Keel v4 · Kalshi mispricing audit · topic: Politics · report date 2026-09-17 · run kalshi-mispricing-politics-10k-4x4sz

Kalshi Politics — Mispricing Audit (v4, archetype-targeted)

Universe: 120 pre-classified target series (targets.json, generated 2026-09-17 12:07Z) · horizon 45 days (close by 2026-11-01) · capital $10,000

0picks deployed

$0allocated (0% of capital)

$10,000held in cash (100%)

7Stage-2 candidates

23logged rejects

Verdict: no trade this week. Only 7 of 899 open markets across the 120 allowlisted series survived the Stage-0 cuts inside the 45-day horizon, and none survived the v4 entry bands. The two genuinely live questions, the Bank of Canada's Oct 28 decision and the Anthropic IPO Nov 1 leg, both showed the same shape: the side with edge sits under the 40¢ floor or inside a cluster that already carries live picks. Cash is the position. The 8 existing picks are untouched and none appears below as a new pick.

1. Stage 0 — universe and mechanical cuts

All 120 series in targets.json were pulled live from the Kalshi public API (899 open markets). Cuts applied in order: beyond horizon (the bulk of Tier-1 volume sits in Dec 31 / Jan 1 rungs), existing picks, mentions backstop, vol24h < 500 live, spread > 5¢, YES bid ≥ 95¢, YES ask ≤ 5¢, close more than 1h after expected expiration, inside 3h of expiration.

Cut	Markets	Notes
Beyond 45-day horizon	~812	KXSAVEACT, KXRETIREMM (closes Nov 3), KXTRUMPADMINLEAVE, KXSANCTIONRUSSIA, KXBILLS and every other high-volume Tier-1 series close after Nov 1 — the reason this slate is thin.
vol24h < 500 (live)	62	Includes every Tier-1 rung inside the horizon except Kash Patel Nov 1. Zero-volume books are the documented phantom-edge cohort.
YES ask ≤ 5¢ / YES bid ≥ 95¢	12	Settled-in-all-but-name rungs (BoJ Sep hike, CNB hold, Hegseth Oct 1, Iran deal Oct 1).
Spread > 5¢	2	Anthropic OCT24 (28/35), Ebola-USA (5/11).
Already an active pick	2	Anthropic OCT01, OCT17.
Close after expiration	1	KXHEGSETHOUT-26AUG-26NOV01 — trades a month past its Oct 1 expected expiration.
Survivors to Stage 2	7	BoC Oct H0 + H25, Anthropic NOV01, Kash Patel NOV01, Trump-criticizes-Warsh OCT01 + NOV01, Tesla/SpaceX NOV01.

2. Picks

0 picks No market cleared Stage 3 and Stage 3.5 together. Cash is a position: $10,000 (100%) held, reserve requirement (≥10%) trivially met.

3. Stage 2 diligence — the two live questions

Bank of Canada, Oct 28, 2026 decision scheduled-processTier 2rejected: <40¢ band

Legs: KXCBDECISIONCANADA-26OCT-H25 (hike 25bp) 37/38¢ · vol24 3,007 · OI 26,848 · KXCBDECISIONCANADA-26OCT-H0 (hold) 63/64¢ · vol24 3,327 · OI 24,439

Resolver: Kalshi off the BoC press release at 09:45 ET on Oct 28; both legs close at 13:29Z. Fixed-date, mechanical, no clarification text. Precedents (11 rows, Jun–Sep 2026): every rung matched consensus; the store's Sep 16 row already flagged this exact Kalshi-vs-OIS gap as "observable, not deployable".

Primary evidence (dates): Sep 2 hold at 2.25%, seventh straight (BoC release). Sep 14 August CPI: headline 3.0% unchanged, CPI-trim 1.9%, CPI-median 2.0%, gasoline +22.8% y/y (StatCan via TradingEconomics). Sep 16 summary of deliberations: a hike "could" be required only "if higher energy prices did spill over into other components of the CPI"; CPI ex-gasoline 2.2%, core "around 2%"; Council worried the re-escalated US tariff dispute "would weigh on consumer and business confidence". Sep 16 LSEG: 60% odds of an October hike; Sep 17 BoC-odds tracker: 59% hike / 41% hold. Economists split: Scotiabank (Holt) has October "in play"; Desjardins (Mendes) sees nothing until H1 2027.

Prob: P(hike) 45% pre-market → 50% final (blended toward a deep OIS market at 59–60%). The Bank's own trigger is core spillover and the August print showed none, while the tariff shock cuts the other way; OIS is a market view, not a Bank signal, but it is a far better-informed book than Kalshi's.

Band test: H25 YES: p=50 vs ask 38 → 12 pts of paper edge, but 38¢ entry is a forbidden tail. H0 NO: p(NO)=51 vs NO ask 37 → same tail. H0 YES at 64¢: p=49 → no edge. Result: both legs logged as rejects. Price path (YES, last 8 sessions): H0 10:84 → 11:79 → 12:79 → 13:70 → 14:64 → 15:68 → 16:69 → 17:64; H25 10:15 → 11:22 → 12:22 → 13:27 → 14:35 → 15:32 → 16:42 → 17:37.

Precedents consulted: KXCBDECISIONCANADA-26JUL-H0/H25/H26/C25/C26, -26JUN-H0/H25/H26/C25/C26, -26SEP-H0 (store rows); 26SEP-H0/H25 results verified live (yes/no).

Anthropic confirms an IPO before Nov 1, 2026 corporate-eventTier 2rejected: cluster occupied + band

Leg: KXIPOANTHROPIC-DATE-26NOV01 53/54¢ · vol24 3,481 · OI 88,693 · YES bid depth 4,006 at 53¢.

Resolver: mechanical — SEC declares the S-1 effective, OR the deal is priced, OR an exchange assigns a ticker, before Nov 1 04:59Z (rules_secondary). Precedents: JUL01 and JUN01 resolved NO despite the Jun 1 confidential-filing headline; KXIPO-26-SPACEX did not resolve on a proposed symbol in a public S-1; the store's SEP01 row maps a Kalshi rung to the Polymarket "trading by" rung dated one day later.

Primary evidence (dates): EDGAR full-text search Aug 1–Sep 17: zero public S-1 / S-1/A for Anthropic (verified this run). Reuters via CNBC Sep 5: marketing "mid-October at the earliest", listing "days before the U.S. midterm elections" (Nov 3), prospectus "not expected until late September". Bloomberg Sep 13: Nasdaq chosen; IPO "as soon as October". Bloomberg Sep 16: "status of Anthropic IPO as AI fears mount". Polymarket (gamma API, Sep 17): trading by Oct 15 1.9%, by Oct 31 52.5% ($885k), by Nov 15 74.5%, by Dec 31 86.5%.

Prob: P(YES) 45% pre-market → 49% final. The current plan lands pricing around Oct 27–29, inside the window with only days of slack; a one-week slip pushes it past the midterms into Nov 9+, so the outcome is bimodal. Two liquid books (Kalshi 53/54 with 88.7k OI, Polymarket 52.5%) agree, and my evidence is not more specific than theirs.

Why rejected: (1) the anthropic-ipo-announce cluster already carries two live NO legs (OCT01, OCT17) — a NOV01 NO is the third leg on one thesis, the exact multi-leg pattern v4 bans; (2) NO at 47¢ is in the 40–59¢ band, which requires HIGH conviction and a Tier-1 process archetype — this is Tier-2 corporate-event at MEDIUM; (3) ~4 pts of edge is inside noise. Price path (YES): 10:72 → 11:70 → 12:45 → 13:55 → 14:57 → 15:53 → 16:56 → 17:53 — note the Sep 12 print at 45¢ that reverted within a day.

Precedents consulted: KXIPOANTHROPIC-DATE-26JUL01, -26JUN01, -26SEP01, KXIPO-26-SPACEX.

Band-only rejects (diligence kept light by design)

Kash Patel out before Nov 1 (KXKASHOUT-26JUL-26NOV01, 6/7¢ · vol24 5,702 · OI 30,049). Patel sat through a five-hour Senate Judiciary hearing on Sep 15 (Washington Post, CBS) with no removal signal and Trump still behind him. Model 5% YES agrees with the book. NO at 93–94¢ is deep-favorite band; YES 7¢ is a tail. Precedent: KXKASHOUT-26APR-JUL01/AUG01/SEP01 all NO; far rungs systematically overpriced.
Trump publicly criticizes Kevin Warsh before Oct 1 / Nov 1 (KXDJTATTACKWARSH-26SEP-26OCT01 4/11¢ · vol24 32,326 · OI 25,288; -26NOV01 18/20¢ · vol24 16,833 · OI 10,604). Speech-act market — the v4 mentions backstop applies whatever the filing category. For the record: the Fed hiked 25bp unanimously on Sep 16 with Warsh voting for it; Trump demanded 1% rates on Truth Social but said he "still has confidence" in Warsh (CNBC, Axios Sep 16). Lean YES 15% / 30% (the Oct 27–28 FOMC sits inside the Nov 1 window), both tails, both excluded by archetype.
Tesla/SpaceX definitive agreement before Nov 1 (KXCOMPANYACTIONMERGER-27-26NOV01, 8/11¢ · vol24 1,052 · OI 70,049). Rules require an official announcement of a signed definitive agreement and explicitly exclude Musk intent statements; his Sep 15 All-In Summit remarks do not count. Model 4% YES. NO 89–92¢ deep-favorite; YES 11¢ tail.

4. Reject log (Stage 2 survivors, side formed)

model_prob_pct is the probability of the side named. Prices are the ask of the side named.

Market	Archetype	Side	Price	Model	Cluster	Reason
`KXCBDECISIONCANADA-26OCT-H25` Bank of Canada hikes 25bp at the Oct 28, 2026 meeting?	scheduled-process	YES	38¢	50%	boc-oct28-decision	<40c tail band. P(hike) 45% pre-market -> 50% final (blended toward OIS at 59-60%, LSEG Sep 16 / BoC-odds Sep 17). YES ask 38c carries ~12 pts of paper edge but sits under the 40c floor; the NO side (62c) has negative edge. Same undeployable dislocation the Sep 16 precedent row documented.
`KXCBDECISIONCANADA-26OCT-H0` Bank of Canada holds (0bp) at the Oct 28, 2026 meeting?	scheduled-process	NO	37¢	51%	boc-oct28-decision	<40c tail band on the edged side (NO ask 37c vs 51% model) and cluster runner-up to the H25 leg; the YES/hold side at 64c has no edge against a 49% model. Core CPI at 1.9/2.0 and a fresh US-tariff shock keep a hold live, but not at 64c.
`KXIPOANTHROPIC-DATE-26NOV01` Anthropic confirms an IPO (S-1 effective / priced / ticker assigned) before Nov 1, 2026?	corporate-event	NO	47¢	51%	anthropic-ipo-announce	Cluster already carries two live NO legs (OCT01, OCT17) - a third leg is the multi-leg pattern v4 bans. Also fails the 40-59c band test: Tier-2 corporate-event at MEDIUM conviction, not HIGH + Tier-1. P(YES) 45% pre-market -> 49% final; NO at 47c is ~4 pts of edge against a Polymarket twin at 52.5% ($885k) - inside noise.
`KXKASHOUT-26JUL-26NOV01` Kash Patel out as FBI Director before Nov 1, 2026?	departure/ouster	NO	94¢	95%	patel-fbi-exit	No playable band: NO 93-94c is deep-favorite (>=90c reject), YES 7c is a <40c tail. Model 5% YES agrees with the book - Patel sat through a 5-hour Senate Judiciary hearing Sep 15 with Trump still behind him; precedent KXKASHOUT-26APR-JUL01/AUG01/SEP01 all resolved NO.
`KXDJTATTACKWARSH-26SEP-26OCT01` Trump publicly criticizes Kevin Warsh before Oct 1, 2026?	multi-news-other	YES	11¢	15%	trump-criticize-warsh	Mentions-type speech-act market ('will X say Y') - v4 backstop skip. Also a <40c tail and the live spread widened to 4/11c (>5c). Trump attacked the Sep 16 Fed hike on Truth Social but said he 'still has confidence' in Warsh (CNBC Sep 16); next FOMC is Oct 27-28, after this rung closes.
`KXDJTATTACKWARSH-26SEP-26NOV01` Trump publicly criticizes Kevin Warsh before Nov 1, 2026?	multi-news-other	YES	20¢	30%	trump-criticize-warsh	Mentions-type speech-act market - v4 backstop skip; YES is a <40c tail. The Oct 27-28 FOMC sits inside this window and a second hike could draw a by-name attack, so the lean is YES vs a 20c ask, but the archetype is excluded regardless of the read.
`KXCOMPANYACTIONMERGER-27-26NOV01` Tesla/SpaceX announce a definitive merger or acquisition agreement before Nov 1, 2026?	corporate-event	NO	92¢	96%	tesla-spacex-merger	No playable band: NO 89-92c deep-favorite, YES 11c tail. Rules exclude Musk intent/hypothetical statements (Sep 15 All-In Summit remarks do not count); no definitive agreement reported. Model 4% YES.

Stage-0 mechanical cuts logged for the record (no side formed)

Market	Archetype	YES px	Reason
`KXHEGSETHOUT-26AUG-26NOV01` Hegseth out as SecDef before Nov 1, 2026?	multi-news-other	11¢	Stage-0: close after expiration (close Nov 1 04:59Z vs expected expiration Oct 1 14:00Z)
`KXIPOANTHROPIC-DATE-26OCT24` Anthropic confirms an IPO before Oct 24, 2026?	corporate-event	35¢	Stage-0: spread>5c (28/35)
`KXIPOANTHROPIC-DATE-26OCT17` Anthropic confirms an IPO before Oct 17, 2026?	corporate-event	16¢	Stage-0: already an active pick
`KXIPOANTHROPIC-DATE-26OCT01` Anthropic confirms an IPO before Oct 1, 2026?	corporate-event	1¢	Stage-0: already an active pick
`KXHEGSETHOUT-26AUG-26OCT01` Hegseth out as SecDef before Oct 1, 2026?	multi-news-other	4¢	Stage-0: yes ask<=5c
`KXUSAIRANAGREEMENT-27-26OCT` US agrees to a new Iran nuclear deal before Oct 1, 2026?	scheduled-process	1¢	Stage-0: yes ask<=5c
`KXCOMPANYACTIONMERGER-27-26OCT01` Tesla/SpaceX definitive agreement before Oct 1, 2026?	corporate-event	2¢	Stage-0: yes ask<=5c
`KXDJTATTACKWARSH-26SEP-26SEP18` Trump publicly criticizes Warsh before Sep 18, 2026?	multi-news-other	4¢	Stage-0: yes ask<=5c; mentions-type market
`KXCBDECISIONJAPAN-26SEP17-H25` BoJ hikes 25bp at the Sep 2026 meeting?	scheduled-process	100¢	Stage-0: yes bid>=95c
`KXCBDCZECH-26SEP17-HOLD` Czech National Bank holds at the Sep 2026 meeting?	scheduled-process	98¢	Stage-0: yes bid>=95c; inside 3h halt
`KXEBOLACOUNTRY-OCT26-USA` Human Ebola case reported in the US before Oct 1, 2026?	multi-news-other	11¢	Stage-0: spread>5c (5/11)
`KXAGICO-COMP-26Q3` Any company announces AGI achieved before Oct 1, 2026?	multi-news-other	4¢	Stage-0: yes ask<=5c; underspecified rules
`KXKASHOUT-26JUL-26OCT01` Kash Patel out as FBI Director before Oct 1, 2026?	departure/ouster	5¢	Stage-0: vol24h<500 (217); NO side 96c deep-favorite anyway
`KXNUMAPPROP-26OCT01-AL1` At least 1 FY27 appropriations bill enacted before Oct 1, 2026?	confirmation/enact	6¢	Stage-0: vol24h<500 (200)
`KXCBDECISIONJAPAN-26OCT29-H25` BoJ hikes 25bp at the Oct 2026 meeting?	scheduled-process	48¢	Stage-0: vol24h<500 (402); spread 39/48
`KXCBDECISIONMEXICO-26SEP24-HOLD` Banxico holds at the Sep 24, 2026 meeting?	scheduled-process	96¢	Stage-0: vol24h<500 (351); yes bid>=95c region

5. Position sizes, limits and EV (for the record)

Ticker	Side	Limit	Contracts	Cost	Max win	EV %
No positions. $10,000 cash (100%). Reserve ≥ $1,000 satisfied.

6. Cluster table

Cluster	Legs surveyed	Deployed leg	Status
boc-oct28-decision	H0, H25	—	Both legs under the 40¢ floor on the edged side
anthropic-ipo-announce	NOV01 (OCT01, OCT17 live from prior runs; OCT24 spread cut)	—	Cluster already carries two live NO legs; no third leg
patel-fbi-exit	NOV01 (OCT01 vol cut)	—	Deep-favorite / tail
trump-criticize-warsh	OCT01, NOV01 (SEP18 ask cut)	—	Mentions-type, excluded
tesla-spacex-merger	NOV01 (OCT01 ask cut)	—	Deep-favorite / tail

7. Conviction exposure and risk profile

Conviction	Picks	Cost	% capital
HIGH	0	$0	0%
MEDIUM	0	$0	0%
Cash	—	$10,000	100%

Portfolio risk is zero by construction. The live book of 8 existing picks (Anthropic OCT01/OCT17 NO, crypto structure NO, FISA NO, Hormuz NO, Suez YES) is unchanged by this run. Nothing surfaced this week contradicts the Anthropic NO legs: no public S-1 on EDGAR as of Sep 17 and marketing not before mid-October leave the OCT01 and OCT17 triggers out of reach on any reported timeline.

8. Execution notes

Nothing to execute. Do not chase the BoC hike leg at 38¢ or the hold NO at 37¢; if either edged side lifts into the 40–59¢ band before Oct 28, the HIGH + Tier-1 requirement still blocks it (Tier-2 scheduled-process). This ladder is a shadow-track for the band rule, per the Sep 16 precedent note.
Watch-list for next run: Sep 30 / Oct 1 rungs roll off; Nov 3-and-later Tier-1 rungs (KXRETIREMM-26 closes Nov 3, KXTRUMPADMINLEAVE Dec 31) enter the 45-day window in early October and carry the volume this run lacked.
Calendar inside the window: Canada jobs Oct 9, BoC Business Outlook Survey ~Oct 19, Canada CPI ~Oct 20, FOMC Oct 27–28, BoC decision + MPR Oct 28, Anthropic roadshow mid-October per Reuters.

9. Precedents consulted and written back

Consulted: 23 store rows across KXCBDECISIONCANADA (11), KXIPOANTHROPIC/KXIPO (4), KXKASHOUT (1), and the departure/ouster archetype (7). Written back to /tmp/out/precedents.json: KXCBDECISIONCANADA-26SEP-H0 (result yes, verified live) and -26SEP-H25 (result no) with the Sep 2 hold facts, the Sep 16 deliberations hike condition, the Aug CPI core prints and the Kalshi-vs-OIS gap history.

10. Sources

Methodology: Keel v4 — Stage-0 allowlist universe, probability committed before opening the book, precedent retrieval before diligence, one leg per cluster, 40–84¢ entry band with 85–89¢ at half size, live-book liquidity floor. Zero picks is an accepted outcome.