Keel v4 · report date 2026-09-14 · category focus: Politics · horizon 45 days · capital $2,500
Kalshi Politics — Mispricing Audit (v4, archetype-targeted)
One pick, fifteen logged rejects, 94% cash. Stage 0 scanned 902 open markets across the 120 allowlisted series and left 10 mechanical survivors in five clusters. Of those five, only the Suez weekly transit ladder offered a deployable leg outside the market's own spread and inside the 40–84¢ band; the FOMC dissent ladder, Anthropic's October 24 IPO rung, the Bank of Canada October hold and the Hegseth October rung all failed a band, spread or edge test and are logged with the side and probability I would have taken.
Method notes for this run
- Universe.
targets.json(generated 2026-09-14T16:07Z, 120 series, 75 Tier 1 / 45 Tier 2). Live public-API quotes were used for every cut; 823 of 902 open markets close beyond the 45-day horizon, 813 had under 500 contracts of 24h volume, 71 close after their expected expiration. - Probability before price. Pre-market probabilities were written to a file at about 16:12Z (container clock) before any orderbook or candlestick was fetched: Suez T270 YES 90, T260 YES 95, FOMC 0-dissent 42, Anthropic by Oct 24 YES 27, BoC October hold 50, Hegseth by Oct 1 YES 4.
- Classifier note.
KXSUEZWEEKLYis filed as archetype legal-action in the allowlist; it is a scheduled-process ladder identical in mechanics toKXHORMUZWEEKLY(85 precedent rows). The archetype string is reported as filed; the fix belongs in the classifier. - Continuous-number screen. The one pick is a count-threshold market. It is kept because the thesis is not a trend extrapolation toward a nearby line: the line sits ~4 standard deviations below a stationary ten-week series and the resolver mechanics were verified on two settled weeks this run. This is the documented tension in v4 (transit counts are named as a winning archetype; count-pace theses are named as a reject) and the shadow record should adjudicate it. Conviction is capped at MEDIUM for that reason.
Picks
KXSUEZWEEKLY-26SEP20-T270
More than 270 Suez Canal transit calls (IMF PortWatch), Sep 14-20, 2026?
Thesis. Ten straight PortWatch weeks printed 286-301; the 270 line sits ~4 sd below a stationary series, yet the book charges 22% for a miss.
Rationale. Prob: 90% pre-market -> 86% final (blended toward a 2.5k-contract book). IMF PortWatch daily Suez counts (data through Sep 6, feed edited Sep 9) sum to 295/289/291/286/291/287/286/294/301/291 for the last ten Mon-Sun weeks; the lowest week clears 270 by 16 and the weekly sd is ~4.6, and totals held through the Aug 11 Tihamah and Aug 24 Yanbu attacks. Deductions: PortWatch data gap or late revision -3, Bab el Mandeb decline (302->202 traceable, Lloyd's List Sep 3) propagating north -3, canal blockage/security shock -4. Precedent: KXHORMUZWEEKLY-26JUL19 ladder - PortWatch is the only count that settles, price rungs off the official series' gap to the threshold, not headlines. Cleanest tail: a multi-day canal closure or a Houthi escalation that halts Red Sea traffic mid-week.
PortWatch weekly Suez totals, last ten Mon–Sun weeks (red line = 270)
Source: IMF PortWatch Daily_Chokepoints_Data, Suez Canal, data through Sep 6 (layer last edited Sep 9 12:34Z). Daily counts range 27–57; weekly sums range 286–301.
Resolver profile. Resolver: Kalshi, mechanically off the IMF PortWatch daily series summed Mon–Sun; rules_secondary says settlement comes "no earlier than Tuesday 9 AM … once complete data is available". Verified this run: the Aug 30 week settled Wed Sep 2 15:07Z (T300 YES on a 301 print) and the Sep 6 week settled Wed Sep 9 14:48Z (T290 YES / T300 NO on a 291 print), each within about two hours of the PortWatch feed update. The resolver reads its own rule literally and waits for the feed; a feed lag is a timing risk, not a directional one. No dated clarification in rules_secondary.
Risk deductions. PortWatch data gap or late revision −3; Bab el Mandeb decline (302→202 traceable transits, Lloyd's List Sep 3) propagating north −3; multi-day canal blockage or security shock −4. Total −10 from a stationarity prior of ~100, before blending toward the book.
Precedents consulted. KXHORMUZWEEKLY-26JUL19-T60/T75 (PortWatch is the only count that settles; rungs priced off tracker headlines are wrong); KXHORMUZNORM-26MAR17-B260715 (price threshold rungs off the gap between the live series and the line, not headlines); KXSUEZWEEKLY-26AUG30-T300 and -26SEP06-T290 (verified this run and written back: settlement on PortWatch, Wednesday timing, knife-edge rung near the mean is a coin flip).
Cleanest way to be wrong. A Houthi escalation or canal incident that halts Red Sea traffic for two or more days inside Sep 14–20; the ten-week sample contains two attack weeks but no closure.
Position sizes and limits
| Ticker | Side | Limit | Contracts | Cost (max loss) | Max win | Cap binding |
|---|---|---|---|---|---|---|
KXSUEZWEEKLY-26SEP20-T270 | YES | 78¢ | 184 | $143.52 | $40.48 | Depth cap: 553 visible at 78¢ → 1/3 = 184 (MEDIUM 7% cap would allow 224) |
| Cash | — | — | — | $2,356.48 | — | Reserve ≥ 10% satisfied (94.3% cash) |
EV for the record
| Ticker | p (final) | Price | EV / contract | EV % | Expected $ |
|---|---|---|---|---|---|
KXSUEZWEEKLY-26SEP20-T270 | 0.86 | 78¢ | +8.0¢ | +10.3% | +$14.72 |
Sizing is conviction-based and EV-agnostic; EV is recorded for calibration only.
Cluster table
| Cluster | Deployed leg | Runner-ups logged | Cost |
|---|---|---|---|
| suez-weekly-sep20 | KXSUEZWEEKLY-26SEP20-T270 YES | T260 (cluster runner-up) | $143.52 |
| fomc-sep-dissent-count | — | 0 (tail band), 1 (inside spread), 2 (thin edge / 48h move), 3 (inside spread), 4 (deep favorite) | $0 |
| anthropic-ipo-timing | — | OCT24 (edge ≤ slippage) | $0 |
| boc-oct28-decision | — | H0 (tail band on NO), H25 (spread > 5¢) | $0 |
| hegseth-departure | — | OCT01 (deep favorite), DEC01 (close after expiration) | $0 |
Conviction exposure
| Conviction | Picks | Cost | % of capital | Cap |
|---|---|---|---|---|
| HIGH | 0 | $0 | 0% | 15% per pick |
| MEDIUM | 1 | $143.52 | 5.7% | 7% per pick |
| Cash | — | $2,356.48 | 94.3% | — |
Risk profile
- Worst case (pick loses): −$143.52, 5.7% of capital.
- Best case: +$40.48.
- Correlation: single cluster, no shared thesis with the existing live pick (KXCRYPTOSTRUCTURE-26JAN-OCT NO).
- Timing: window Sep 14–20; PortWatch data typically lands 3–8 days later; expect settlement Wed Sep 23 or later, after the nominal Sep 22 expiration.
Execution notes
- Limit 78¢ YES, 184 contracts. Visible at 78¢: 553 (NO bids at 22¢). Do not chase to 79¢; the edge at 80¢ is under 6¢ and the thesis does not improve with price.
- If PortWatch prints Sep 7–13 (visible around Sep 16) below 280, cut the position: that would be the first sub-286 week in the sample and would invalidate the stationarity prior.
- The T260 leg at 86¢ is the fallback if T270 fills are unavailable; it is a half-size band (3% cap for MEDIUM) and the runner-up, not an add.
Rejected (logged, shadow-tracked)
| Market | Archetype | Side | Price | Model % | Reason |
|---|---|---|---|---|---|
KXSUEZWEEKLY-26SEP20-T260More than 260 Suez transit calls, Sep 14-20, 2026? | legal-action | YES | 86¢ | 93 | cluster runner-up: same thesis as T270 with less edge (7c) in the 85-89c half-size band; one leg per cluster |
KXFOMCDISSENTCOUNT-26SEP-0Exactly 0 dissents at the Sep 16, 2026 FOMC? | multi-news-other | YES | 31¢ | 40 | <40c tail band. Pre-market 42% (FedWatch ~85% hike on Sep 12; the July dissenters Hammack/Kashkari/Logan get their hike, no dove has dissented since April). Book already repriced 12c->31c on Sep 11 CPI; YES at 31 fails the band, NO at 70 is negative edge |
KXFOMCDISSENTCOUNT-26SEP-2Exactly 2 dissents at the Sep 16, 2026 FOMC? | multi-news-other | NO | 78¢ | 17 | edge 5c rests on a subjective hike-with-two-doves decomposition; whole ladder moved >10c in the last 72h on public CPI news (48h price-move check); specialist-dense ladder (16k vol24 on sibling) |
KXFOMCDISSENTCOUNT-26SEP-3Exactly 3 dissents at the Sep 16, 2026 FOMC? | multi-news-other | NO | 85¢ | 13 | inside spread: NO ask 85 vs model 87; 3-dissent is the hold-scenario base case (15% hold odds) and the 2c edge is below slippage |
KXFOMCDISSENTCOUNT-26SEP-1Exactly 1 dissent at the Sep 16, 2026 FOMC? | multi-news-other | NO | 75¢ | 27 | inside spread: model 27% matches the 25-27c book; no edge |
KXFOMCDISSENTCOUNT-26SEP-4Exactly 4 dissents at the Sep 16, 2026 FOMC? | multi-news-other | NO | 93¢ | 3 | deep-favorite band: NO costs 93c |
KXIPOANTHROPIC-DATE-26OCT24Anthropic confirms IPO (S-1 effective / priced / ticker assigned) before Oct 24, 2026? | corporate-event | NO | 67¢ | 30 | edge 3c <= slippage on public-info-only thesis. Pre-market 27% YES: no public S-1 on EDGAR as of Sep 14, Reuters (Sep 4) puts the public prospectus in late September and marketing mid-October at the earliest with a listing days before Nov 3; the 15-day rule plus a roadshow makes pricing by Oct 23 a best-case path. Book already fell 57c->33c on Sep 11-12. Precedent: KXIPOANTHROPIC-DATE-26JUL01 / KXIPO-26-SPACEX - only effectiveness, pricing or formal exchange assignment counts |
KXCBDECISIONCANADA-26OCT-H0Bank of Canada holds at 2.25% on Oct 28, 2026? | scheduled-process | NO | 33¢ | 50 | <40c tail band. Pre-market 50% hold vs CORRA swaps at 36% hold / 64% hike (Sep 14) after Macklem's Sep 2 'multiple increases' warning and NBF/Scotia October-hike calls; Kalshi hold leg is 30+ pts too rich vs OIS but the only mispriced side is a sub-40c NO. Aug CPI on Sep 15 is the next catalyst. Precedent: KXCBDECISIONCANADA-26JUL-H0 - hold leg is carry, not a view |
KXCBDECISIONCANADA-26OCT-H25Bank of Canada hikes 25bp on Oct 28, 2026? | scheduled-process | YES | 34¢ | 48 | Stage-0: bid-ask spread 6c (28/34) exceeds 5c; also <40c tail band |
KXHEGSETHOUT-26AUG-26OCT01Hegseth out as Secretary of Defense before Oct 1, 2026? | multi-news-other | NO | 96¢ | 4 | deep-favorite band: NO costs 96c (YES at 6c is a <40c tail). Trump 'fantastic job' after Driscoll/Tillis (Sep 1-2); precedent KXHEGSETHOUT-26APR-SEP01 - criticism volume is not a White House signal |
KXTARIFFBILL-26JUL-26NOV01New/increased tariff legislation becomes law before Nov 1, 2026? | confirmation/enact | — | 45¢ | — | Stage-0: closes 48 days out, beyond the 45-day horizon (not researched) |
KXKASHOUT-26JUL-26NOV01Kash Patel out as FBI Director before Nov 1, 2026? | departure/ouster | — | 4¢ | — | Stage-0: beyond 45-day horizon and YES ask <= 5c |
KXHOUSEADJOURN-26-26SEP26House leaves for pre-election recess by Sep 26, 2026? | departure/ouster | — | 99¢ | — | Stage-0: YES bid 98 >= 95c (deep favorite) |
KXHEGSETHOUT-26AUG-26DEC01Hegseth out as Secretary of Defense before Dec 1, 2026? | multi-news-other | — | 28¢ | — | Stage-0: close after expiration (close_time Dec 1 vs expected_expiration Oct 1) and beyond horizon |
KXCBDECISIONJAPAN-26SEP17-H25Bank of Japan hikes 25bp on Sep 17, 2026? | scheduled-process | — | 98¢ | — | Stage-0: YES bid 97 >= 95c; nothing left to earn |
Diligence notes by cluster
FOMC dissent count (Sep 16)
CME FedWatch had a 25 bp hike at roughly 85% on Sep 12 (up from ~40% on Aug 21 and 57% after Jackson Hole). July's 9–3 hold had Hammack, Kashkari and Logan dissenting for a hike (Fed statement Jul 29; minutes: "several participants favored an increase"); no member has dissented for a cut since Miran in April, and Waller said before the blackout that the CPI report would decide his support for a hike (PBS). Under a hike the hawkish bloc is satisfied and a unanimous vote is plausible, so my pre-market 42% for zero dissents sits above the 31¢ book, but YES at 31¢ is a tail-band buy and the whole ladder already repriced on Sep 11. No deployable leg.
Anthropic IPO by Oct 24
EDGAR full-text search on Sep 14 shows no public Anthropic S-1. Reuters (via CNBC, Sep 5) moved the public prospectus to late September, marketing to mid-October at the earliest and the listing to days before the Nov 3 midterms, pending a $15bn revolver (Bloomberg Sep 3). Bloomberg reported Nasdaq as the venue (Sep 13) and Nvidia weighing a $10bn anchor (Sep 11); neither is a resolution trigger under rules_secondary. With the 15-day public-filing rule and a one-week roadshow, pricing by Oct 23 is the best-case path. Pre-market 27% YES, blended to 30% against a book that fell from 57¢ to 33¢ on Sep 11–12; NO at 67¢ is a 3¢ edge, below slippage.
Bank of Canada Oct 28
The Bank held at 2.25% on Sep 2 (seventh straight hold) with Macklem warning that "multiple" increases could be needed if inflation, at 3%, does not cool. CORRA swaps now price a 64% hike / 36% hold (bluegamma, bankofcanadaodds, both Sep 14); National Bank and Scotiabank forecast an October hike. Kalshi's hold leg at 67–69¢ is 30+ points richer than swaps, but the only mispriced side is a NO at 33¢ in the tail band, and the 25 bp leg is 28/34 (spread 6¢). Aug CPI prints Sep 15. Precedent (KXCBDECISIONCANADA-26JUL-H0): the hold leg is carry, not a view.
Hegseth by Oct 1
Army Secretary Driscoll resigned Sep 1 and Sen. Tillis called for replacement Sep 2; Trump: "fantastic job". The Oct 1 rung sat 4/6 through it (35k OI). Precedent KXHEGSETHOUT-26APR-SEP01: criticism volume is not a White House signal. NO at 96¢ is a deep-favorite reject; YES at 6¢ is a tail.
Data and API notes
Query API schema fetched first; tables: trading_markets, trading_events, signals, signal_snapshots, kalshi_schedules, resolution_precedents (158 rows). Public API `*_dollars`/`*_fp` fields used throughout; candlesticks from `/series/{series}/markets/{ticker}/candlesticks`. The mirror still lists the settled Aug 30 and Sep 6 Suez ladders as active; settlement was confirmed from the public API. Two precedent rows written to precedents.json for KXSUEZWEEKLY.