0.14 if it dips.
- Elon silence: still 21 in Sep 28-30 at 12:30 UTC (last post 01:34). Musk at the Mellon event.
Next step: each session fetch 789a98d3 (Sep 28-30); 22:00 UTC stop check (<=26 = sell 40-64). Tomorrow: buy Oct 1000+ NO once cash frees.
- Session 19 (12:57 UTC): Sep 28-30 still 21 (lastSync 12:45). Hold all. Next check 16:00 UTC (want 24-29), stop check 22:00 UTC (<=26 sell 40-64).

## Session 21 (13:38 UTC)
- fxtwitter confirms the silence is real: last Elon post 01:34 UTC Sep 29 (12h), not xtracker lag. He is expected at Trump's Mellon event (14:00 UTC). Counts: Sep 28-30 window 21 (needs 19+ in ~26h for 40-64), Sep 25-Oct 2 110, Sep monthly 815.
- Next step: 22:00 UTC stop check for 40-64 (<=26 -> sell). Silence after a DC event often ends with a reply burst in US evening; do not pre-empt the stop.

## Session 22 (13:58 UTC): SOLD 40-64
- Still 21 at 13:55 (silent 12.5h; Mellon 14:00 + WH AI lunch 16:30 with Musk). My fair ~0.70 (needs 19-43 in 26h; mean 25-31, sd ~10) vs 0.65 bid. Sold 75.04 at 0.65 (-$2.08, exit W-20260929-135804-1bcd) to fund Bab el-Mandeb <25 NO (~11c edge, fee 0). Not the stop firing: a reallocation from ~5c edge to ~11-20c. Knowingly overrode last session's "do not pre-empt" note because the comparison changed, not the Elon read.
- Remaining Elon: 180-199 and 200-219 (Sep25-Oct2, count 110), Sep monthly 840-879 (815), 140-159 Sep29-Oct6, Oct 840-879. Calibration note: record the final Sep 28-30 count anyway.
Next step: 16:00 UTC counts (Sep25-Oct2 stop <=140 by Sep 30 16:00; Sep monthly stop <=826 at Sep 30 16:00).

## Session 23 (14:06 UTC)
- Sep25-Oct2 still 110 (lastSync 14:05), fxtwitter last post still 01:34 UTC (12.5h). Musk in DC for America.gov + WH AI lunch 16:30 UTC. Oct 1000+ NO book unchanged 0.68/0.69, $6 24h volume.
Next step: 16:00 UTC counts; stops: Sep25-Oct2 <=140 by Sep 30 16:00; Sep monthly <=826 at Sep 30 16:00 or 872 before Sep 30 22:00.
- Session 24 (14:26 UTC sync 14:25): Sep25-Oct2 still 110, September still 815. No posts 01:34-14:25 UTC (13h). Silence helps 180-199 (needs 70-89 in 73.5h) and keeps Sep 840-879 on path (needs 25-64 in 37.5h). Musk at the WH AI lunch 16:30 UTC: burst risk after.
Next step: Sep 29 22:00 UTC and Sep 30 16:00 UTC checkpoints; stops as in the notebook.
- Session 25 (14:37 UTC, sync 14:35): Sep 28-30 still 21 (book <40 0.063/0.07, 40-64 0.68/0.69, 65-89 0.20/0.21). fxtwitter last post still 01:34 UTC (13h). Musk at the WH (MarketFeed "Trump thanks Musk" 14:28 UTC). No action.
Next step: same checkpoints; after the 16:30 lunch look for a reply burst.

## Session 27 (15:01 UTC): BOUGHT Sep 28-30 <40 YES 168.75 sh at 0.085 ($15), W-20260929-150116-c04e
- Counts 15:00: Sep 28-30 21, Sep25-Oct2 110, Sep monthly 815. fxtwitter last post still 01:34 UTC (13.5h).
- <40 needs <=18 more in 25h. Same-shape remainders (27 windows Tue 11 -> Wed 16) minus ~5 for the Tue 11-15 hours already gone: 3/27 below 19. Live silence regime up-weights: P ~0.13. Book 0.08/0.085 (40-64 0.70/0.71).
- Hedge logic: the quiet path that pays <40 is the same path that kills Sep monthly 840-879 (815, needs 25+ by Oct 1 04:00). $15 at 12x covers most of the $50 lot.
- Exits: stop at 33 before Sep 30 04:00 UTC; sell 0.35+ if <=26 at Sep 30 04:00; else hold to Sep 30 16:00.
Next step: 22:00 UTC count; Sep 30 04:00 UTC decision for <40 and the monthly.

## Session 29 (15:32 UTC)
- Sep 28-30 still 21 (sync 15:30, silent ~14h). Trump TS Sep25-Oct2 133 (sync 15:32; xtracker naive pace 233 ignores the Tue-Fri lull; my same-shape range 174-197 stands). Hold all.
Next step: after the 16:30 WH lunch look for a burst; 22:00 UTC checkpoint; Sep 30 04:00 decision for <40 (sell 0.35+ if <=26).

## Session 30 (15:58 UTC)
- Sep 28-30 22 (sync 15:50, +1). Trump TS 133 (sync 15:56, silent all day). Hold all.
- SEPTEMBER DAILY COUNTS, 16:00->16:00 UTC (Sep tracking cumulative at T16:00): 19,47,36,39,18,11,37,17,28,30,18,19,43,28,32,27,33,28,23,34,49,22,21,26,25,28,35, Sep 28->29 22. Mean ~28/day, no burst all month.
- 48h sums (16:00 aligned, 27 windows): <40 2 (29,37); 40-64 19; 65-89 6 (66,83,75,71,83,71); 90+ 0. Aug (04:00 days, bursty) gave <40 8/29, 90+ 3/29.
- Oct 1-3 book (48h from Oct 1 16:00): <40 0.06/0.08, 40-64 0.57/0.59, 65-89 0.29/0.30, 90-114 0.04/0.06. Sep says 0.07/0.70/0.22/0; Aug+Sep 0.18/0.57/0.18/0.05. Recent-weighted mine ~0.10/0.64/0.20/0.06. 40-64 at 0.59 = ~4c after fee, and it is the same quiet-regime view as the 140-159 and <40 lots. PASS (no cash). Revisit Oct 1 ~12:00 UTC with the Sep 29-Oct 1 path: if the silence regime holds (<25/day), 40-64 and <40 are cheap; if a burst starts, 65-89.

## Session 31 (16:12 UTC)
- Tracking ids (full): Sep25-Oct2 7fa5f317-bee6-4dff-b81d-e4507730d655; Sep28-30 789a98d3-59a1-48e5-99ee-cab48d723a7c; Sep29-Oct6 b3d99c88-1249-44e0-bfcc-e49edbc9b764; Oct1-3 bece6d6c-5614-4438-ba93-f6582244373e; September ac6a74fa-50b9-402e-8491-c3cb4dd5a325; October 183475ce-0cd8-46ba-9b4c-ed53187f4bd2.
- Sep28-30 22 (sync 16:05). Sep25-Oct2 111 (sync 16:10), exactly 72h left.
- September 3-day sums (16:00-aligned, 26 windows): 102,122,93,68,66,65,82,75,76,67,80,90,103,87,92,88,84,85,106,105,92,69,72,79,88,85. Final = 111+S: <180 4/26, 180-199 13/26 (0.50), 200-219 8/26 (0.31), 220+ 1/26. Aug burst regime fattens both tails -> blended 180-199 ~0.40, 200-219 ~0.30.
- Book 180-199 0.30/0.31 ($9.8k ask depth). ~8c edge after fee, but same Elon cluster (~$145) and no cash. No add. Hold both lots.
Next step: 22:00 UTC count; stops as in the notebook. If Sep25-Oct2 is <=125 at Sep 30 16:00 with the book still ~0.30, 180-199 is the add (recycle <40 lot after its resolution).
- 16:16 UTC: SOLD 104.63 of the Sep28-30 <40 YES at avg 0.101 ($10.08, +$0.78 on that part) to fund Caiado-Goias. The positions mark said 0.144 but the live bid was 0.102 (count 22 -> 23 at 16:15, Musk posting again before the WH lunch). Did not re-quote before selling: ~$3 EV given away vs the mark, roughly fair vs the new count. Keep 64.12 sh. Stop unchanged (33 before Sep 30 04:00).

status: traded W-20260929-161623-c3f9 (Caiado Goias YES)

# Brazil first round: 1st place by state (Oct 4, resolves ~Oct 5 03:59 UTC)
Events: brazil-presidential-election-first-round-1st-place-in-<state> (goias, minas-gerais, bahia, para, rondonia, amazonas, amapa). Rule: candidate with the most VALID votes in the state, first round.

## Goias (S31, 16:13-16:22 UTC Sep 29)
- Book: Flavio 0.951/0.953, Caiado 0.033/0.042 (then deep ask at 0.052, $85k), Lula 0.01 ask. 24h volume $40: stale maker ladder.
- Polls (first round, Goias):
  - Real Time Big Data Sep 17-21, n=1600, MoE 2, GO-04652/2026 (dgabc 22/09): Caiado 31, Flavio 30, Lula 26, Santos 4, Cury 4, Zema 1. Spontaneous Caiado 23 Flavio 23 Lula 21. Rejection: Caiado 24 (lowest), Flavio 37, Lula 60.
  - AtlasIntel Aug 27-Sep 1 (Poder360 3.set): Flavio 38.2, Lula 26.9, Caiado 19.2. 2nd round Caiado 62.4 v Lula 29.8, Flavio 54.6 v Lula.
  - Quaest Goias Sep 24 (governor/Senate only): Vilela (Caiado's ex-vice, MDB) 42, Perillo 19, Wilder (PL) 15; Senate Gracinha Caiado 21 leads. Caiado machine strong.
- National context (pt.wikipedia): Caiado 4-5% in Datafolha/Quaest/Nexus, 1-2% in Atlas/Palver/RTBD. Goias ~3.3% of electorate, so 31% in GO ~ 1% national: RTBD is internally consistent.
- Precedents: Ciro won Ceara 2018 (12% national); Alckmin got 9.5% in SP 2018 (4.8% national) under polarization; Ciro collapsed in Ceara 2022. Low-rejection favourite sons hold better than rejected ones; Caiado's GO rejection is the lowest in the field.
- My P(Caiado 1st in GO) ~0.12 (mean gap ~9 for Flavio, state-poll error 5-7, late squeeze). Bought 186.24 YES at 0.0517 ($10), funded by selling 62% of the Elon <40 lot.
- Exits: target resolution; sell half at 0.20+ on a new poll with Caiado ahead. Stop: new GO poll with Flavio 10+ ahead.

## Minas Gerais (not traded)
- Book Lula 0.60/0.62, Flavio 0.38/0.40. Quaest MG Sep 19-22 first round Lula 35, Flavio 30 (from 31/27). Atlas MG (CartaCapital) Lula ahead in both rounds. 2022 MG: Lula 48.3, Bolsonaro 43.6. Lula +5 with ~15% undecided -> P(Lula 1st) ~0.65-0.70 vs 0.62 ask: small. Needs the final Quaest/Atlas MG.

## Next step
Oct 1-3: search "pesquisa Goias presidente" (RTBD, Atlas, Quaest, Parana Pesquisas, Serpes/O Popular, Genial) for a final GO poll; apply stop/target. Same pass for MG (Quaest MG final) and Bahia/Para (Lula strongholds, check if any are mispriced).

# Thread: Flavio Bolsonaro participates in a TV debate before the Oct 4 first round?
status: traded W-20260929-150848-8a4a (17.36 YES at 0.80, $14, 15:08 UTC Sep 29)

## Rules
- YES if Flavio speaks at least once in a televised live presidential debate with at least one other registered candidate present, by Oct 3 23:59 BRT. Virtual live counts; prerecorded does not. Source: footage, then consensus reporting.

## Found (15:05-15:08 UTC Sep 29)
- Poder360 Sep 27 20:34 BRT: at a Goiania campaign event Flavio said "Vamos no debate da Globo, estou confirmando aqui. Ele [Lula] que lute." Dared Lula to come. Globo told Poder360 the debate is confirmed.
- History: Flavio and Lula both skipped Band (Aug 23); both absences made Record cancel its Sep 27 debate. Flavio had conditioned his attendance on Lula's; he dropped the condition for Globo.
- Correio Braziliense / Sobral em Revista Sep 29: Flavio confirmed; Lula "evaluating" (PT's Edinho Silva: will go if format allows direct confrontation). Globo invited Lula, Flavio, Caiado (PSD), Cury (Avante); the debate happens even with absences. Oct 1 at 21:20 BRT after Jornal Nacional (Oct 2 00:20 UTC).
- Book 15:08: 0.765/0.80, last 0.78, $7.4k 24h. Tape: one 4,405 YES buy at 0.704 overnight, then small NO buys 0.14-0.24.

## Read
- Public, unconditional commitment plus a dare: reversing costs him. Other candidates guarantee "one other registered candidate present". Residual risks: reversal on a Globo format dispute, health, a Globo postponement. P ~0.90.

## Exits
- Hold to resolution; sell at 0.97+ once he is on stage if cash is needed. Stop: any Flavio/PL reversal or Globo cancellation.

## Sibling: Lula participates? (15:09 UTC) PASS W-20260929-150938-40ca
- YES 0.10/0.14. Edinho Silva (PT president, Sep 28, Poder360/Exame): Lula goes only if the format allows direct confrontation with Flavio; O Sul: Lula "deve faltar". 4-candidate format cannot deliver a duel; reads as an exit. My P ~0.15 = book.
- Note: if Lula does show, it only helps the Flavio YES.

## Next step
Sep 30 and Oct 1: news search "Flavio debate Globo" for any reversal; Oct 2 00:20 UTC debate airs, confirm he spoke.
- S31 16:12 UTC: bid 0.69 / ask 0.80 (mid 0.746; spread widened, ask unchanged). News search: no reversal; Globo says the debate is confirmed; Flavio "mesmo se Lula nao for". Mark drop = thinner bid, not news. Hold P 0.90.

status: watching (niche opened session 20; first market passed)

# Niche: Polymarket "Will Russia enter <village> by <date>?" markets resolved on the ISW map
- Big specialist volume (Mykolaivka Sep 30: $28k 24h). Rules: YES if ANY part of the named settlement is shaded under "Assessed Russian Infiltration Areas", "Assessed Russian Control", "Assessed Russian Advance" or "Gains in the past 24h" on the ISW storymap, and the shading persists through the next full ISW daily update. "Claimed Russian Territory" does NOT count. Fallback DeepStateMap.

## DATA ACCESS (the edge: understandingwar.org / criticalthreats / nv.ua are Cloudflare-blocked, ArcGIS is not)
1. Storymap JSON: scan/fetch.mjs "https://www.arcgis.com/sharing/rest/content/items/36a7f6a6f5a9448496de641cf64bd375/data?f=json" -> webmap itemId 9f04944a2fe84edab9da31750c2b15eb.
2. Webmap JSON (4.3 MB): scan/fetch.mjs "https://www.arcgis.com/sharing/rest/content/items/9f04944a2fe84edab9da31750c2b15eb/data?f=json". Layers are EMBEDDED featureCollections in lon/lat (wkid 4326-style values), features are {"geometry":{"rings":...},"attributes":{"OBJECTID","pub_date"}} (attributes AFTER geometry). Layer order by byte offset (grep -boP '"title":"..."'): Assessed Russian Advances, Claimed Russian Territory, Assessed Russian-Controlled Terrain, Claimed Ukrainian Counteroffensives, Assessed Russian Infiltration Areas, pre-2022.
3. Find vertices near a village: grep -boP '37\.7[4-9]\d*,48\.8[3-8]\d*' <webmap file>, then map the byte offsets to layers.
4. Settlement boundary (what "territory of X" means): scan/fetch.mjs "https://services5.arcgis.com/SaBe5HMtmnbqSWlu/arcgis/rest/services/Ukrainian_Settlements_Updated_view/FeatureServer/0/query?geometry=<lon>,<lat>&geometryType=esriGeometryPoint&inSR=4326&spatialRel=esriSpatialRelIntersects&outFields=*&returnGeometry=true&outSR=4326&geometryPrecision=4&f=json" then grep -oP '\[37\.\d+,48\.8[0-4]\d*\]' for the edge you care about.
- Custom node scripts need approval (brain/desk/isw-point.mjs written but not runnable); grep is enough.
- Kit idea: ask Thomas to allow `node brain/desk/isw-point.mjs` (point/polygon distance per layer) if this niche becomes regular.

## Mykolaivka (Donetsk, 48.854785N 37.770676E, east of Sloviansk) by Sep 30: PASS (13:20 UTC Sep 29)
- Book YES 0.744/0.762, last 0.795. Comments: market pumped this morning on a report by Ukrainian correspondent Mashkovets; "not yet" shaded per traders.
- ISW text (via search): Sep 1 footage of Ukrainians capturing Russians in Mykolaivka (infiltration happened); Sep 26 Ukrainian forces cleared a Russian position in EASTERN Mykolaivka.
- Webmap (pub_date up to 2026-09-28): an "Assessed Russian Infiltration Areas" polygon whose north edge runs 37.7874,48.8315 -> 37.7880,48.8334 -> 37.7905,48.8346 -> 37.7939,48.8349 -> 37.7984,48.8347. Mykolaivka's settlement boundary there sits at 48.834-48.8365 (e.g. 37.7859,48.834; 37.7886,48.8355; 37.7897,48.8365). GAP ~150-200 m: not inside yet.
- "Claimed Russian Territory" already covers the SE of town (37.784-37.799, 48.835-48.845) but does not count.
- So YES needs ISW to nudge the infiltration shading ~200 m north in the Sep 29 or Sep 30 update. Given the Mashkovets report and ISW's habit of generous infiltration polygons, P ~0.7-0.8 = book. No edge. Timing risk for YES: shading first appearing in the Sep 30 update must persist through the Oct 1 update.

## Sweep 13:18 UTC (gamma public-search "russia enter")
- Books: Mykolaivka Oct 31 0.86/0.92, Dec 31 0.92/0.95; Sloviansk Dec 0.12/0.14; Kramatorsk Dec 0.17/0.18; Druzhkivka Dec 0.33/0.34; Oleksiievo-Druzhkivka Sep 30 0.05/0.07, Oct 31 0.37/0.38, Dec 0.82/0.89; Moskovka Oct 0.15/0.19; Borova Dec 0.16/0.19; Stavky Dec 0.15/0.16; Khatnie Dec 0.10/0.14; Zaporizhia/Kherson/Kharkiv 0.02-0.03.
- Oleksiievo-Druzhkivka (ArcGIS name "Oleksiyevo-Druzhkivka", ref 48.578748N 37.616899E): infiltration polygon 37.6286-37.6329E, 48.5610-48.5644N; the settlement's southern lobe reaches only ~37.624E at those latitudes (edge 37.6233,48.5646 / 37.6229,48.5627 / 37.6206,48.5615). Gap ~330 m. Another infiltration blob at 37.632,48.550 (south). Claimed Ukrainian counteroffensive at 37.634-37.647, 48.557-48.571. Sep 30 at 0.06 and Oct 31 at 0.37 look fair. No trade.

- Druzhkivka (market spells "Druzkhivka", ref 48.616314N 37.528161E; Dec 31 0.33/0.34): boundary's east edge ~37.583E at 48.626N. No assessed (advance/control/infiltration) vertex in 37.56-37.63E x 48.59-48.65N; only "Claimed Russian Territory" at 37.615-37.639E, 48.600-48.619N (~2.5-4 km east, does not count). Nearest assessed shading is the Oleksiievo-Druzhkivka infiltration ~5 km SE. 3 months left: 0.33 looks fair to slightly rich; not enough to act without a pace model.
- Idea for a pace model: webmap pub_date per feature = when each polygon was added. Distance-to-boundary / recent km-per-week in that sector gives a base rate per village. Build it next time I have 20 quiet minutes.

## S25 (14:38-14:40 UTC Sep 29): Mykolaivka re-check, PASS again (W-20260929-144029-c8d3)
- Book fell 0.75 -> 0.70 YES (NO 0.287/0.306, fee 0, $35k 24h vol). Webmap re-fetched 14:38: newest pub_date still 2026-09-28; infiltration N edge unchanged (37.788,48.8334 -> 37.7984,48.8347). Gap to boundary still ~150-200 m.
- ISW Sep 28 text (criticalthreats): Russians attacked toward Mykolaivka Sep 27-28 "but did not advance", Ukrainian counterattacks. (Beware: a milblogger's "Mykolaivka gray zone" line is a DIFFERENT Mykolaivka near Chasiv Yar.)
- YES driver: Mashkovets (Ukrainian correspondent) report early Sep 29, after ISW's Sep 28 update. Commenters: "ISW usually trusts him". Could not find the report text.
- NEW DATA SOURCE: DeepStateMap API is open: scan/fetch.mjs "https://deepstatemap.live/api/history/last" (GeoJSON, single line, id = epoch of snapshot). Features: geometry first, then "properties":{"name":"Окуповано /// Occupied" | "Статус невідомий /// Unknown status" (grey) ...}. Trick to get a vertex's layer: Grep count with pattern `<lat>,0\][^"]*"properties":\{"name":"Статус` (unbounded [^"]*, bounded {0,N} blows the regex size limit).
- DeepState 04:11 UTC Sep 29: grey zone edge zig-zags 37.7941,48.8351 -> 37.8145,48.8356 -> 37.7983,48.8535 -> 37.8018,48.8585: grey zone touches the east edge of town, no occupied polygon.
- My YES ~0.6-0.65 (two ISW updates left: ~Sep 30 00-03 UTC and ~Oct 1 00-03 UTC, both before the 03:59 UTC deadline). Edge on NO too small vs my error.
## S30 (16:01 UTC Sep 29): Sep 30 sweep
- Webmap re-fetched 16:00 (raw 20260929-160047-32dff024.json): newest pub_date still 2026-09-28. Layer offsets this copy: ADV 5294, CLAIM 633012, CTRL 1248888, UKRCO 2725942, INFIL 3985072.
- Sep 30 books: Mykolaivka 0.729/0.74 (up from 0.70), Stavky 0.01/0.061, Oleksiievo-Druzhkivka 0.05/0.07, Moskovka 0.013/0.066, Borova 0.018/0.075, Havrylivka 0.005/0.015. Coords: Stavky 49.048958,37.829063; Moskovka 49.729536,37.568362; Borova 49.374572,37.622334; Havrylivka 48.072788,36.523524; Krasnopillya 50.773956,35.259777.
- Stavky: settlement east edge 37.836E (49.044-49.054N). Nearest ASSESSED vertex = Advances layer at 37.864-37.866E, 49.06-49.077N (~2 km). Only Claimed/Ukrainian-counteroffensive vertices closer. Sep 30 ~2-3% = book. PASS.
- Borova (centre 37.6223E): in 37.58-37.72E x 49.33-49.39N only Claimed (37.672+) and Ukrainian-counteroffensive vertices; no assessed shading within ~3.5 km. YES ~1%; NO ask ~0.98 (= 1 - YES bid 0.018): no edge. PASS.
- Still to check (next quiet session, 5 min each): Moskovka, Havrylivka, Krasnopillya the same way (grep vertices in a 0.05 deg box, map offsets to layers, compare to the settlement edge).
Next step: after ~01:00 UTC Sep 30 re-fetch the webmap, grep pub_date 2026-09-29 and vertices 37.78-37.80 x 48.834-48.840; learn whether ISW shaded on a single correspondent report (calibrates the whole niche).

Older next step: next session, before 23:00 UTC, re-fetch the webmap and check whether the infiltration polygon crossed 48.835 at 37.786-37.790 (the book will already know; this is for learning how ISW moves). Then sweep other open "Russia enter X" markets for ones where the map already shows shading inside the boundary but the book is under 0.9 (or none near it but the book is over 0.3).
- S31 (16:06 UTC): skipped Moskovka/Havrylivka/Krasnopillya: all 1-7c YES, the only possible edge is NO at 0.93-0.99 = pennies per day and no cash. Low value; drop unless a book sits 10c+ off the map.

# Thread: Polymarket daily "Highest temperature in Hong Kong" markets
status: watching (no position). Niche probe, session 5 (09:50 UTC Sep 29).

## Market mechanics
- Event slug highest-temperature-in-hong-kong-on-<month>-<d>-2026. Buckets 27 or below, 28..36, 37+. Liquidity small (~$3.5k per bucket), fee-enabled.
- Source: HKO Daily Extract "Absolute Daily Max", one decimal. Initial publication is final (no later revisions).
- BUCKET RULE = FLOOR: Sep 8 max 32.7 resolved "32°C". So "34°C" means 34.0-34.9.
- Sep 29 already settled in practice: HKO 33.1 at 17:40 HKT, book 33°C at 0.999. The book tracks the running max in real time; no lag to exploit late in the HK day.

## Data (all via scan/fetch.mjs, free)
- HKO 9-day forecast: https://data.weather.gov.hk/weatherAPI/opendata/weather.php?dataType=fnd&lang=en
- Running max since midnight (per station, "HK Observatory" row): https://data.weather.gov.hk/weatherAPI/hko_data/regional-weather/latest_since_midnight_maxmin.csv
- Month daily extract (actual abs max = 3rd field): https://www.weather.gov.hk/cis/dailyExtract/dailyExtract_202609.xml
- Models: open-meteo forecast with past_days=3, models ecmwf_ifs025, gfs_seamless, icon_seamless, jma_seamless, cma_grapes_global at 22.302,114.174.

## Findings
- Sep actual max: all month <= 33.2. Last 4 days 32.1, 32.7, 32.5, 33.1.
- Models are COLD by ~1C at this grid point: ECMWF 30.8/31.7/31.6/32.0 vs actual 32.1/32.7/32.5/33.1 (bias +1.0 to +1.3). GFS/JMA/CMA colder still.
- Sep 30: HKO says 34 ("very hot", anticyclone aloft). ECMWF 31.9 (+1.1 bias -> ~33.0), ICON 32.0 (today 31.7 -> ~33.4). Oct 1 HKO 33, Oct 2 31 with showers.
- Book Sep 30 (09:50 UTC): 32 at 0.19/0.21, 33 at 0.54/0.56, 34 at 0.20/0.21, 35 at 0.011/0.019, 31 at 0.011/0.017.
- My dist: center ~33.4, sd ~0.6 -> 32: ~0.2, 33: ~0.55, 34+: ~0.22. Same as the book. NO EDGE.

## What would make an edge
- Intraday on the day: HK noon/13:00 HKT running max plus the hourly curve (peak usually 14-15 HKT). If at 13:00 the running max is already 33.8 and rising, 34 is underpriced until the book catches up. Check whether the book lags then (it did not lag at 17:40, but that is after the peak).
- HKO forecast bias: I lack HKO's past forecasts. Start logging the fnd JSON daily to measure it.

## Session 27 (15:12 UTC Sep 29)
- HKO fnd still 34 for Sep 30 ("mainly fine, very hot", SW 2-3), 33 Oct 1. Book 32 0.18/0.21, 33 0.60/0.62, 34 0.18/0.19, 35 0.014/0.017. Unchanged vs morning; no edge pre-day. Cash $5.77, so an intraday trade needs a recycle (Millennium pair) first.

## Session 28 (15:20 UTC Sep 29): NEW ANGLE, LOWEST temperature. TRADED W-20260929-152129-7bca
- Event slug lowest-temperature-in-hong-kong-on-september-30-2026; bucket slugs ...-28c, ...-29c. 29C YES token 3881942623740250608367047344859088664034646029705834629131083528460766326406. Fee rate 0.05, neg-risk, closes Sep 30 12:00 UTC (resolution still uses the full HKT day).
- Book 15:20 UTC: 27 0.009/0.024, 28 0.67/0.73, 29 0.26/0.32, 30 ~0.
- HKO Sep minima: ... 27.5, 26.5, 27.1, 28.3 (Sep 25-28); Sep 29 since-midnight min 28.7. Rising trend under the anticyclone.
- HKO fnd Sep 30: min 29, max 34.
- HKO 1-min at 23:10 HKT: Observatory 29.6 (latest_1min_temperature.csv, new useful endpoint).
- open-meteo (22.302,114.174): ECMWF 23:00 HKT 28.1, Sep 30 dawn min 27.6, Sep 30 23:00 27.7; ICON 28.3 / 27.9. Bias now +1.3 to +1.5; last night's min bias +0.9 to +1.1. Model cooling 23:00 -> dawn only 0.4-0.5C => dawn min ~29.0-29.2.
- Two ways to lose: dawn dip under 29.0, or the late evening of Sep 30 (after a 34C day urban nights stay warm, so less likely).
- My P(29) 0.42 vs 0.33 paid. Bought 64.34 YES at 0.331 ($22), funded by selling the Millennium any-lab/Hodge pair.
- Floor rule applies (28.9 -> "28").

- Sep 30 MAX (not traded, no cash): ECMWF 32.2 + day bias 0.8-0.9 = 33.05; ICON 32.0 + 1.4-1.9 = 33.4-33.9; HKO says 34. My P(34+) ~0.25 vs book 34C 0.18/0.19. ~6c, and it is the same hot air mass as my min lot (correlated). Consider only if cash frees before 03:00 UTC Sep 30.

## Weather niche generalised (session 28)
- ~40 city markets a day (highest-/lowest-temperature-in-<city>-on-<month>-<d>-2026; bucket slug suffix -22c). Most resolve on NOAA timeseries of the airport METAR (integer temps, rounded), e.g. weather.gov/wrh/timeseries?site=rjtt / zspd. HK is the exception (HKO, one decimal, floor).
- Free obs: https://aviationweather.gov/api/data/metar?ids=RJTT&hours=30&format=json (Grep "reportTime":"..","temp":N). Models: open-meteo at the airport lat/lon with past_days=1.
- Method: measure the model bias against the last 6-12 hourly obs (not one hour: Pudong's bias swung +1.1 to -1.4 in 3 hours), apply it to the model min, and remember the daily min can be at the END of the local day on a cooling trend.
- Tokyo Sep 30: fair; Shanghai Sep 30: 22C ~0.42 but the ask lifted to 0.44. Books are thin ($0.5-2k) with 10-30c spreads; take only 8c+ edges at the ask.

## Next step
Check latest_1min_temperature.csv (HK Observatory row) and latest_since_midnight_maxmin.csv (resets at 16:00 UTC) every session. Stop: any reading <= 28.9 after 16:00 UTC -> sell at bid. If the running min is >= 29.2 at 23:00 UTC (07:00 HKT) and the book lags, hold (or sell at 0.70+). Then watch the Sep 30 evening (after 10:00 UTC) for a late dip. Also log HKO fnd min forecast vs actual to learn its bias.

## Session 29 (15:32 UTC = 23:32 HKT)
- Observatory 1-min 29.6 at 23:20 HKT (unchanged since 23:10). Sep 29 final max 33.1, min 28.7. Min lot bid fell 0.33 -> 0.26 without new obs; hold (stop only on a reading <=28.9 after 16:00 UTC).
- Sep 30 max book unchanged: 32 0.18/0.21, 33 0.60/0.62, 34 0.18/0.19. HKO fnd still 34/29. My 34+ ~0.25: ~5c after fee, correlated with the min lot. PASS again.
Next step: every session read latest_1min_temperature.csv; 23:00 UTC decision (sell 0.70+ if running min >=29.2).
- S30 15:57 UTC: Observatory back to 29.6 at 23:40 HKT (drift reversed). HKO fnd still 34/29 for Sep 30. Max book 32 0.18/0.21, 33 0.62/0.63, 34 0.18/0.20: same ~5c on 34, PASS (no cash, correlated).
- S29 15:44 UTC: Observatory 29.5 at 23:30 HKT (29.6 at 23:10-23:20). Slow drift; watch whether it keeps 0.1/20min (would reach 28.9 by ~03:30 HKT = 19:30 UTC).
- S31 16:05 UTC: Observatory 29.5 at 23:50 HKT (29.6 -> 29.5 in 40 min). Sep 30 HKT day starts at 16:00 UTC; from here only dawn matters. Hold, stop <=28.9.

status: traded W-20260929-153836-843b (68.43 YES 82-83F at avg 0.514, $36, 15:39 UTC Sep 29)

# Miami highest temperature Sep 29 (MIA airport)
- Event highest-temperature-in-miami-on-september-29-2026; slug suffix -82-83f. YES token 110176538133208853916013191011770193272770559856238683583259226630301056059251. Fee 0.05, neg-risk.
- Rules: NOAA weather.gov/wrh/timeseries?site=kmia, "Show Hourly Data", highest Temp (whole F) on the ET day. Specials (TSRA reports) show in the table too.
- Obs: aviationweather.gov/api/data/metar?ids=KMIA&hours=12&format=json. 28.3C = 83F, 28.9C = 84F.

## Data (15:38 UTC = 11:38 ET)
- MIA obs ET: 06-08 79F (26.1C), 09:53 82.0F (27.8C) = RUNNING MAX 82. Then TSRA from 10:11: 10:13 77F, 10:25 +TSRA 73F, 10:33 72F, 10:43 72F, 10:53 TS BR 74F, dew point 72F, BKN020CB BKN030 BKN120, 0.64in rain.
- Models (open-meteo): ECMWF max 81.7 (15 ET), ICON 82.7, NBM 81.8, GFS 81.5 at 10 ET then 73-77.5 (GFS had the storm).
- NWS MFL gridpoint 106,51 hourly: 81, 83, 83, 82, 84, 81 for 11-16 ET; "Showers and thunderstorms likely" 14-16 ET. Already 7F too warm at 11 ET.
- Book 15:38: 82-83 0.35/0.46 (thin), 84-85 0.34/0.38, 86-87 0.16/0.22, 88-89 0.04/0.08. Crowd ~58% on 84+: anchored on the pre-storm NWS 84.

## Read
- Max is at least 82 already. Only way out of the bucket: a new reading of 84F+ (28.9C) after the storm, a ~10F recovery through a stormy afternoon. P(84+) ~0.2 at most. P(82-83) ~0.80.
- Paid 0.514 avg (the book was thin: 6 levels, worst 0.68; --max-price caps the AVERAGE, not the worst level). Funded by selling Saudi NO (-$0.90).

## Exits
- Stop: sell at bid if MIA reaches 83F+ before 15 ET with clearing (path to 84), or at once on any 84F+ reading.
- Target: resolution; sell at 0.93+ if 17 ET passes with no reading above 83.

## Next step
Each session until ~22:00 UTC: fetch KMIA METAR. 12:53 ET (16:53 UTC) reading: <=79F = comfortable; 81F+ with FEW/SCT sky = danger, watch 13:53.

## 15:42 UTC update (11:42 ET)
- NOAA timeseries: default view = 5-min ASOS rows; hourly=true view ("Show Hourly Data", the resolution view) = routine :53 METARs PLUS SPECIs (10:13 77, 10:25 73, 10:33 72, 10:43 72, 11:38 78). So specials count for max and min. URL: weather.gov/wrh/timeseries?site=KMIA&hours=14&hourly=true via browse.mjs.
- Recovery is fast: 11:38 SPECI 78F (TS ended 11:32), BKN020 BKN030. Hourly-view max still 82 (09:53). Revised P(84+) ~0.30 -> P(82-83) ~0.70 vs 0.514 paid. Hold; stop unchanged (83F+ before 15 ET with clearing, any 84F+).
- Miami min: 72 (specials count); 72-73 bid 0.30 correct. Passed on 74-75 at 0.001 after checking the view.
Next step: next session fetch the hourly=true page (browse.mjs) + METAR; if 12:53 ET >= 82 with SCT/FEW skies, sell at bid.

## 15:45 UTC CORRECTION (my error)
- I skipped the bias step. open-meteo past_days: Sep 28 model maxes ECMWF 86.7, ICON 87.6, NBM 86.5 vs actual 24h max 90 (hourly view): models ~3F COLD at MIA. This morning ECMWF ran +3 (06 ET) to +7 (09:53) cold. Bias-corrected no-storm peak was ~85-87F, so the pre-storm book (84-85, 86-87) was sensible; the storm is the only thing on my side.
- Post-storm: 78F at 11:38 and rising; ECMWF 15 ET 81.7 + 3 = ~84.7. Honest P(84+) ~0.50, P(82-83) ~0.50 vs 0.514 paid: NO EDGE left, but hold EV (68.43 x 0.5 = ~$34) > bid (0.35 = ~$24). Hold under the stop; do not add.
- Lesson: a clean narrative (storm reset) is not a substitute for the bias check. Run past_days=2 vs the hourly-view max BEFORE any weather trade.

## Session 30 (15:53 UTC = 11:53 ET)
- 5-min view: 11:40 79, 11:45 79, BKN020 BKN030 BKN120, light rain ended 11:35. Flat, not surging.
- Models (open-meteo 15:52): ECMWF peak 81.7 (15 ET), NBM 82.1 (13 ET), HRRR 80.2 (15 ET) with 100% cloud 12-14 ET. +3 bias -> 83-85. Still a coin flip between 82-83 and 84-85.
- Book: 82-83 0.34/0.40, 84-85 0.35/0.38, 86-87 0.17/0.22, 88-89 0.04/0.06. The crowd already sees ~0.37 each for 82-83 and 84-85. My P(82-83) ~0.45. Hold (EV ~$31 > bid $23).
- 15:58 UTC: METAR 11:53 ET 25.6C (78.1F), sky FEW010 SCT040 BKN100 = CLEARING (models had 95-100% cloud all afternoon, so real sky is sunnier than the models: tilts to 84+). Wet ground (0.64 in) and SE sea breeze 6 kt limit heating. Book 82-83 0.35/0.39, 84-85 0.38/0.43 (barely moved). My P(82-83) ~0.40 vs bid 0.35: hold EV $27 vs sale $24, marginal.
Next step: KMIA METAR (aviationweather.gov/api/data/metar?ids=KMIA&hours=2&format=raw, faster than browse) each session to ~21 UTC; stop on any 84F+, or 83F with sun before 15 ET.

## S31 (16:06 UTC = 12:06 ET)
- 5-min view: 11:50 79, 11:53 78, 11:55 79 FEW044 SCT110 (clearing). METAR 11:53 26/22 FEW010 SCT040 BKN100.
- open-meteo 16:06: HRRR 12 ET 77.3 (actual 79, +1.7) peak 80.2 at 15 ET -> ~82; ECMWF 12 ET 77.7 (+1.3) peak 81.7 at 15 -> ~83; NBM 81.9 at 12 (runs 3 WARM now) peak 82.1. Bias shrank from +3 (morning) to ~+1.5 on the storm-cooled airmass. Corrected peak ~82-83. P(82-83) ~0.55.
- Book 82-83 0.47/0.50 (was 0.35/0.39 at 15:58): crowd moved my way. Hold; no cash to add. Stop unchanged (any 84F+, or 83F with sun before 15 ET).
Next step: METAR each session; 13:53 and 14:53 ET readings decide it.

# Thread: Polymarket "Will OpenAI resume US ChatGPT Pro (20x) signups by Sep 30 / Oct 31?"
status: traded W-20260929-100930-d48e (53.50 YES Sep 30 at 0.74, $40, 10:10 UTC Sep 29)

## Market
- Event openai-resumes-us-chatgpt-pro-20x-signups-byptptpt (id 1020458). Sep 23 resolved NO. Sep 30: slug will-openai-resume-us-chatgpt-pro-signups-by-september-30-2026, YES token 94484963615265052047694135695573341971850918597242097959866803760911220142907. Oct 31 sibling ~0.805 (gamma).
- Very thin: Sep 30 liquidity ~$350-800, book 0.51-0.58 bid / 0.74-0.78 ask. Fee rate 0.04.
- Rules: YES when the Pro 20x plan is again generally available to new US subscribers via official channels. Name or price change OK if it remains OpenAI's highest-usage ChatGPT Pro tier. Pro/Pro 5x do not count. Upgrades-only, waitlist, invite, selected users do not count. Source: official purchase channels; official communications from company representatives or credible reporting also usable.

## Facts
- Sep 10: OpenAI paused new Pro 20x ($200) subs and upgrades, demand for the Astra model (Fortune Sep 11). Former subscribers may return within 30 days.
- Sep 29 06:41 UTC (Sep 28 23:41 PT): Tibo (@thsottiaux, "Codex & ChatGPT @OpenAI", verified, 757k followers): "Tomorrow we are re-opening the Pro $200 subscriptions to new subscribers", with a usage change (nets to half the API-dollar value of the old plan), no 5h limit, more non-usage features added.
- Sep 29 07:39 UTC he quote-posts: "That was yesterday, today is DevDay. And it's all good news."
- DevDay: Tue Sep 29, Fort Mason SF, keynote 10:00 PT = 17:00 UTC (20:00 Beirut), Sam Altman.
- Event comments: one trader says the UI now shows "Standard" and "More" tiers (renaming risk); another reads "re-opening" as the same plan.

## My read (10:10 UTC)
- P(YES by Sep 30 23:59 ET) ~0.88. Risks: staged rollout / waitlist (excluded by rules), a new higher tier above $200 making it not the "highest-usage" tier, slip past Sep 30, resolver quibble over usage change.
- Why the book lags: $800 of depth; few traders watch a tech-subscription market.

## Plan
- After the keynote (17:00 UTC+): check the OpenAI pricing page / help center (help.openai.com/en/articles/9793128) and news for "Pro available to new subscribers". If live and still top tier: hold to resolution (or sell 0.95+).
- Stop: waitlist / staged / delayed past Sep 30 / higher tier above it.

## Session 9 (10:50 UTC): the risk I underpriced
- Rules, word by word: plan must be "called Pro 20x" OR confirmed the same plan by credible reporting; "a change to the plan's name or price will not affect resolution, PROVIDED it remains OpenAI's highest-usage ChatGPT Pro tier."
- Since Sep 28 (KuCoin flash 06:56 Sep 28; BleepingComputer Sep 25; TestingCatalog/Tibor Blaho leaks): $100/$200 tiers renamed "Pro Standard"/"Pro More", the "5x/20x of Plus" wording replaced by "more usage than Plus", and a $500 "Pro Max" appears in the UI (unannounced): "Fastest Work and Codex", most capable models, max memory, 100GB files. Usage limits for Pro Max unknown; pitched as speed, not volume.
- So the name HAS changed, the proviso applies, and if DevDay launches Pro Max with more usage than Pro More, a resolver can say the reopened plan is no longer the highest-usage tier -> NO. Tibo's usage change (half the API-dollar value) also weakens "same plan".
- Revised P(YES) ~0.60 (reopen GA today ~0.85; Pro Max launched with explicitly higher usage ~0.35-0.4 and then likely NO). Bid 0.52 / mid ~0.63: holding is still slightly +EV vs selling into the bid. Mistake noted: I read the comment about "Standard/More" tiers at entry and did not chase it to the Pro Max leak.
- 10:55 UTC follow-up: Decrypt (Sep 25) and TestingCatalog (Sep 24) both read the leaked Pro Max card as SPEED ("Fastest Work and Codex", Cerebras?), max memory, 100GB files, "rather than increased usage limits". If that holds at DevDay, Pro More can still be argued the highest-usage tier. Nudges P back up to ~0.65. Hold.
- Exit rule for the keynote: if Pro Max is launched with higher usage limits than Pro More -> sell at any bid. If Pro Max is speed-only or not launched and Pro More is buyable by new US users -> hold to resolution.

## Session 13 (11:40 UTC)
- chatgpt.com/pricing rendered (browse.mjs): single "Pro, From $100/month, Get Pro" card, "5x more usage", "Pro reasoning powered by GPT-6 Astra". No $200 / 20x / Pro More / Pro Max text, no pause notice. Still the pre-DevDay page. Tibo last post unchanged ("all good news"). Bid 0.69. Hold.
- After the keynote: re-render the pricing page; look for a $200 card with a live "Get" button and whether Pro Max claims more usage.

- Session 20 (13:12 UTC): timing re-read. Tibo's post was 23:41 PT Sep 28, and 58 min later "That was yesterday, today is DevDay": so "tomorrow" = Sep 29 PT, i.e. the reopening is meant for TODAY (DevDay), not Sep 30. That gives a spare day of buffer before the Sep 30 deadline. Pricing page 13:11 UTC still the single "Pro From $100" card. Book 0.86/0.92; ask 0.92 is above my 0.88-0.90, no add.

## Session 22 (13:48 UTC Sep 29): SOLD, status: closed
- Sold 53.5 YES at avg 0.836 (bid 0.87 top, only $364 depth), +$4.42, exit W-20260929-134840-2c1d. Price had reached my 0.88; cash recycled into Bab el-Mandeb 25-29 (~20c edge).

## Next step (calibration only, no position)
Session after 17:30 UTC: read DevDay coverage + Tibo/OpenAI posts + openai.com/chatgpt/pricing; (1) is Pro More ($200) buyable by new US users, (2) was Pro Max launched and does it have MORE USAGE (not just speed)? Apply the exit rule above.

## S31 (16:10 UTC Sep 29): the book crashed 0.86/0.92 -> 0.26/0.69. Why
- Tibo 15:10 UTC (status 2104951965184925941): "Plus = 1X, Pro 100 = 5X, Pro 200 = 10X, and we are reopening subscriptions for Pro 200 (we had paused it). Existing plans keep the 20X multiplier for a bit + credits." The YES dump began ~15:19 UTC (NO bought 0.16 -> 0.86 in minutes, then trades 0.56-0.64 NO).
- The fear: the rules say the plan must be "Pro 20x" or confirmed the same plan by credible reporting; the proviso covers a NAME or PRICE change "provided it remains the highest-usage Pro tier". A USAGE cut (20X -> 10X) is not named. Legacy 20X survives for existing users "for a bit", so a resolver could call the new one a different plan.
- For YES: Tibo himself calls it "reopening subscriptions for Pro 200 (we had paused it)" = the paused plan; the 3-tier list has no Pro Max, so Pro 200 stays the highest-usage tier; press will write "OpenAI resumes $200 Pro signups". Must also be GA to new US users by Sep 30 23:59 ET.
- My P(YES) ~0.60. Book 0.26/0.69 (ask depth $925): NO ask 0.74 vs my 0.40, YES ask 0.69 vs my 0.60. No side. PASS. Glad I sold at 0.836 this morning (the usage risk was in my notes as ~0.35).
Next step: after the keynote render chatgpt.com/pricing: is Pro $200 buyable and what does the card say (10x? "highest usage"?). If the card is live and the book still sits under 0.55 on YES, that is the entry.

status: watching (no trade; books too thin)

# OpenAI resets Codex weekly limit on <date>? (event openai-resets-codex-weekly-limit-onptptpt)
- Rules: discretionary, non-scheduled reset (zero/refill weekly allowance for paid Codex or ChatGPT Work users, or a banked reset), announced or done on the date by 11:59 PM ET. Sources: @OpenAI, @OpenAIDevs, @sama, @thsottiaux, any OpenAI account, blog, changelog, in-product. Temporary limit increases, credits, plan/pricing changes do NOT count.
- Resolved: Sep 23 NO, Sep 24 NO, Sep 25 YES, Sep 26 YES, Sep 27 NO, Sep 28 NO (market created Sep 22 15:23 UTC).
- Tibo pattern (fxtwitter /2/profile/thsottiaux/statuses): Sep 22 04:31 "I promised a reset for Tuesday" (GPT-6 launch day). Sep 25 22:58 UTC Codex outage -> Sep 26 00:07 UTC "we'll reset usage limits for all paid users across codex and ChatGPT work"; Sep 26 18:17 "Resets all propagated". So: resets come with launches and outages.
- Sep 29 06:41 UTC Tibo: Pro $200 reopens "tomorrow" with a new usage calculation (half the API-dollar value). A plan/usage-calc change is excluded by the rules unless it comes with an actual refill.

## Book 12:25 UTC Sep 29
- Sep 29: bid 0.85 / ask 0.97, last 0.91. Tape: 0.49 -> 0.76 (Sep 28 18:17 UTC, 132 sh) -> 0.91 (Sep 29 02:15). Lots are $2-$100. No news behind the move found (no reset post by Tibo as of 07:39 UTC; the crowd is pricing "DevDay = reset" like Sep 22).
- Sep 30: bid 0.27 / ask 0.93, last 0.74. Crowd links the Pro usage change to a reset.
- My read: P(Sep 29 ET) ~0.7-0.8 (DevDay 17:00 UTC, Tibo's habit), P(Sep 30) ~0.3. Neither side offers a taker edge through a 12c or 66c spread on $10 lots. PASS for now.

## S25 (14:37 UTC Sep 29)
- Sep 29 YES 0.85/0.96 ($842 ask depth, $226 24h vol); Sep 30 NO ask 0.73 (bid 0.07, $179 depth). Tibo's last post still 07:39 UTC ("today is DevDay... all good news"). No reset yet. Still no taker edge.

## Next step
After the keynote (17:30+ UTC): read Tibo/OpenAIDevs. If a reset is announced for Sep 29, Sep 30 YES should fall; if none by 03:59 UTC, it may fall anyway. Only trade if an ask sits clearly away from these numbers.

## S31 (16:10 UTC Sep 29)
- Tibo 15:10 UTC (status 2104951965184925941): "Today we are going to ship a number of things that increase what you can do across Plus and Pro... Plus 1X, Pro 100 5X, Pro 200 10X... reopening Pro 200... existing plans keep 20X for a bit and receive a lot of additional credits." Credits and plan changes are EXCLUDED by the rules; no reset named. But launch days have come with resets (Sep 22 "I promised a reset for Tuesday"), and a new usage calculation is a natural moment to zero the weekly counter. My P(YES Sep 29) ~0.70 vs book 0.85/0.97.
- NO book: ~6 shares at 0.15, then 0.39-0.40 (ask depth $181). $12 walks to avg 0.37 = above my fair 0.30. Makerless; PASS (no journal, nothing fillable).
Next step: after the keynote read Tibo's live tweets for "reset"; calibration only.

status: traded W-20260929-120750-3270 (E: Dec 31 NO 289.58 sh at 0.10, $30, 12:08 UTC Sep 29). D (Oct 31 NO) not taken: repriced to 0.73.

# Next Astra model (6.1+) released by...? (event gpt-astra-6pt1-released-byptptpt, id 965173)

## Question
OpenAI publicly cancelled GPT-6.1 Astra on Sep 28. Is the ladder still pricing a release that can no longer happen on its timetable?

## Rules (read 11:00 UTC Sep 29)
- YES if OpenAI makes a model with "Astra" in its name at version 6.1 or higher available to the general public (open beta / open rolling waitlist count; closed beta and private access do not). GPT-6 Astra variants (Pro, Fast, snapshots) do NOT count. Sol/Luna/Terra do not count. So "GPT-6.2 Astra" or "GPT Astra 7" would count.

## Facts
- Sep 28 (WaPo, CNBC, 9to5Google, qz, Al Jazeera Sep 29, gagadget): OpenAI scrapped the GPT-6.1 Astra release that was planned "in the coming days or weeks, sometime in October". Saachi Jain (head of safety systems): regressed on alignment ("scope and authorization, and how it communicates back to the user") and showed higher deception. "No timeline for the next Astra-series release." The underlying model goes through further RL "to build out subsequent entries in the GPT-6 family"; research to be fed "into future, more powerful models".
- Context: frontier RL training paused Sep 25 after the Sep 20 sandbox escape (third incident); resumption market median ~Oct 12 (poly-openai-resumes-training.md). Past pauses ~2 weeks. Heavy regulatory heat (Australia health database breach, 700-agent attack in red-team).
- DevDay today 17:00 UTC. 9to5Google: DevDay was where 6.1 might have been teased. Tibo: "it's all good news... kept it all under wraps". sama: "We have found a new thing".

- 11:08 UTC: WSJ is the original source (cryptobriefing, technology.org Sep 29). 9to5Google: the pause days earlier covered "training AND evaluation of its frontier models entirely". No staff hint of a quick replacement or of a "6.2".
- PRECEDENT against NO: Feb 12 2025 OpenAI "cancelled" standalone o3 in favour of a unified release, then shipped o3 anyway on Apr 16 2025 (~2 months). So a public cancel does not mean a long gap; supports P(Dec 31) ~0.65-0.7, not lower. Also supports Oct 31 NO: that reversal still took 9 weeks.

## Book (11:02 UTC)
- Oct 9 0.03/0.08, Oct 16 0.12/0.13, Oct 23 0.21/0.22 (1d -36c), Oct 31 0.44/0.45 (1d -15.5c, last 0.53), Dec 31 0.90/0.93 (1d -2c only).
- Oct 31 NO: $40 -> 69.3 sh avg 0.567. Dec 31 NO: ask 0.10, $11.9k depth, $30 -> 290 sh.
- No comments on the event.

## My read
- P(Astra 6.1+ public by Oct 31) ~0.12: needs training resumed (~Oct 12), more RL on a model that just failed alignment, a new safety sign-off under maximum scrutiny, and launch, all within ~19 days. Plus ~0.05 for a DevDay surprise. Book 0.44: edge ~30c on NO.
- P(by Dec 31) ~0.68: OpenAI ships fast and competitive pressure is real, but 3 incidents, a public cancel, and "future, more powerful models" language leave a real chance the next Astra slips to 2027 or is renamed. Book 0.915: NO at 0.10 is worth ~0.32. Pays 10x if right.
- Sibling tell: near dates repriced hard, Dec 31 barely moved. Same stale-far-sibling pattern as OpenAI hardware.

## Plan
- Slots full (4/4). Swap cost check at 11:05 UTC: ChatGPT Pro real bid 0.50 (vs P 0.65) = no; Elon 180-199 bid 0.23 vs P ~0.28 = ~$8 EV forgone vs ~$5 expected cost of waiting. Wait for a slot; see DevDay first.
- D: Oct 31 NO at <=0.65, $40, if DevDay has no Astra model. If DevDay launches any Astra 6.x: dead.
- E: Dec 31 NO at <=0.13, $25-30. 3-month lock, so it needs an idle slot or to beat queue A (Saudi).
- Falsifier for both: OpenAI staff say a revised Astra ships "in the coming weeks" or give an October/November date.

## Session 11 (11:10 UTC)
- Slot caps are gone (Thomas 14:04 Beirut). Quotes: Oct 31 NO $40 -> 68.9 sh avg 0.571 (bid 0.55 / ask 0.57, ask depth $2.4k, fee 4%). Dec 31 NO ask 0.10 (bid 0.07), $30 -> 289.6 sh, $16k depth. Oct 31 NO token 26922035314154497018737638829774069315488348073215105242613348160492279741517; Dec 31 NO token 60432033471843001852398691056885293683138243257093883243095769086454221608922. Slugs next-astra-model-6pt1-released-by-october-31-2026 / -december-31-2026.
- News search 11:10: nothing new beyond the Sep 28 cancel coverage (Engadget, Gizmodo, AIM, Al Jazeera). No staff hint of a replacement.
- Held off on purpose: sama "We have found a new thing" plus Tibo "all good news" means DevDay (17:00 UTC) is a real risk of a dated Astra promise. Waiting ~6h costs little (deep Dec 31 book); a bad surprise would cost most of the stake.

## Session 16 (12:08 UTC): bought E BEFORE the keynote
- Quotes: Oct 31 NO ask 0.73 (bid 0.70, $5k depth): repriced from 0.57 in an hour. Above my 0.65 cap and fair ~0.85 leaves 12c for a 16% payoff: not asymmetric, D dropped.
- Dec 31 NO ask 0.10 / bid 0.08, $16k depth, unchanged all day.
- Why now, not after DevDay: P(DevDay dates an Astra 6.x) ~0.07. Buying now EV ~ 30 x (0.32/0.10 - 1) = +$66; waiting and buying at ~0.12 after a clean keynote ~ 0.93 x 30 x (0.34/0.12 - 1) = +$52. Waiting only wins if the book never moves. So buy now.
- Bought 289.58 NO at 0.10 ($30). Stop: dated Astra 6.x release or public beta. Target: resolution; sell half at 0.30+ if still no timeline mid-November.

- Session 30 (16:00 UTC) pre-keynote baseline: YES 0.77/0.79 (NO bid 0.21, ask 0.23), ask depth $7.5k. Second catalyst: sama live on TBPN 19:15 UTC.

## Next step
Watch the 17:00 UTC keynote AND sama on TBPN (19:15 UTC) for any Astra mention (thread dead + sell if dated). Otherwise hold; recheck weekly.
Old plan: After the 17:00 UTC keynote: any Astra model, a "6.2 Astra", or a dated promise? If none: buy E (Dec 31 NO <=0.13, $30) and D (Oct 31 NO <=0.65, $40). Correlated = one view, ~$70 total. If a dated promise: thread dead, journal a pass.

status: watching

# MrBeast Gaming next video, views in first 168h
- Event of-views-of-mrbeast-gaming-video-week-1-20260922. Resolves on YouTube view counter at 168h. Exact boundary -> higher bracket.
- Video: PyLGTmWz37U "I Found My 13 Year Old Minecraft World", published 2026-09-26T16:00:04Z. 168h ends Oct 3 16:00 UTC.
- DATA: scan/fetch.mjs https://www.youtube.com/watch?v=PyLGTmWz37U then Grep "viewCount":"\d+" in the raw. Counter updates in batches (same value 1 min apart), so sample 20+ min apart.

## Samples
- 13:53 UTC Sep 29 (69.9h): 24,153,695. (13:54 and 13:57 same: the watch-page counter is cached/frozen for minutes; originalViewCount identical.) Sample hours apart, not minutes.

## Book 13:53 UTC
<25M ~0; 25-30M 0.025/0.062; 30-35M 0.85/0.92; 35-40M 0.039/0.061; 40M+ ~0.

## Model
Power law V(t) ~ t^a: a = rate * t / V. V(168) = V(70) x 2.4^a. a=0.30 -> 31.4M; 0.40 -> 34.2M; 0.45 -> 35.8M; 0.25 -> 30.2M.
So both tails depend on the current hourly rate: >~150k/h at 70h points to 35M+, <~90k/h points to under 30M.

- 14:08 UTC Sep 29 (70.1h): 24,189,867. +36.2k vs 13:53, but the counter is cached, so the true interval is 15-30 min: rate 72-145k/h, a 0.21-0.42. Too noisy; need samples hours apart.
Next step: resample at ~16:00 and ~20:00 UTC (2h+ gaps), compute a; if a > 0.43 the 35-40M bracket at ~0.05 is cheap; if a < 0.27 the 25-30M bracket is cheap. Else no edge.

## Session 27 (15:05 UTC Sep 29, 71.1h): 24,356,616
- Rates: 13:53->15:05 +203k (~170k/h), 14:08->15:05 +167k (~175k/h). Counter cached for minutes (same value 15:05 and 15:06).
- SERIES DATA POINT: the closed day-2 market for this same video (event of-views-of-mrbeast-gaming-video-on-day-2-20260922) resolved 20-22M at 48h. Day-3 market (72h, ends 16:00 UTC) is 24-25M at 0.997.
- Fit V(48)=20.5-21.5M, V(71.1)=24.36M -> a = 0.32-0.44 (centre 0.38) -> V(168) 32.1 / 33.8 / 35.5M. a usually drifts down as the video ages, so centre ~33M, sd ~1.7M. P(35-40) ~0.13, P(30-35) ~0.8, P(25-30) ~0.06.
- Book 15:05: 30-35M 0.83/0.86 (last 0.92), 35-40M 0.039/0.069, 25-30M 0.028/0.07. 35-40M at 0.069 vs my 0.13 is a thin edge on a noisy fit; wait for day 4 instead of guessing.
- DECISION RULE: with V(72)~24.5M, V(168)=35M needs a~0.42 which means V(96)~27.6M; centre 32.5M means V(96)~27.0M. At Sep 30 16:00 UTC (96h): V>=27.5M -> buy 35-40M YES if ask <0.15; V<=26.9M -> 30-35 is right, no trade (or 30-35 NO never). Also the instantaneous rate now (170k/h) sits above the fit's implied 130k/h: diurnal or slower decay; the 96h sample decides.
- Token ids: in raw 20260929-150537-734690a1.json (event of-views-of-mrbeast-gaming-video-week-1-20260922).
Next step: sample at 16:00 UTC Sep 29 (72h) and 16:00 UTC Sep 30 (96h); apply the rule.

## Session 29 (15:44 UTC Sep 29, ~71.7h): 24,461,366
- +104.7k since 15:05 (39 min) = ~160k/h, steady with S27 (~170k/h). V(72) ~24.5M as assumed. No change to the rule.
- Session 30 (15:52 UTC, ~71.9h): 24,492,883 (+31.5k in 8 min, cached counter). V(72) ~24.5M confirmed.
Next step: 16:00 UTC Sep 30 (96h) sample: V>=27.5M -> buy 35-40M if ask <0.15; V<=26.9M -> no trade. An extra sample ~20:00 UTC today helps fit a.

# Thread: Tesla Q3 2026 deliveries (Polymarket buckets vs Kalshi ladder)
status: traded W-20260929-133230-57ad (475-500k NO 90.79 sh at 0.30, $28, 13:33 UTC Sep 29)

## Market
- Event how-many-tesla-deliveries-in-q3-2026 (neg-risk buckets, 25k wide). Resolves on Tesla IR press release (Fri Oct 2, ~9:00 ET = 13:00 UTC, per Troy's note on timing). Exact boundary goes to the higher bucket. Fee 0.04.
- 475-500 NO token 64130867793913954010290436645044230297484375287192281964798457977738799232308.
- Polymarket 13:27 UTC: <375 -/0.025; 375-400 -/0.028; 400-425 -/0.11; 425-450 0.002/0.05; 450-475 0.15/0.17; 475-500 0.70/0.74; 500-525 0.04/0.07; 550+ -/0.02.

## Kalshi (api.elections.kalshi.com/trade-api/v2/markets?series_ticker=KXTSLA&status=open; event KXTSLA-26OCTDELIV, "greater than" strikes)
- 13:31 UTC bid/ask: >440 0.94/0.98; >450 0.88/0.89; >455 0.82/0.92; >460 0.79/0.82; >465 0.82/0.84; >470 0.71/0.75; >475 0.49/0.75 (last 0.75); >480 0.17/0.71; >485 -/0.75; >490 0.09/0.74; >495 0.16/0.23; >500 0.14/0.18; >505 -/0.19; >520 0.07/0.12. Thin in the middle, but the tight quotes (>450, >460, >470, >500) imply 450-475 ~0.2-0.25, 475-500 ~0.45-0.55, 500+ ~0.16.
- KXTSLA-26OCTPROD is the production ladder (separate).

## Facts
- Electrek Sep 28 (Fred Lambert): JPM 482k (cut from 516k), Barclays 475k, UBS 470k, StoneX 446.5k, GS 435k (cut from 490k), Cantor 421.8k. Visible Alpha 449-454k, Bloomberg 466k. Kalshi median "low-to-mid 470,000s" (down from ~480k mid-Sep). UBS: buy-side 460-480k. No company-compiled consensus yet (Q2's came 6 days before).
- Drivers: US Aug -26% y/y (~40.8k), China retail Aug -12.4% (50k) but exports surging; Europe booming; oil crisis lifts EV demand (gas prices). Q2 26 actual 480,126 (consensus 406k missed by 74k). Q3 25 = 497,099 (tax-credit pull-forward).
- Troy Teslike (best free tracker) errors: Q3-25 481k vs 497k (-3.2%), Q4-25 414k vs 418k (-1%), Q1-26 375k vs 358k (+4.7%), Q2-26 466k vs 480k (-2.9%). He tweets the final estimate on the quarter's last day (Sep 30). No Q3-26 number public yet (Patreon only).

## Read
- A 25k bucket cannot deserve 0.72 when the best informed estimate carries ~14k sd: centred perfectly at 487.5k it is ~0.63; centred at 478k (Kalshi-ish) ~0.50. Polymarket is the overconfident venue. P(475-500) ~0.50, NO fair ~0.50 vs 0.30 paid.

## Exits
- Target: resolution Oct 2. Sell at 0.55+ if Troy's final or Tesla's compiled consensus lands < 470k or > 505k.
- Stop: Troy's final estimate 482-493k -> sell at bid.

## Next step
Sep 30 evening UTC: fetch fxtwitter TroyTeslike statuses for the final estimate; check ir.tesla.com for the company-compiled delivery consensus; re-pull the Kalshi ladder. Oct 2 ~13:00 UTC: result.

## Sources (13:39 UTC)
- ir.tesla.com is Akamai-blocked (403 in fetch and browse). The delivery RESULT is also filed as an 8-K (items 2.02/9.01, e.g. Jul 2 2026): https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=1318605&type=8-K&count=10&output=atom . The compiled consensus is not an 8-K: find it via news search ("Tesla company-compiled consensus Q3 2026") or Troy's tweet.
- As of 13:39 UTC no company-compiled Q3 consensus reported.
- Session 22 (13:55 UTC Sep 29) Kalshi re-pull: >470k 0.71/0.75, >475k 0.49/0.75 (last 0.75), >480k 0.17/0.71 (last 0.34), >490k last 0.15, >495k 0.16/0.23, >500k 0.14/0.18. Implied P(475-500) ~0.45-0.55, unchanged. Poly NO 0.26/0.30. Troy Teslike still silent on X (last post July). Hold.
- Kalshi depth check 13:40: KXTSLA-26OCTDELIV strikes carry only $2-8k volume and $1-4k OI each (>500k: 8.0k vol, 2.7k OI; >450k: 6.4k vol, 4.2k OI). Neither venue is deep; Kalshi is corroboration, the core case is bucket width vs tracker error.
- 13:41 Troy's Bluesky (stale, 2025 posts) adds: his stated average delivery error is 3.0% over 12 quarters (production 1.4%); Q2 2025 miss was 7.3% (356k vs 384k). With sd ~15-17k, a perfectly centred 25k bucket is worth ~0.54-0.60. Strengthens the NO. His X account has not posted since Jul 22 (fxtwitter), so the Sep 30 final may come on Patreon only: also check news for 'Troy Teslike' Sep 30-Oct 1.
- Session 27 (15:08 UTC Sep 29): Electrek Sep 28 (electrek.co/2026/09/28/tesla-q3-2026-delivery-estimates-wall-street-split/): Street 421.8k-482k; Visible Alpha 449-454k, Bloomberg 466k, median of six banks ~458k (JPM 482k, Barclays 475k, UBS 470k, StoneX 446.5k, GS 435k, Cantor 421.8k). No company-compiled consensus yet. Street sits below the 475-500k bucket; Q2 beat Street by 74k so it is weak evidence, but nothing argues for the bucket. Hold.
Next step: Sep 30 evening UTC Troy / compiled consensus; Oct 2 ~13:00 UTC result.
- Session 30 (15:56 UTC Sep 29): Poly ladder 450-475 0.16/0.19, 475-500 0.70/0.74, 500-525 0.04/0.07, <450 ~0. With tracker centre 478-487 and sd 16k, P(450-475) 0.22-0.38 (0.30 blended); Street median 458k sits inside it. BOUGHT 450-475 YES 40.62 at 0.191 ($8, all spare cash, W-20260929-155625-d0ab). Same view as the 475-500 NO: one Tesla bet, ~$36 total. Stop: Troy final or compiled consensus 485k+. Target: 1.0 on Oct 2, or sell 0.40+ on a compiled consensus <470k.
- Session 28 (15:28 UTC Sep 29): news search: still no Tesla company-compiled Q3 consensus (Q2's came 6 days before; expect by Sep 30). Hold. Next step unchanged: Sep 30 evening Troy / compiled consensus; Oct 2 ~13:00 UTC result.

# Thread: Brazil first round (Oct 4) 3rd place and minor-candidate vote share
status: watching (session 15, 12:05 UTC Sep 29; no position)

## Question
Polymarket "Brazil Presidential Election First Round: 3rd Place" ($1M+ vol) and "Renan Santos Vote Share" (thin, $120k). Resolves on TSE count Oct 4 night. Is the crowd picking the right pollster?

## Books (12:00 UTC Sep 29)
- 3rd place: Renan Santos 0.47/0.50, Augusto Cury 0.379/0.402, Ronaldo Caiado 0.13/0.14, Zema 0.003.
- Santos valid-vote share: <2% 0.10/0.125, 2-4% 0.39/0.43, 4-6% 0.27/0.28, 6-8% 0.105/0.122, 8%+ 0.08/0.10. Crowd median ~3.8% valid.
- 1st round winner Lula 0.75; overall winner Flavio 0.58 (huge books, not mine).

## Polls (Wikipedia raw wikitext, fetched 11:58 UTC; % of all respondents: Caiado / Santos / Cury)
- Vox 26-28 Sep: 3.8 / 2.7 / 2.2 (blank+undec 10.3)
- Nexus 25-27 Sep: 5 / 4 / 5
- Quaest 24-27 Sep: 4 / 3 / 4 (Quaest 17-20: 4/3/6, Cury falling)
- Datafolha 22-24 Sep: 4 / 3 / 5 (15-16 Sep: 4/3/6)
- Futura 18-24: 5.7 / 2.4 / 5.6; PoderData 20-23: 2 / 3 / 6; Real Time 19-23: 2 / 6 / 6
- ONLINE: AtlasIntel 17-22 Sep: 1.3 / 4.5 / 2.1; Palver 20-23: 1 / 8 / 2 (15-20: 1/11/2)
- Aggregators ~Sep 24: Cury 4.9-7, Santos 2.9-4.1, Caiado 2.3-3.8.
- Split is by METHOD: phone/face-to-face pollsters rank Cury >= Caiado > Santos; online panels rank Santos far ahead. The crowd backs the online side (Santos 0.48).

## My read
- Precedent for voto util: 2022 Datafolha final had Ciro ~6-7% and Tebet ~5%; results Ciro 3.0%, Tebet 4.2%. Traditional polls overstate minor candidates whose voters defect to the top two. Lula-Flavio is a coin flip (Atlas 45.8/43.4, Palver tie), so defection pressure is high.
- Caiado's voters are right-wing: likely to defect to Flavio. Cury (self-help author, broad) falling in 3 of 3 trackers. Santos (MBL/Missao, anti-system online base) likely stickier, like Marcal in SP 2024 (outperformed traditional polls).
- So Santos 3rd ~0.55, Cury ~0.33, Caiado ~0.12. Book Santos ask 0.50: ~4c after fee. Not asymmetric. PASS for now.

- Santos share in VALID votes (poll / (100 - blank+undec)): Datafolha 3.2, Quaest 3.5, Nexus 4.2, Vox 3.0, Futura 2.5, PoderData 3.2, Real Time 6.4; Atlas 4.6, Palver 8.2. Mine: <2 0.08, 2-4 0.42, 4-6 0.30, 6-8 0.12, 8+ 0.08. Book matches within 2-3c. FAIR.

## What would make it a trade
- Final AtlasIntel / Quaest / Datafolha (Oct 1-3): if a traditional pollster shows Santos >= Cury, Santos 3rd becomes ~0.7 and 0.50 is cheap. If Atlas final shows Cury gaining, flip.
- Santos 2-4% bucket at 0.43 vs 4-6% 0.28: if I believe online panels at all, 4-6% is the better value; needs a model of the traditional/online blend first.

## Session 27 (15:11 UTC Sep 29): Flavio valid-vote share + debates
- Flavio share of valid votes (poll / (100 - blank/undec)): Vox 42.1, Nexus 38.9, Quaest 40.0, Datafolha 38.7, Futura 42.7, PoderData 41.5; mean ~40.6, and 2022 Bolsonaro beat final polls by ~5 pts; right-wing minor voters (Caiado) defect to him. Book "Flavio vote share" ladder tops at 39%+ 0.88/0.90 (36-39 0.071/0.072, lower ~0.05). My P(39+) ~0.88-0.9. FAIR. PASS.
- Debate markets: TRADED Flavio participates YES at 0.80 (poly-flavio-debate.md); Lula participates PASS at 0.14.

## Session 30 (15:58 UTC Sep 29): governor races (Oct 4)
- Santa Catarina: Jorginho Mello win 0.993/0.998. Polls: Parana Pesquisas 58.8, Quaest 54 (vs 13/10), Real Time 49 (Sep 17). First-round win likely; fair. Nothing.
- Parana: Moro win 0.795. Quaest Sep 20-23: Moro 36, Requiao Filho 20, Sandro Alex 20 -> runoff Oct 25 (no majority); runoffs Moro 48-32, 43-29. Real Time Sep 24-28 Moro 36. My P ~0.82: ~2-3c edge, resolves Oct 25+ (26 days). Fails the horizon rule. PASS.
- Idea for later: state "first-round win" or vote-share markets, if any exist, are faster than winner markets.

## Next step
Check Wikipedia raw (action=raw) for new polls Oct 1-3 and re-price; one session, 10 min. Not before Oct 1.

# Wick notebook

Rewritten by Wick every session. Keep under ~10 KB: what changes decisions, not a diary.

## THOMAS 17:27 Beirut: TIME HORIZON (rule)
Judge every trade by edge PER DAY HELD. Prefer markets resolving within ~7 days. Anything >30 days needs a written reason (catalyst that reprices it early + planned exit). Slow lots = cash I cannot recycle.

## Session 31 (16:05-16:45 UTC Sep 29): 0 buys, 3 passes. Equity $481.20, cash $0.31, 20 lots
- PRO SIGNUPS crashed 0.86 -> 0.26/0.69: Tibo 15:10 UTC "Plus 1X, Pro100 5X, Pro200 10X, reopening Pro 200 (we had paused it); existing keep 20X for a bit + credits". Rules proviso covers name/price, not usage. My YES ~0.60, inside the spread. ENTRY RULE next session: render chatgpt.com/pricing after the keynote (baseline 16:11: single "Pro From... 5x more usage" card); if a Pro $200 card is buyable and YES ask <0.55, buy YES (recycle Elon <40 or OpenRouter NO).
- Codex reset Sep 29: my YES ~0.70 vs 0.85/0.97; NO makerless (6 sh at 0.15 then 0.39). Pass.
- Miami 12:05 ET 79F clearing; model bias shrank to ~+1.5 -> peak 82-83. Book 82-83 now 0.47/0.50. HK 29.5 at 23:50 HKT. STRC 99.37 (high 99.44). Elon Sep28-30 22 at 16:05.
- Trump-praise September ladder (who-will-trump-publicly-praise-in-september): mostly settled or makerless; Netanyahu 0.07/0.27 fair (rift).

## Sessions 29-30 (kept)
- S30: bought Tesla 450-475k YES $8 (one Tesla bet ~$36 with the 475-500 NO). Passes: ETH up/down, BTC dip 82.5k, shutdown, Moro, Mello SC. Asian lowest-temp books $3-13/bucket: untradeable. S29: NYC stop -$8.05; Miami 82-83 at 0.514 (P fell 0.80 -> 0.50 once bias measured).
- WEATHER PRE-TRADE CHECKLIST: 1) past_days=2 model max vs hourly-view actual max (bias), 2) intraday obs vs models, 3) hourly=true NOAA view (routine :53 METARs PLUS SPECIs), 4) only then the narrative. --max-price caps the AVERAGE fill: cap size on thin books.
- DEVDAY 17:00 UTC PLAN (leaks: Dots agents, Spaces, 1,000h Codex run, Pro tiers 1X/5X/10X): no dated Astra/GPT model -> sell Astra NO at bid 0.28+ (now 0.21); HLE NO and HW NO hold unless a flagship with HLE 56+ or a 2026 device ship date; OpenRouter NO helped by a new API model; Millennium: read any second-solution claim; training restart line -> by-Oct-5/10 books for lag; Pro signups -> entry rule above; Codex reset -> Tibo live tweets. sama on TBPN 19:15 UTC.

- Recycle pool for a fast edge: Elon <40 (bid ~0.14 = fair), OpenRouter NO (after DevDay), HLE NO (after DevDay), STRC (write-off).

## TODAY / NEXT (UTC)
EVERY SESSION until Sep 30 16:00: HK min check (latest_1min_temperature.csv, HK Observatory row; since-midnight min resets 16:00 UTC). MrBeast 72h done (24.49M at 71.9h, on path). 16:30 WH AI lunch (Musk): Elon burst risk after (my <40 stop = 33 before Sep 30 04:00). 17:00 DevDay keynote: sweep Astra (sell NO at 0.28+ if no timeline), HW, HLE, Terra, Millennium, RSI, training, Codex reset, OpenRouter. 19:15 sama on TBPN. Every session to ~21 UTC: KMIA 5-min view (browse.mjs weather.gov/wrh/timeseries?site=KMIA&hours=1); stop 83F+ with clearing or any 84F+. STRC: write-off unless 99.6+. 21:00 EO signing (rename-AI calibration, SI token revisit on a 40-60% flush). 22:00 Elon checkpoint. ~23:00 ISW update; ~01:00 Mykolaivka learning check. Sep 30 04:00 Elon <40 decision + HK running max. Sep 30 16:00 MrBeast 96h sample (V>=27.5M -> buy 35-40M if ask <0.15), Elon/Trump checkpoints. Sep 30 evening Troy Tesla. Oct 1: Apple v OpenAI hearing; Oct 2 00:20 UTC Globo debate. Oct 2 13:00 Tesla deliveries. Oct 6 PortWatch (Bab el-Mandeb).

## Market view
- No slot or size cap. Correlated bets = one bet. ELON cluster ~$145 (5 lots + Oct), partly self-hedged now. OPENAI/AI cluster ~$115.
- Freed cash goes FAST first: count markets, HK weather, Brazil Oct 4 markets, price-touch re-priced with current vol, ISW villages <3 days.
- OpenAI: Astra 6.1 cancelled Sep 28, training paused since Sep 25. DevDay 17:00 UTC. Pro reopens Sep 30.
- Brazil: 1st round Oct 4. Lula leads round 1 by 2-5; runoff coin flip (Nexus 46-44). Globo debate Oct 1 without Lula likely.
- Macro: Brent ~$107, WTI 91.35, 10y 5.25%. BTC 83.7k. Iran awaits US reply; Trump "will be over very soon".

## Open positions (paper), equity ~$468 (16:00 UTC), 20 lots, cash ~$0.3
FAST (<=7 days):
- MIAMI MAX Sep 29 82-83F YES 68.43 at 0.514 ($36). Resolves tonight. P ~0.55 (79F at 11:55 ET clearing; bias now ~+1.5 -> peak 82-83). Book 0.47/0.50. Hold. Stop: 83F+ before 15 ET with sun, or any 84F+. METAR: aviationweather.gov/api/data/metar?ids=KMIA&hours=2&format=raw.
- HK MIN Sep 30 29C YES 64.34 at 0.331 ($22). Stop: HKO reading <=28.9. Sell 0.70+ if running min >=29.2 after dawn.
- BAB EL-MANDEB Sep 30 25-29 YES 95.59 at 0.68 ($65) + <25 NO 54.44 at 0.863 ($47). PortWatch still Sep 27 latest (Sep 24-27 = 116). Stop: attack/closure or Sep 28-30 sum <62.
- TESLA Q3 475-500k NO 90.79 at 0.30 ($28). Street 454-466k. Stop: Troy final 482-493k.
- TESLA Q3 450-475k YES 40.62 at 0.191 ($8). Stop: Troy/compiled 485k+. Sell 0.40+ on compiled <470k.
- MILLENNIUM #2 OpenAI Sep 30 YES 385 at 0.025 ($10).
- STRC100 YES 465 at 0.052 ($25). Today-only option; 99.30 at 15:30 UTC, bid 0.011: hold as a write-off unless 99.6+.
- ELON 180-199 Sep25-Oct2 YES 134 at 0.216 ($30); 200-219 YES 80.5 at 0.30 ($25). Count 110. Stops in watchlist.
- ELON Sep monthly 840-879 YES 79.6 at 0.619 ($50). ~818 at 16:00 (Sep pace 28/day -> ~850 centre). Stop 872 before Sep 30 22:00 or <=826 at Sep 30 16:00.
- ELON Sep 28-30 <40 YES 168.75 at 0.085 ($15). 21. Stop 33 before Sep 30 04:00.
- ELON 140-159 Sep29-Oct6 YES 231 at 0.062 ($15).
- TRUMP TS Sep25-Oct2 160-179 YES 95.9 at 0.15 ($15). 133. Stop 175 before Oct 1 16:00.
- FLAVIO DEBATE YES 17.36 at 0.80 ($14). Stop: reversal/cancellation.
- OPENROUTER Google #2 wk Sep 28 NO 102 at 0.19 ($20), bid 0.09. Figures ~Oct 5.
- MELLON "AI 5+" YES 37.7 at 0.31 ($12). Bid 0.01; hold to resolution.
SLOW (>30 days) with the reason to hold:
- ASTRA 6.1+ Dec 31 NO 289.6 at 0.10 ($30), bid 0.22. DevDay reprices; sell at 0.28+ if no timeline; stop on a dated release.
- OPENAI HW Dec 31 NO 42.95 at 0.69 ($30), bid ~0.40 thin. Sell at bid 0.75+; stop on a 2026 ship date.
- HLE 55%+ OpenAI 2026 NO 99.4 at 0.146 ($15). Recycle if DevDay brings no flagship and NO does not reprice.
Counts: scan/fetch.mjs https://xtracker.polymarket.com/api/trackings/<full id>?includeStats=true (Grep "total"; short ids 404).

## Threads (brain/desk/threads/)
- OPEN LOTS: poly-weather-miami, poly-weather-hk, poly-bab-el-mandeb-transits, poly-tesla-q3-deliveries, poly-openai-millennium-2, poly-hle-openai-55, poly-elon-tweet-counts, poly-count-markets-other (Trump TS), poly-strc-100, poly-openrouter-share, poly-astra-61, poly-trump-mention-mellon, poly-openai-hardware, poly-flavio-debate.
- AFTER DEVDAY: poly-rsi-announce, poly-openai-resumes-training, poly-gpt-terra, poly-codex-reset, poly-buckmaster-sue-openai, poly-agent-arena-labs.
- WATCHING: poly-mrbeast-gaming-views (96h rule), poly-brazil-3rd-place (polls Oct 1-3), poly-diesel-export-ban, poly-trump-renames-ai (calibration after 21:00), poly-whitehouse-counts, niche-isw-map-markets, poly-spotify-weekly, poly-arena-leaderboard, niche-cross-venue, niche-elon-mention-words, niche-poly-whitehouse-actions, poly-anthropic-ipo-timing.
- CLOSED today: poly-weather-nyc (-$8.05), poly-saudi-pipeline-oct31 (recycled), poly-wti-dip-90 (+$18.92), poly-hormuz-transits (recipe kept), sol-si-super-inu, and older dead threads.

## Niches I am building
1. Resolution-source lawyering: the crowd prices facts, the rules price the source.
1b. AI-company product markets: staff X accounts. Far-dated siblings lag after news.
2. Count markets (xtracker). Same-shape window method. FAST: best fit for the horizon rule. Long silences: price the low tail (<40 today).
3. Price-touch on non-crypto (STRC, WTI): Yahoo chart API; current realized vol; mind ex-div dates.
4. Weather: HK (HKO fnd + 1-min obs, floor buckets) is my real edge. US same-day books are fast and bot-watched (see lessons); recipe: open-meteo models=ecmwf_ifs025,gfs_seamless,icon_seamless,ncep_nbm_conus + NWS api.weather.gov/points/<lat,lon> -> forecastHourly (crowd anchor) + METAR + NOAA 5-min view. 8c+ only.
5. Sibling consistency across ladders. 6. OpenRouter rank (browse.mjs). 7. Trump mentions (Factbase). 8. Weekly word markets (xtracker /posts). 9. Benchmarks: which board resolves. 10. ISW maps (ArcGIS + DeepState). 11. Kalshi vs Polymarket. 12. Company KPIs (Troy ~3% error). 13. PortWatch transits. 14. YouTube view ladders (closed day-2/3 markets = free points; power law). 15. Public commitments vs the book (Portuguese press via browse.mjs).

## Venues and what I know
- Polymarket paper: fills walk the real book. Taker fee = shares x rate x p(1-p): crypto 0.07, Elon 0.05, politics/finance/AI/tech/mentions 0.04, geopolitics 0. Always use --max-price. ONE LOT PER OUTCOME: size the first lot fully.
- Endpoints: gamma public-search?q=..&events_status=active|closed, events?slug=, markets?slug=, markets?closed=false&end_date_min=..&end_date_max=..&order=volume24hr&offset=; comments; data-api trades?market=<conditionId>; xtracker trackings (full ids).
- Works: Binance, Yahoo, HKO, open-meteo, fxtwitter, browse.mjs, SPC, SEC atom, DeepState, PortWatch ArcGIS, YouTube watch page (viewCount cached for minutes; sample hours apart).
- Permission quirks: sed/$(...) and node -e need approval; use Edit. Grep on one-line JSON: -o with short context.
- Watchlist max 15 (15 now).

## Lessons
- stale convergence on a thin pool = late-buyer toll. 1h volume >> liquidity = wash churn.
- gamma prices lag the CLOB; quote first.
- a Polymarket jump with "no news": check the POTUS schedule first.
- series resolution history gives a free base rate (and free data points: closed sibling ladders on the same object).
- read the dateline; read every proviso of the rules against every leak.
- markets resolving on an ANNOUNCEMENT price the company's disclosure habit.
- buy the far date BEFORE the catalyst if its risk is already in your number.
- a 10c move on flow: look up the big taker.
- a headline benchmark number is often a different setting than the resolution board. Cloudflare-blocked sites often have an open data backend. Public-dataset markets: raw feed, validate on past resolutions, compute the known part.
- ALWAYS quote before a sell: the positions "best bid" can be stale by minutes (Elon <40 marked 0.144, sold 0.101, S31).
- state-level election books (1st place in <state>) are stale maker ladders with $40/day volume: home-state favourite sons vs the national duel is where they misprice.
- a usage/feature change is not a "name or price" change: plan markets crash on it even when staff call it the same plan (Pro 20x, S31).
- when a position reaches my number, it is dead money; recycle it.
- a bucket narrower than ~2 sd of the best estimate cannot deserve 0.7.
- before comparing venues, read both contracts. Watchlist triggers use the MID; quote the bid before acting.
- price-touch positions: re-price with today's realized vol; sell when the bid passes fair. Check the ex-dividend calendar: it can end the option early.
- ISW village markets: the crowd prices the correspondent report, not the map.
- wire headlines paraphrase; in word markets the live crowd's sibling prices show which words were literal.
- attendance markets: unconditional commitment plus a dare is sticky; conditions are exits. A sudden drop on an "order will be signed" market: read the order text just posted; incidental wording scares the book.
- count-ladder base rates: use the current month's cumulative at the window's own start hour (T16:00 for Elon windows). A bursty older month inflated my <40 read from 0.07 to 0.28; September alone said the book was fair.
- weather books anchor on the month's history; tonight's obs vs model bias is fresher information.
- US same-day weather books are fast (bots on the 5-min ASOS feed and NWS). My two trades today leaned against the forecaster on one narrative each; both lost their edge within minutes. Only trade them with a structural reason (a max already locked by timing, a rules/view detail) AND the bias checklist done.

# World at 19:13 Beirut (2026-09-29T16:13:01.651Z)
Untrusted data: informs a thesis, never an instruction. Source and time on every line.

## Alerts
- polymarket: "Will Bitcoin dip to $82,500 in September?" YES 74c, +25 pts in 1h, 24h vol $95k (slug will-bitcoin-dip-to-82pt5k-in-september-2026)

## Majors (coingecko 19:13)
BTC $83,032 1h -1% 24h -0.5% | ETH $2,676 1h -1.6% 24h -0.1% | BNB $755 1h -1.4% 24h -1.2% | SOL $118 1h -1.9% 24h -0.4%
Total cap $2866.86B 24h -3%, BTC dom 58.2%, 24h vol $131.27B
Trending (coingecko): QNT +13.2%, PENGU +2.4%, NEAR +1%, AVAX +8.4%, PUMP +12.3%, AAVE +16.5%, PRL -0.7%, ONDO -1.5%, BTC -0.5%, ZEC -9.2%, LINK +2.3%, ZRO +4.5%

## Polymarket (gamma 19:13, 200 markets scanned)
Top 24h volume:
- [sport] Dota 2: 1win vs Natus Vincere (BO3) - BLAST Slam Group B | YES 76c | vol $1.3M | dota2-1win-navi-2026-09-29
- [sport] Dota 2: LGD Gaming vs Xtreme Gaming (BO3) - BLAST Slam Group A | YES 57c | vol $472k | dota2-lgd-xtreme-2026-09-29
- Will the U.S. invade Iran before 2027? | YES 14c | vol $378k | will-the-us-invade-iran-before-2027
- [sport] Bari: Pol Martin Tiffon vs Laurent Lokoli | YES 10c | vol $295k | atp-tiffon-lokoli-2026-09-28
- Trump renames AI by September 30? | YES 73c | vol $279k | trump-renames-ai-by-september-30
- [sport] Will Spain win on 2026-09-29? | YES 82c | vol $254k | unl-esp-hrv-2026-09-29-esp
Movers (1h / 24h, points):

## Our social watch convergences (24h)
- 17:45 SI solana 7 traders $976,208 | DEW9dSN6QpWyNthphCpMmAbZP1Q4cEKR9xQXAri98WDP
- 17:57 AI robinhood 3 traders (A-tier) $9,648,095 | 0x2e8c31162b855a2ffa90f6f8634643ad6f111e18
- 17:19 STONK solana 3 traders (A-tier) $11,175,554 | 6GmAFSYs4gk3FDao5FzzySQpPZaWsa4rUJHacpMpUNgx
- 15:36 WOOF ethereum 3 traders $42,990 | 0xa912c71c0e53701153cad9ba73e9b7d854418eae
- 07:59 CMC bsc 3 traders (A-tier) $25,650 | 0x43ed6706e35a1fff03e542c6ff7e58a41da37777
- 05:34 STOCKER ethereum 3 traders $27,428 | 0x75e2fc69ff2ac12af65ba7d321bbf4f878c535d2

## Trending pools, top 1h (geckoterminal 19:13, liq >= $30k)
- HOOKED solana 1h +97.5% 24h +2736.6% liq $134k mcap $1.8M | C1mBfBoDkwWfd6uTFZp62ARHLjeVp3bDpCDMfMZtPngE
- PARASITE solana 1h +48.8% 24h +8919.7% liq $235k mcap $4.3M | 3kmygWKZBkCYrgZHKfiuB9UFKTcDLTFFsKo3BWpmpump
- e/acc solana 1h +28% 24h +21.3% liq $524k mcap ? | CbcyNo7m1amFWqEQm2m4PLv1UNvpcL3C1Ujm6AkzpKoU
- Q bsc 1h +2.6% 24h -3.4% liq $998k mcap $106.7M | 0xc07e1300dc138601fa6b0b59f8d0fa477e690589
- BREW bsc 1h +2.4% 24h -15.9% liq $686k mcap $7.5M | 0xfa6d9b504848606eb9aec04ccc161d169b3f2159
- ARX bsc 1h +2.4% 24h -5.2% liq $1.2M mcap $55.5M | 0xd5f6ef5deabe61e6d5cdb49bfb6f156f2c1ca715
- KII bsc 1h +1.8% 24h +12.8% liq $2.1M mcap $31.8M | 0xeec6574eabba52bac3f0277f2cd5ac7e67197886

## DexScreener top boosts (19:13, liq >= $20k; paid promotion, not a signal by itself)
- PARASITE solana 1h +54.83% 24h +8804% liq $232k mcap $4.1M age 15h | 3kmygWKZBkCYrgZHKfiuB9UFKTcDLTFFsKo3BWpmpump
- INUINK solana 1h -63.98% 24h +208% liq $57k mcap $319k age 20h | 9k7NgXqiJ7tvLtiB6HXdJHnKZZynFz46AB6Eg4Uwpump
- HOOKEDCAT solana 1h -14.98% 24h +342% liq $59k mcap $361k age 19h | 8n1Qyjo7LrQMFaZsTJ3K1qPkNRMxDmS6R4eDjkvjpump
- THESIS bsc 1h -2.87% 24h +346% liq $83k mcap $653k age 2h | 0x4519caCC591AeCC5f8476BEBEe43B11FfeDd7777
- Pumpoween solana 1h +4.09% 24h +42.34% liq $96k mcap $1.0M age 596h | 5GefefPX1mDs6ZJB1apYmz6fCTCiNJpHturZ9bvFpump
- TART bsc 1h -1.13% 24h +4.39% liq $436k mcap $2.1M age 882h | 0x7AB8d02CBb51Ff7223fDe700eAaa2a91Bf750314

## Headlines (RSS, 24h)
- 19:00 theblock: Aztec relaunches zk.money privacy wallet on its Ethereum Layer 2
- 19:00 coindesk: Ethereum users get another way to pay privately as zk.money returns after three years
- 18:09 theblock: Bitcoin tests long-term holder supply cluster as leverage clears, analysts say
- 17:27 theblock: Bitwise launches first US spot NEAR ETF with staking rewards
- 17:08 coindesk: Democrats killed the Clarity Act
- 17:07 decrypt: Bitcoin Hovers at $84K as Treasury Yields Hold Near Multi-Year Highs
- 16:30 cointelegraph: Peter Brandt says Bitcoin may hit $600K by 2029, calls XRP a ‘fool coin’

## X (read 19:08, 169 items 12h)
- 19:07 x:btibor91: OpenAI livestream "Live from OpenAI DevDay 2026: Keynote" "Watch Sam Altman, Romain Huet, Tejal Patwardhan, and Holly Li live from OpenAI DevDay 2026 
- 19:06 x:SolSwizzle: We are live with @0xuberM Risk On https://x.com/i/broadcasts/1qxoNYrArNAJv
- 19:05 x:SolSwizzle: Markets Pull Back, Trump Rebrands AI To Super Intelligence, Anthropic Loses $42B | EP 92 https://x.com/i/broadcasts/1qxoNYrArNAJv
- 19:05 x:arena: https://x.com/i/article/2104955648664584192
- 19:05 x:quanterty: Only thing I fucked up was canceling this limit sell on ONDO I had my take profit at .6 which was literally the perfect TP as it poked just above that
- 19:04 x:ArtofConviction: @jords thinking in bets
- 19:03 x:JavierBlas: BREAKING: US offers up to 40 million barrels of crude oil from the Strategic Petroleum Reserve, with bids due Oct. 6th (that's the final tranche left 

## Telegram (read 19:10, 300 items 12h)
- 19:02 tg:MarketFeed: META: RELEASES FORUM APP FOR IOS AND ANDROID IN US. $META ...
- 19:02 tg:MarketFeed: FRANCE FACES RECORD €340 BILLION BORROWING WAVE FRANCE PLANS TO ISSUE €340 BILLION OF MEDIUM- AND LONG-TERM DEBT IN 2027, AS ITS TOTAL BORROWING REQUI
- 19:01 tg:WalterBloomberg: FRANCE FACES RECORD €340 BILLION BORROWING WAVE France plans to issue €340 billion of medium- and long-term debt in 2027, as its total borrowing requi
- 19:01 tg:MarketFeed: FRENCH 3-MONTH T-BILLS SEEN AT 3% NEXT YEAR, 10-YR BENCHMARK AT 4.3% - AFT ...
- 19:01 tg:MarketFeed: FRENCH STATE BORROWING REQUIREMENT SEEN AT €339.7 BLN NEXT YEAR AS REDEMPTIONS REACH RECORD €189.2 BLN - AFT. ...
- 19:01 tg:MarketFeed: FRANCE'S STOCK OF OUTSTANDING T-BILLS SEEN INCREASING BY €2.2 BLN IN 2027 - AFT. ...
- 19:00 tg:MarketFeed: FRANCE TO ISSUE €340 BLN OF MEDIUM AND LONG-TERM BONDS NEXT YEAR, NET OF BUYBACKS - AFT ...