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

Report date 2026-08-18 · Horizon 45 days (through 2026-10-02) · Capital $2,500 · Category focus: AI (advisory) · Universe: Stage-0 allowlist (targets.json, generated 2026-08-18) · Methodology: Keel v4 — probability before price, precedent-fed, one leg per cluster

Summary

1 pick, 12 logged rejects, $63.00 deployed (2.5% of capital), $2,437 cash (97.5%). A thin, efficient week: of 92 in-horizon markets across the allowlisted series, only four passed the mechanical cuts (vol24h ≥ 500 live, spread ≤ 5¢, tradeable band) and one of those is an existing pick. Nearly everything in-horizon is either a zero/low-volume book — the documented phantom-edge cohort — or priced as a tail/deep favorite with no playable side. The single deployment is a NO on Anthropic confirming an IPO before Oct 1, a precedent-backed mechanical-trigger decay trade in the core 60–84¢ band. Cash is a position.

Pick 1 of 1 — BUY NO · KXIPOANTHROPIC-DATE-26OCT01

NO @ 84¢ limit MEDIUM archetype: corporate-event · Tier 2 cluster: anthropic-ipo-timing KXIPOANTHROPIC-DATE-26OCT01

When will Anthropic officially announce an IPO? — Before Oct 1, 2026 (closes 2026-10-01 03:59Z)

Prob (final)NO 86%
Prob (pre-market)NO 88%
Entry (NO)84¢ limit · ~83.7¢ VWAP
Contracts75
Max cost / loss$63.00
EV (record only)+2.0¢/ct · +2.4%
Yes bid/ask18¢ / 22¢
Depth ≤ limit227 contracts
Vol 24h2,027
Open interest111,455

Prob: 88% pre-market → 86% final (blended toward a liquid book pricing NO ≈ 80).

Resolver & mechanism. Kalshi's own rules_secondary makes resolution purely mechanical: YES fires only if, before Oct 1, (1) the SEC declares the S-1 effective, (2) the IPO is priced, or (3) an exchange assigns a ticker. Nothing journalistic counts. Verified on EDGAR today: no public S-1 flip exists as of Aug 18 — the Jun 1 filing remains a confidential draft. SEC guidance requires the public flip ≥15 days before a roadshow; the fastest mega-IPO comp (SpaceX) still took ~4 weeks flip-to-pricing. So YES needs a public flip by ~Sep 8 and a maximally fast roadshow with zero slippage. Latest timeline data points the other way: TradingView's IPO calendar lists a planned offer date of Oct 23; futuresearch's median listing forecast moved out to Nov 30 (Aug 6 update); Forbes (Aug 13) says October. Every flip-less day mechanically decays YES.

Why the price ignores this. YES spiked 15¢ → 32¢ over Aug 9–14 (22.7k contracts on Aug 11) with no identifiable catalyst, and is decaying (21¢ close Aug 18). That is the series' documented headline-flow pattern: buyers of "Anthropic IPO coming" stories systematically overpay against triggers that lag by months.

Resolver risk pricing. −3 pts for a loose "ticker assigned" reading if the public flip discloses an intended symbol — bounded by the SpaceX precedent, where the Jun 4 roadshow ticker disclosure did not settle the market (triggers counted Jun 11–12). No other deductions; Kalshi's settle-lag quirk favors NO. Total deductions 3 < 15 ✓. 48h check: YES drifted 25¢ → 21¢ — decay, not a news event; entry remains below model.

Precedents consulted: KXIPOANTHROPIC-DATE-26JUL01 resolved NO — the Jun 1 confidential draft S-1 announcement met none of the three triggers; headline YES-buyers were wrong within 24h (130k contracts repriced 5¢→3¢→1–2¢). KXIPOANTHROPIC-DATE-26JUN01 resolved NO — "officially announce" is mechanical, not journalistic. KXIPO-26-SPACEX resolved YES only at S-1 effectiveness/pricing, not at roadshow ticker disclosure.

Position sizes & limits

TickerSideLimitContractsMax cost% of capitalCaps applied
KXIPOANTHROPIC-DATE-26OCT01NO84¢75$63.002.5% Depth cap binding: ⅓ × 227 visible ≤ 84¢ = 75 (MEDIUM cap would allow $175 / 208 cts)
Cash reserve$2,437.0097.5%≥10% reserve rule trivially met

EV for the record (not used for sizing)

TickerSideEntryModel probEdgeEV/contractEV on cost
KXIPOANTHROPIC-DATE-26OCT01NO84¢86%+2.0¢+2.0¢+2.4%

Sizing is conviction-weighted and EV-agnostic per v4; this table exists for the resolved-record audit trail.

Clusters

ClusterDeployed legRunner-up legs (logged rejects)
anthropic-ipo-timingKXIPOANTHROPIC-DATE-26OCT01 (NO)KXIPOANTHROPIC-DATE-26SEP01

One leg per cluster enforced; only one cluster deployed this run.

Conviction exposure

ConvictionPositionsCost at risk% of capitalCap
HIGH0$00%15%/pick
MEDIUM1$63.002.5%7%/pick
Cash$2,437.0097.5%

Risk profile

Execution notes

Rejects log (12)

TickerArchetypePxReason
KXNJGOLFVISITCOUNT-26AUG-3multi-news-other66 continuous-number thesis — Bedminster count stands at 3 (Aug 1–2, 7–9, 14–16 verified via WH pool/Factbase; the alleged Aug 4–5 NJ stop was actually Trump National LA) with two weekends left; both sides reduce to visit-pace extrapolation, no mechanical lock. NO side ~34¢ is also a <40¢ tail. Model prob (YES) ~45% if forced.
KXNJGOLFVISITCOUNT-26AUG-4multi-news-other24mechanical: spread 7¢, vol24h 413; cluster runner-up of the exactly-3 strike
KXCOMPANYACTIONMERGER-27-26OCT01corporate-event8no playable band — YES 8¢ tail (<40¢, tails 0-for-8 lifetime); NO 93¢ deep-favorite (≥90¢)
KXCOMPANYACTIONMERGER-27-26SEP01corporate-event3deep-favorite band; cluster twin of the Oct strike
KXPIRROOUT-26AUG-26SEP01multi-news-other7already an active pick
KXHEGSETHOUT-26APR-SEP01multi-news-other3already an active pick
KXIPOANTHROPIC-DATE-26SEP01corporate-event99cluster runner-up (Oct 1 strike is the better leg); NO ≥98¢ deep-favorite band; vol24h 361 < 500
KXUSAIRANAGREEMENT-27-26OCTscheduled-process4deep-favorite band — YES 3/4¢ tail, NO ≥96¢; liquidity fine but no playable side
KXUSAIRANAGREEMENT-27-26SEPscheduled-process1deep-favorite band — YES 0/1¢ tail, NO ≥99¢
KXMAMDANIEO-26AUG22-T0confirmation/enact2no playable band — YES 1/2¢ tail, NO 98/99¢; high volume (7,987) does not create a band
KXGOVTFULLFUND-26AUG-26SEP26confirmation/enact60zero-volume book (vol24h 0) — the documented phantom-edge cohort; spread 5¢ borderline
KXCBDECISIONJAPAN-26SEP17-H25scheduled-process76liquidity floor — vol24h 189 < 500 (hard rule); otherwise attractive scheduled-process book at 76/77

Universe & method notes

Stage-0 universe: targets.json (generated 2026-08-18T15:06Z, min OI 5,000) — ~70 Tier-1 and ~45 Tier-2 series. 92 open markets close within the 45-day horizon; 4 passed all mechanical cuts. Tier-1 in-horizon markets were almost universally tails/deep favorites (departure and enactment strikes at ≤5¢ YES) or zero-volume books — no playable Tier-1 leg existed this run, so the sole deployment is Tier-2 (corporate-event), consistent with the AI category focus. Precedent store consulted for KXIPOANTHROPIC (2 rows) + corporate-event archetype (1 row); KXNJGOLFVISITCOUNT and multi-news-other had no precedents (cold derivation noted in reject). Probabilities were committed pre-book per v4 discipline; the pre-market and final numbers are shown on the pick. No new settled-market facts were established this run — precedents.json is empty.

Sources relied on: SEC EDGAR (no public Anthropic S-1 as of Aug 18); Kalshi rules_primary/secondary and orderbooks (live API); Roll Call Factbase / WH pool schedules (Bedminster visits); Forbes Aug 13 (October IPO); TradingView IPO calendar (Oct 23 offer date); futuresearch.ai (median listing Nov 30, Aug 6 update); Anthropic newsroom (confidential draft S-1, Jun 1).

Keel v4 · run kalshi-mispricing-ai-v2-h9f6v · 1 pick / 12 rejects · $63.00 of $2,500 deployed