# Kalshi AI — Mispricing Audit (v4, archetype-targeted) — 2026-08-25

Date: 2026-08-25
Source: https://kalshi-audits.pages.dev/2026-08-25-kalshi-ai-mispricing-audit-v4

---
# Kalshi AI — Mispricing Audit v4 · archetype-targeted

Report date 2026-08-25 · Horizon 45 days (through 2026-10-09) · Capital $2,500 · Category focus: AI (advisory) · Universe: `targets.json` allowlist (118 series, generated 2026-08-25 15:06 UTC)

## Executive summary

**Zero picks this run. 100% cash.** The Stage-0 allowlist yielded 103 open markets inside the horizon; the v4 liquidity floor (live vol24h ≥ 500) and price-band screens cut 100 of them, and all three Stage-2 survivors failed the band/edge tests after full diligence. Every playable expression this week was either a sub-40¢ tail, a ≥90¢ deep favorite, or a thin favorite whose ask already exceeded my blended probability. An efficient week — cash is the position.

Picks0
Rejects logged15
Deployed$0
Cash$2,500 (100%)
Precedent rows written2

Notable context: the allowlist's highest-volume series (KXSAVEACT vol24 83.8k, KXRETIREMM 39.5k, KXRDUVISITS 16.2k) all close beyond the 45-day horizon (Nov 2026–Feb 2027) — verified live against the public API, not just the mirror. The in-horizon slate was structurally thin.

## Stage-2 rejects — the three that earned full diligence

### KXCBDECISIONJAPAN-26SEP17-H25 scheduled-process REJECT — no edge at ask

Will the Bank of Japan hike 25bps at the September 17–18, 2026 meeting? · Entry context: YES 87 / 89, spread 2¢, vol24h 1,290, OI 24,033, ask depth ≈ 22 contracts · closes 2026-09-18 02:29 UTC

**Prob: 84% pre-market → 85% final** (blended toward a liquid book at 87–89). Rate is 1.0% since the June 16 hike; July 30–31 was a hold with a statement flagging core inflation "clearly above 2%" in H2 FY2026; a Reuters sources story says the BoJ is "eyeing a September hike, faster pace of tightening"; the Aug 25 underlying-inflation print stayed above target; Polymarket 84%. **Precedent:** KXCBDECISIONJAPAN-26JUN15-H25 resolved YES — "the tradeable window is the run-up (signal-following), not the announcement; by the time swaps read ~73% the edge is gone." At 89¢ ask the leak is fully public and priced. YES needs p>89; NO at 12–13¢ is a banned <40¢ tail and is exactly the June losing trade (over-owned hold vs a telegraphed BoJ). Ask depth of ~22 contracts would cap the position at ~7 contracts anyway.

### KXCBDISRAEL-26AUG31-HOLD scheduled-process REJECT — bands + resolver timing watch: mispriced, untradeable

Will the Bank of Israel hold at the August meeting? · Entry context: YES 90 / 92, vol24h 664, OI 31,827 · closes 2026-08-31 12:59 UTC, latest expiration 2026-09-07

**Prob: 64% pre-market → 78% final** (blended hard toward a 32k-OI book that presumably carries local analyst consensus). The interesting find of the run: the decision publication has moved from Aug 31 to **Sep 1** (Governor Yaron on Aug 18: the next meeting "in about two weeks" is a "live decision"). The rules' delay clause ("delayed past the expiration date → No-change resolves YES") keys off expiration, and latest expiration is Sep 7 — so this almost certainly settles on the _actual_ Sep 1 decision, one meeting after two consecutive cuts to 3.50% with the BoI's own research department projecting 3.0% within a year. YES at 92¢ has not repriced Yaron's "live" remark. But: the value side (NO ≈ 9–10¢) is a banned <40¢ tail, YES at 92 is the banned ≥90 deep-favorite band, and the expected-vs-latest expiration reading is itself >15 pts of resolver-timing risk. Logged, not traded; a precedent-store note records the observation for scoring.

### KXNJGOLFVISITCOUNT-26AUG-3 multi-news-other REJECT — no playable side

Exactly 3 Trump visits to Bedminster in Aug 2026? · Entry context: YES 86 / 88, spread 2¢, vol24h 611, OI 16,680, ask depth ≈ 12 contracts · closes 2026-08-31 14:00 UTC

**Prob: 70% pre-market → 78% final.** Count verified at exactly 3 from primary-source schedules: the Jul 31 arrival with a stay spanning Aug 1–2 counts under the "in progress at month start" rule (#1), plus arrivals Aug 7 (#2) and Aug 14 (#3); the Aug 21–23 weekend was Myrtle Beach / Trump National DC / Freedom 250 Grand Prix, not Bedminster. The entire question is a 4th arrival Aug 28–31; the weekend press schedule is unpublished as of Tuesday and nothing is announced either way. He went 3 of the month's first 4 weekends — I can't get P(4th trip) below ~20% on evidence, so YES at 88 has no edge (needs p>88), and NO at 14¢ is a banned tail. Also flagged by the v4 thesis screen as a count/pace bet. Sister leg KXNJGOLFVISITCOUNT-26AUG-4 (NO at 92¢) rejected as cluster runner-up in the deep-favorite band. Correlated context, not a cluster conflict: active pick KXLAGODAYS-26AUG-0 YES rides the same "Trump summers in NJ, not FL" fact.

## Reject log (all)

| Ticker | Archetype | Side | Px | Model %YES | Reason |
|---|---|---|---|---|---|
| KXCBDECISIONJAPAN-26SEP17-H25 | scheduled-process | YES | 89 | 85 | p < ask; run-up edge gone per June precedent; NO is banned tail; depth ~22 |
| KXCBDISRAEL-26AUG31-HOLD | scheduled-process | NO | 10 | 78 | value side is <40¢ tail; YES is ≥90 deep favorite; expiration-reading resolver risk >15 pts |
| KXNJGOLFVISITCOUNT-26AUG-3 | multi-news-other | YES | 88 | 78 | no playable side: YES needs p>88, NO is 14¢ tail; count/pace thesis |
| KXNJGOLFVISITCOUNT-26AUG-4 | multi-news-other | NO | 92 | 20 | cluster runner-up; NO entry 92¢ deep-favorite band |
| KXHEGSETHOUT-26APR-SEP01 | multi-news-other | — | 1 | — | already an active pick |
| KXPIRROOUT-26AUG-26SEP01 | multi-news-other | — | 8 | — | already an active pick |
| KXIPOANTHROPIC-DATE-26OCT01 | corporate-event | — | 24 | — | already an active pick |
| KXCOMPANYACTIONMERGER-27-26SEP01 | corporate-event | — | 1 | — | YES ask ≤5¢ tail; NO deep favorite (vol24 30k noted) |
| KXCOMPANYACTIONMERGER-27-26OCT01 | corporate-event | — | 3 | — | YES ask ≤5¢ tail; NO deep favorite |
| KXUSAIRANAGREEMENT-27-26SEP | scheduled-process | — | 1 | — | YES ask ≤5¢ tail; NO deep favorite |
| KXUSAIRANAGREEMENT-27-26OCT | scheduled-process | — | 2 | — | YES ask ≤5¢ tail; NO deep favorite |
| KXIPOANTHROPIC-DATE-26SEP01 | corporate-event | — | 1 | — | YES ask ≤5¢ tail; adjacent leg already active |
| KXLEAVEPOWELLGOV-26SEP01 | departure/ouster | — | 2 | — | YES ask ≤5¢ tail; NO deep favorite |
| KXZELENSKYYOUT-26OCT01 | departure/ouster | — | 5 | — | vol24h 346 < 500; spread >5¢ |
| KXKASHOUT-26APR-26SEP01 | departure/ouster | — | 2 | — | vol24h 309 < 500 |

Not itemized: ~85 further Stage-0 mechanical cuts, almost all on the vol24h ≥ 500 floor (the entire KXEBOLACOUNTRY, KXCBDECISIONCANADA/MEXICO ladders, KXGOVTFULLFUND, KXHOUSEADJOURN, KXLOCALLAW, and the low-volume departure/ouster tail). The floor exists because 98 of 136 historical picks entered zero-volume books and that cohort had negative edge.

## Cluster table

| Cluster | Legs seen | Deployed leg | Outcome |
|---|---|---|---|
| boj-sep-2026-decision | 1 (H25; other rungs failed vol floor) | none | reject — no edge at ask |
| boi-aug-2026-decision | 1 (HOLD; other rungs failed vol floor) | none | reject — bands + resolver timing |
| trump-bedminster-aug-count | 2 (buckets 3, 4) | none | reject — no playable side; runner-up logged |

## Position sizes & exposure

| Conviction | Cap/pick | Positions | Cost |
|---|---|---|---|
| HIGH | $375 (15%) | 0 | $0 |
| MEDIUM | $175 (7%) | 0 | $0 |
| **Cash** | — | — | **$2,500 (100%)** |

Cash is a position; 40–70% cash is expected in an efficient week and this one cleared 100%. Reserve requirement (≥10%) trivially satisfied. No EV is booked for the record because nothing deployed.

## Risk profile & execution notes

- **Live portfolio note:** nothing new correlates against the 14 existing picks. The Bedminster reject shares a fact pattern with the active KXLAGODAYS-26AUG-0 YES (Trump's August base is NJ) but no new capital touches it.
- **Watch item (next run):** KXCBDISRAEL — if a September-meeting event lists after the Sep 1 decision, the "live decision" repricing failure seen here is the template. Check whether the book carries the same hold-side complacency inside the 40–84¢ band.
- **Watch item:** KXCBDECISIONJAPAN October/December rungs — the precedent says the payoff is following the leak _early_; a sources-story day is the entry window, not meeting week.
- **Slate breadth:** Tier-1 in-horizon markets were almost uniformly dead books (<500 vol24h). If this repeats, the weekly classifier should consider whether the min-OI=5000 filter is admitting series whose in-horizon legs are illiquid shells.

## Precedents consulted

- KXCBDECISIONJAPAN-26JUN15-H25 (YES) / -HOLD (NO) — telegraphed BoJ hikes deliver; the hold side was over-owned at ~27% and lost; edge window is the run-up.
- KXCBDECISIONJAPAN-26JUN15-H25P (NO) — BoJ moves in strict 25bp steps; outer rungs are structural zeros.
- KXFEDDECISION-26JUL ladder — only the two rungs adjacent to the current stance carry premium; a hawkish repricing into a meeting is a fade, not a signal.
- KXCBDECISIONCANADA-26JUL ladder — "safe" hold legs pay 1–3¢; resolver-safe ≠ tradeable.
- KXNJGOLFVISITCOUNT / KXCBDISRAEL: no series precedent in store (cold derivation done from live rules); two rows written back this run (BoJ July hold, BoI July cut).

## Method

Keel v4: Stage 0 allowlist universe (targets.json only) with hard mechanical cuts (vol24h ≥ 500 live, spread ≤ 5¢, 5–95¢ book, 3h halt, close-vs-expiration ≤ 1h, no active picks, no Mentions); Stage 1 precedent retrieval before opinion; Stage 2 probability committed before studying the book, then blended; Stage 2.5 resolver profiling with point-priced risks; Stage 3 reject screens (continuous-number theses, book-derived no-trade band, underspecified rules); Stage 3.5 entry bands (40–84¢ core, 85–89¢ half-size, ≥90¢ and <40¢ banned); Stage 4 one leg per cluster, conviction-capped EV-agnostic sizing. Sources this run: Kalshi query mirror + public API (fp/dollar fields), resolution_precedents store, Reuters/CNBC/Bloomberg/BoI/WH press-pool primary reporting.

  Generated 2026-08-25 · run kalshi-mispricing-ai-v2-vxtdg · Keel v4 · 0 picks / 15 logged rejects / $2,500 cash.

  Key sources: CNBC (BoJ June 16 hike to 1%; July 31 hold), Reuters via Investing.com (BoJ eyeing September hike), Bloomberg (BoJ price gauge Aug 25; Yaron "live decision" Aug 18), Bank of Israel press release (July 6 cut to 3.5%), WH Press Pool / Factba.se / CDN daily schedules (Bedminster arrivals Jul 31, Aug 7, Aug 14; Myrtle Beach–DC–Grand Prix weekend Aug 21–23), Polymarket (BoJ September 84%).