Kalshi Politics — Mispricing Audit
Report date 2026-07-16 · Methodology v3.3 (conviction-weighted, EV-agnostic sizing; 15% cluster cap; 60–90¢ favorites bias; resolver risks priced as point deductions; LOW conviction never deploys) · Capital $10,000 · Horizon 45 days · Picks in this report count toward the public track record and are cleared for auto-copy.
1. How this was researched
Mode: category-match. $CATEGORY = Politics matched trading_events.category directly: 300 active markets closing before 2026-08-30 in the live Kalshi mirror. No theme-keyword expansion was needed.
Pipeline: (1) schema check against /api/db/schema — note the mirror carries no general price-history table this run (only signal_snapshots, keyed to previously logged signals), so 14-day history came from the Kalshi public candlestick API instead; (2) Stage-1 mechanical screen on live Kalshi API data — volume24h ≥ 1,500, bid-ask spread ≤ 5¢, not fully priced (YES bid < 95¢, YES ask > 5¢), no announcer-noise series (Truth Social post counts, late-night posting hours, photo-days, endorsement counts, etc.), and no ticker already live in users' feeds from $EXISTING_PICKS (37 entries excluded up front, including the whole CPI/payrolls complex, KXKASHOUT, KXHEGSETHOUT, KXKNESSET, KXTRUMPATTEND and both Hormuz legacy picks); (3) Stage-2 diligence on the 14 survivors — full rules text, live 10-level orderbooks, 14-day candlesticks, and primary-source news; (4) Stage-2.5 resolver profiling with priced deductions; (5) Stage-3/3.5 reject screens; (6) conviction-weighted sizing.
One methodological note worth advertising: while pulling the IMF PortWatch resolver feed for the Hormuz markets, the obvious ArcGIS query (portid='chokepoint4') returned a plausible-looking daily series that was actually the Bab el-Mandeb Strait. Hormuz is chokepoint6. The two series differ by 3–4× right now (Bab el-Mandeb ~41/day vs Hormuz ~10/day). An audit that skipped the resolver-verification step would have priced every Hormuz market off the wrong strait — the exact class of error that produced this program's costliest historical loss.
2. Markets reviewed
14 Stage-1 survivors (of 300 screened). Prices are live top-of-book at audit time, in cents.
| Ticker | Market | YES bid/ask | Vol 24h | OI | Outcome of diligence |
|---|---|---|---|---|---|
| KXBLANCHEJUDICIARY-27-26AUG01 | Judiciary reports Blanche AG nomination before Aug 1 | 60 / 62 | 1,535 | 6,057 | PICK — BUY YES |
| KXTRUMPWORLDCUP-26-DJT-1 | Trump attends ≥1 World Cup match | 93 / 94 | 30,797 | 104,679 | Reject — duplicates live pick cluster |
| KXHORMUZWEEKLY-26JUL19-T50 | >50 Hormuz transit calls Jul 13–19 | 82 / 85 | 21,411 | 14,864 | Reject — NO side is a sub-35¢ tail; 48h move |
| KXHORMUZWEEKLY-26JUL19-T75 | >75 Hormuz transit calls Jul 13–19 | 53 / 55 | 8,359 | 10,865 | Reject — coin-flip band, not HIGH conviction |
| KXHORMUZWEEKLY-26JUL19-T45 | >45 Hormuz transit calls Jul 13–19 | 85 / 86 | 2,364 | 3,015 | Reject — no edge, 33-contract depth |
| KXHORMUZPEAK-26JUL19-T15 | Peak single day >15, Jul 13–19 | 76 / 77 | 3,134 | 2,939 | Reject — efficiently priced |
| KXHORMUZPEAK-26JUL19-T20 | Peak single day >20, Jul 13–19 | 47 / 49 | 4,589 | 3,566 | Reject — coin-flip band, no edge |
| KXAPRPOTUS-26JUL17-41.2 | RCP approval 41.1–41.3 at 11am Jul 17 | 43 / 44 | 3,217 | 10,368 | Reject — resolver unobservable |
| KXAPRPOTUS-26JUL17-40.9 | RCP approval 40.8–41.0 at 11am Jul 17 | 29 / 30 | 3,671 | 9,276 | Reject — resolver unobservable |
| KXAPRPOTUS-26JUL17-41.3 | RCP approval >41.3 at 11am Jul 17 | 16 / 20 | 4,732 | 10,233 | Reject — resolver unobservable |
| KXAPRPOTUS-26JUL17-40.6 | RCP approval 40.5–40.7 at 11am Jul 17 | 6 / 7 | 4,771 | 8,374 | Reject — resolver unobservable, >90¢ NO |
| KXTRUMPVH-26JUL17-T40.8 | VoteHub approval >40.8% at 10am Jul 17 | 36 / 42 | 1,706 | 3,286 | Reject — resolver unobservable + 6¢ spread |
| KXTRUMPBIBIMEET-26AUG01-AUG01 | Trump meets Netanyahu before Aug 1 | 82 / 83 | 2,842 | 1,344 | Reject — efficiently priced |
| KXBILLSCOUNT-26JUL-0.0 | Exactly 0 bills become law in July | 21 / 22 | 1,861 | 2,358 | Reject — definitional trap |
3. Top picks — detail and thesis
Pick 1 — KXBLANCHEJUDICIARY-27-26AUG01 · BUY YES @ 0.64 limit · MEDIUM CONVICTION
Market: Will the Senate Judiciary Committee report Todd Blanche's nomination to be U.S. Attorney General to the full Senate before Aug 1, 2026?
Current price: 60 bid / 62 ask (YES)
My probability: 70% · Market implied: ~62%
Entry band: 60–90¢ favorite (effective entry 62–64¢)
Cluster: blanche-ag-confirmation
Close: 2026-08-01T03:59Z (16 days)
Mispricing thesis. The market sold off from 68¢ to 60–62¢ on hearing-week wobble coverage (Cornyn "has concerns," Tillis conditional, the post-Graham one-vote margin). But the mechanics favor YES more than 62¢ implies: the hearing is done (Jul 15, witnesses Jul 16), the committee's regular Thursday executive-business calendar offers two windows before the deadline (Jul 23; standard one-week hold-over lands Jul 30 — still before Aug 1), Blanche is already running the department as acting AG, and leadership wants its AG seated before the August recess. Committee reporting is the step nominations essentially always clear once a majority-party chairman has scheduled the hearing and the White House is actively whipping — failures happen by withdrawal before the hearing (Gaetz), not by losing markup votes.
Evidence.
- Confirmation hearing held Wednesday Jul 15, 2026, Hart 216; Day 2 (outside witnesses) Thursday Jul 16 — Senate Judiciary Committee hearing notice, CNN live coverage, Jul 15.
- Sen. Cornyn: will decide at "the confirmation vote in about a week" — i.e., a committee vote is expected ~Jul 22–23, two clear windows before Aug 1 (CNN).
- Sen. Tillis "leaning yes," conditioned on Trump endorsing a legislative kill of the $1.8B "anti-weaponization" fund (CNN) — a demand Trump can satisfy with a Truth Social post; it costs him a program he can rebuild elsewhere.
- Margin context: Sen. Graham's death leaves committee Republicans effectively unable to lose a vote (Roll Call, Jul 14, NPR, Jul 15) — real risk, but it is the same risk the market has already marked down 6–8¢ for.
- Blanche is serving as acting attorney general (Deseret News, Jul 15) — the administration's urgency to regularize the role before recess is the schedule-forcing function.
Resolver profile (Stage 2.5). Resolver: Source Agency reporting that the committee "reports the nomination to the full Senate" — a discrete, well-covered procedural event with no definitional ambiguity (reporting favorably or without recommendation both qualify as "reports"; a tie fails and does not report). Mechanism: first occurrence before Aug 1. Priced deductions from a ~90% mechanical baseline: −12 pts — one GOP defection (Tillis's condition unmet, or Cornyn breaking) either fails the vote or, more likely, causes the chairman to pull it; −8 pts — calendar slip: if votes aren't locked by Jul 30's meeting the next window is past the deadline, and "delayed to September" resolves NO here even if Blanche is ultimately confirmed. Net: 70%.
Tail risks (cleanest way to lose first).
- Trump publicly refuses Tillis's anti-weaponization-fund demand; Tillis withholds; chairman shelves the markup past Aug 1. This is the single most probable loss path and it needs no dramatic news to happen — just two weeks of stalling.
- A second hearing round is demanded after Day-2 witness testimony, pushing the notice period past Jul 30.
- Adverse Epstein-files or pardons revelation between now and markup flips a second Republican outright.
Numbers (for the record — did NOT set size): edge +8¢ vs 62¢ signal (+6¢ vs 64¢ limit); EV ≈ +12.9% on cost at signal; max payout $300 on 300 contracts.
Liquidity / entry context: yes_bid 60 × 100, yes_ask 62 × 10; ask ladder 62×10 → 67×100 → 69×200, so only ~10 contracts fill instantly at ≤64 — this is a resting-limit entry against ~1.5k/day volume, not a sweep; vol24h 1,535; OI 6,057.
Price history (14d): thin until the hearing week — 65¢ (Jul 9), 68¢ (Jul 13–14), 65¢ (Jul 15), 60–62¢ today. The 48h move is toward our entry (wobble news priced in), not away from it.
Sizing: MEDIUM conviction caps cost at 7% of capital ($700); the book caps it lower. 300 contracts at a 64¢ limit = $192 max cost (1.9% of capital), well inside both the conviction cap and the 15% cluster cap.
4. Recommended $10,000 portfolio
One pick cleared every screen this week. That is the correct answer, not a failure: Politics this week is dominated by (a) resolver-snapshot markets I cannot observe, (b) a Hormuz complex that repriced violently intraday on data the resolver hasn't published, and (c) headline markets already trading at my number. Cash is a position — 98.1% held.
| Pick | Side | Limit | Contracts | Max cost | % capital | Conviction / band | Max payout | Edge ¢ | EV % |
|---|---|---|---|---|---|---|---|---|---|
| KXBLANCHEJUDICIARY-27-26AUG01 | BUY YES | 64 | 300 | $192 | 1.9% | MEDIUM · favorite (62–64¢) | $300 | +8 | 12.9 |
| Total deployed | $192 | 1.9% | $300 | ||||||
| Cash reserve | $9,808 | 98.1% | Held for opportunistic adds (see execution notes) | ||||||
Blended EV at model probabilities: +$18 on $192 deployed (+9.4% on cost at limit). Edge and EV are recorded for calibration only; size was set by conviction tier, entry band, and book depth — v3 is EV-agnostic by design because the resolved record showed stated EV was anti-predictive.
Cluster exposure (cap: 15% of capital per cluster)
| Cluster | Cost | % of capital | Cap | Status |
|---|---|---|---|---|
| blanche-ag-confirmation | $192 | 1.9% | 15% | OK — sole pick |
Correlation note against the live feed: this pick is independent of every ticker in $EXISTING_PICKS. The tempting World Cup add was rejected precisely because it would have doubled an existing feed cluster (see Section 5).
Conviction exposure
| Tier | Cost deployed | Per-pick cap | Picks |
|---|---|---|---|
| HIGH | $0 | $1,500 | 0 |
| MEDIUM | $192 | $700 | 1 |
| LOW | $0 — never deploys (v3.3) | — | 0 |
Risk profile
- Worst case: Blanche markup slips past Aug 1 or fails → −$192 (−1.9% of capital). No cluster can flip the book's sign; nothing else is at risk.
- Best case: committee reports Jul 23 → +$108 on cost (+56% on the position, +1.1% on capital), with early resolution freeing capital mid-window.
- Most likely: markup Jul 23 or Jul 30 reports the nomination on party lines after Trump gives Tillis his fund-kill endorsement; position pays $300.
- Concentration: single thesis, 1.9% of capital — concentration risk this run is effectively nil; the real "risk" is opportunity cost of the 98% cash book, which v3 explicitly accepts in efficient weeks.
Execution notes
- Order: resting limit BUY YES at 64¢ (take the 10 at 62, rest the balance). Do not chase above 64 — the ladder jumps to 67/69 and the edge at 69 (+1¢) does not survive fees. If unfilled after 3–4 days of ~1.5k daily volume, let it expire unfilled; a partial fill is an acceptable outcome.
- Upgrade trigger: a Judiciary executive-business notice listing the Blanche nomination for Jul 23, or Trump publicly endorsing the anti-weaponization-fund kill, converts this to ~85% — add toward the full $700 MEDIUM cap if the price is still ≤75¢.
- Invalidation: markup held over past Jul 30 without a scheduled date, a second Republican joining Cornyn in expressed doubt, or any "vote delayed to September" report — exit at market; the calendar is the position.
- Watchlist (no capital): KXHORMUZWEEKLY-26JUL19-T75 NO if it cheapens toward 35–38¢ once PortWatch publishes Jul 13–14 prints (the first two resolver days will collapse the variance); KXTRUMPBIBIMEET YES if a White House confirmation knocks it below ~78¢ intraday; the KXAPRPOTUS Jul 24 successor series if a fresh, verifiable RCP print lands within 24h of the snapshot.
5. What I rejected and why
Every Stage-2 survivor that didn't become a pick is machine-logged in picks.json and shadow-tracked to settlement — if the rejects outperform the picks, the screen itself is falsified, and that's a finding. Thirteen rejects this run, in four stories:
The Hormuz complex (5 rejects) — the entry-band screen doing its job
The resolver is IMF PortWatch daily transit calls for chokepoint6 (verified against the chokepoint reference table — chokepoint4, the obvious guess, is Bab el-Mandeb). The resolver's published series ends Jul 12: the market week (Jul 13–19) has zero published days. Run-rate into the week was 15, 11, 9, 14, 10/day (Jul 8–12). Since then, off-resolver evidence only: a live tracker logged a 41-vessel rush on Jul 14 ahead of the reinstated US blockade (effective 20:00 UTC Jul 14); Iranian missiles hit two VLCCs the same day; Iran re-declared the strait closed Jul 16; the US claims 7 transits Jul 16; 39 tankers are running dark — and dark ships don't appear in an AIS-based count. On that evidence I make the weekly sum ~45–70 central, and the whole Kalshi complex repriced violently today (T50: 43¢→85¢; T75: 22¢→54¢) on exactly this unpublished-data guesswork.
- T50 — my P(sum>50) ≈ 62% vs 82–85¢: the value side is NO at ~17¢, which is a sub-35¢ tail — hard reject, no exceptions (v3.2); the last Hormuz tail (KXMAXSHIPSHORMUZ AL50) lost 75% on tracker-vs-PortWatch slippage. Also fails the 48h-move screen outright.
- T75 — NO at ~46¢ shows my largest nominal edge of the run (model 67 vs 46, +21¢). Rejected anyway: coin-flip band requires HIGH conviction, and this cannot be HIGH — five of seven resolver days are unpublished, a US-escorted-convoy resumption (the US struck Iran Jul 15 explicitly "to protect vessels") could print 30+/day within the window, and today's informed flow is betting exactly that. This is the resolver-lag bet class that produced prior losses. Logged; the shadow track will score it.
- T45 — YES fair (model 87 vs 86¢) but ≤1¢ edge and 33 contracts of ask depth: fails the 3¢-slippage screen at any real size.
- PEAK-T15 (YES, model 79 vs 77¢) and PEAK-T20 (YES, model 52 vs 49¢) — both ride the single question of what PortWatch prints for the Jul 14 rush (historical tracker→PortWatch ratio ~55% would print ~22). Market is at my number on both; T20 is additionally coin-flip band.
The approval snapshots (5 rejects) — unobservable resolver
The KXAPRPOTUS buckets resolve to the literal RealClearPolitics average at 11:00 AM ET tomorrow; KXTRUMPVH to the VoteHub average at 10:00 AM. Both resolver pages returned HTTP 403 to every fetch attempt, and no Wayback snapshot exists for this week. The last public print I can cite is RCP 40.9 on Jul 14 (AOL/RCP) — but the 41.1–41.3 bucket has been bid from 24¢ to 44¢ over 48h by traders who are plainly watching the live page. Betting a snapshot market blind against page-watchers is not an edge in any direction, so all five legs are rejected on resolver-unobservability (Stage 2.5's screen), the VoteHub leg doubly so for its 6¢ spread on a 13×24-contract book.
The definitional trap (1 reject) — KXBILLSCOUNT-26JUL-0.0
The single most instructive market of the run. Title: "Will 0 to 0 bills become law in Jul 2026?" Rules primary: "If the President signs exactly 0 bills into law in Jul 2026…". On Jul 11 the 21st Century ROAD to Housing Act became law without Trump's signature after he cancelled the signing ceremony and declared he'd sign nothing until Congress passes the SAVE America Act (NPR, Jul 10). Under the title's reading NO has already won; under the rules' literal reading YES is alive and arguably cheap at 22¢ given an announced signing strike. When the two readings of the same contract point at opposite resolutions, the market is unpriceable at any conviction tier — Stage 3's "definitional dispute could flip the resolution" reject, verbatim. (Its 3¢→28¢→21¢ two-week path suggests the market hasn't agreed on the reading either.)
Efficiently priced / duplicate cluster (2 rejects)
- KXTRUMPBIBIMEET-26AUG01-AUG01 YES @ 83¢, model 84%. Israeli press reports Netanyahu departs Saturday and meets Trump Monday Jul 20 (Times of Israel), with the Lindsey Graham memorial as a second in-person path (Al-Monitor) — but a senior White House official says no visit is on the schedule (Axios), and the market already took the +7¢ repricing today on these same reports. ≤2¢ of edge is not deployable.
- KXTRUMPWORLDCUP-26-DJT-1 YES @ 94¢, model 95%. Rejected on cluster discipline, not probability: users' feeds already hold KXTRUMPATTEND YES (Trump attends the Jul 19 final) — "attends ≥1 match" with no attendance yet recorded is the same bet through a second market. Stacking near-perfectly correlated legs on one thesis is the exact two-Kash-Patel failure mode the 15% cluster cap exists to prevent, and the 1–2¢ of deep-favorite edge doesn't buy that correlation.
6. Sources
- IMF PortWatch — chokepoint monitoring portal; resolver data pulled directly from the PortWatch ArcGIS FeatureServer (
Daily_Chokepoints_Data,portid='chokepoint6'= Strait of Hormuz; chokepoint reference table used to verify the ID→strait mapping). Hormuz daily prints Jul 8–12: 15, 11, 9, 14, 10. - straits.live — Strait of Hormuz live tracker (Jul 14 pre-blockade rush; 39 dark tankers; US claim of 7 transits Jul 16; PortWatch counts AIS-broadcasting crossings only).
- hormuzstraitmonitor.com (Jul 14: 23 inbound / 18 outbound before the 20:00 UTC blockade; VLCC strikes; war-risk insurance 8× pre-crisis).
- CNN, Jul 15 — US strikes on Iran "aimed at protecting vessels in Strait of Hormuz"; Al Jazeera, Jul 10 — traffic plunges as fighting resumes; Wikipedia — 2026 US naval blockade of Iran.
- Senate Judiciary Committee — Blanche nomination hearing notice (Jul 15, 2026) and Day 2 notice (Jul 16).
- CNN, Jul 15 — Blanche hearing live updates (Cornyn timeline; Tillis condition); Roll Call, Jul 14; NPR, Jul 15; Deseret News, Jul 15.
- NPR, Jul 10 — housing bill becomes law without Trump's signature; signing-strike condition on the SAVE America Act.
- Axios, Jul 4 — Netanyahu requested WH meeting; WH official: visit not on schedule; Times of Israel — Saturday departure, Monday meeting expected; Al-Monitor — visit tied to Trump meeting and Graham memorial; Middle East Monitor, Jul 15.
- RealClearPolling — Trump job approval (page 403s to automated fetch; last public print 40.9 on Jul 14 via AOL); VoteHub approval average (also 403).
- Kalshi live market data and orderbooks:
api.elections.kalshi.com/trade-api/v2; market universe and rules text: Kalshi mirror read-only query API (trading_events,trading_markets).
Data sources: Kalshi mirror DB (markets, events, rules), Kalshi public trade API (live books, candlesticks), IMF PortWatch ArcGIS feed, and the primary news sources linked above. Audit run 2026-07-16 under methodology v3.3.
Disclaimer: all probabilities are subjective estimates by the auditing model, not statements of fact. Prediction-market contracts routinely resolve to zero; the entire cost of any position can be lost. Logged rejects are shadow-tracked for selection-skill analysis and are not recommendations. Nothing here is financial advice.