Events Desk

Kalshi event markets beyond sports — politics, courts, economics, world events.
Status: backtest phase. No live trading until an edge is proven on history.

Alert Queue

2 new · 2 open · 1 taken · 2 dismissed

Flag-for-review blotter — a human triages every alert; nothing auto-trades. Ranked by conviction. Click a row to review the news, reasoning, and suggested action.

27
Will the United States acquire any part of Greenland before 2029?
Logic check · resolution-rules edge · Politics · edge ~16.5pp
Logic new
Suggested action
BUY NO — fade the elevated YES
fair value est. ~5% vs market 22%
Signal
Resolution-rules mispricing (no price move required)
Potential ann. return (if right)
+11% · resolves in 926d
Projected ann. return (prob-weighted)
+8% · vs S&P ~10%
[Blind LLM estimate, knowledge cutoff ~Jan 2026 — VERIFY against current news before trading; large edges on fast-moving markets (appointments, Fed path) are often the model lacking recent facts, not a real mispricing.] Despite unusually elevated political rhetoric (a US administration has repeatedly floated acquiring Greenland), the actual chain of events required for YES is very demanding: Greenland is an autonomous territory of Denmark, both of which have firmly and repeatedly rejected any sale or transfer. Formal US governance or jurisdiction over any part requires either Denmark/Greenland's consent (a treaty, cession, or annexation agreement) or an act of force, both of which face enormous legal, political, and NATO-alliance obstacles within this ~2-year window. However, the secondary rule materially raises the probability versus a pure completed-acquisition market: a mere joint announcement of intent counts, which is more achievable amid ongoing pressure and negotiations over defense/base arrangements — though leasing explicitly does not count, and most plausible near-term deals resemble expanded basing/defense access rather than formal jurisdiction transfer. [Resolution nuance: The secondary rule is the trap in both directions. YES can trigger on just an ANNOUNCEMENT by both the US and the controlling entity that a transfer will happen — not a completed acquisition — which raises odds above a naive read. But it must be about formal US governance/jurisdiction (state/territory/other); a lease, a defense-cooperation pact, or expanded military basing (the most likely concrete outcome) does NOT count. So loud rhetoric or a base deal short of a jurisdiction-transfer commitment resolves NO.]
Open on Kalshi ↗
23
Will Trump buy Greenland?
Logic check · resolution-rules edge · Politics · edge ~14.5pp
Logic new
Suggested action
BUY NO — fade the elevated YES
fair value est. ~5% vs market 20%
Signal
Resolution-rules mispricing (no price move required)
Potential ann. return (if right)
+10% · resolves in 925d
Projected ann. return (prob-weighted)
+7% · vs S&P ~10%
[Blind LLM estimate, knowledge cutoff ~Jan 2026 — VERIFY against current news before trading; large edges on fast-moving markets (appointments, Fed path) are often the model lacking recent facts, not a real mispricing.] Purchasing sovereign territory from an ally is an extraordinarily rare event — the US has not completed a territorial purchase since the Virgin Islands (1917), and no NATO member has sold populated territory in the modern era. Despite Trump's repeated public interest and 2025 pressure campaign, Denmark, Greenland's autonomous government, and Greenland's ~57,000 residents (who have consistently opposed sale and favor eventual independence, not US annexation) all firmly reject a sale. Even a "part of Greenland" purchase would require Danish parliamentary approval, Greenlandic consent, and a complex international negotiation over a ~4-year window — a long chain where each link is unlikely. The more probable paths to expanded US presence (military bases, mineral deals, association agreements) would NOT satisfy the resolution criterion of an actual purchase from Denmark. [Resolution nuance: Resolution requires an actual PURCHASE of at least part of Greenland FROM DENMARK. A defense/basing agreement, a critical-minerals deal, a "free association" compact, or Greenland independently choosing US alignment would NOT resolve YES — only a completed sovereign land transaction paid to Denmark counts. Also note the buyer is "the United States," not Trump personally, and the deadline is the exact 2029 inauguration date.]
Open on Kalshi ↗
95
Will the Fed cut rates 0 times?
Logic check · resolution-rules edge · Politics · edge ~60.8pp
Logic taken
Suggested action
BUY NO — fade the elevated YES
fair value est. ~15% vs market 76%
Signal
Resolution-rules mispricing (no price move required)
Potential ann. return (if right)
+656% · resolves in 174d
Projected ann. return (prob-weighted)
+526% · vs S&P ~10%
[Blind LLM estimate, knowledge cutoff ~Jan 2026 — VERIFY against current news before trading; large edges on fast-moving markets (appointments, Fed path) are often the model lacking recent facts, not a real mispricing.] Entering 2026 the Fed carried an easing bias, with its own dot-plot projections and market consensus pointing to at least one 25bp cut during the year. Zero cuts across all ~8 FOMC meetings requires a sustained "higher for longer / no easing" year, most plausibly driven by an inflation resurgence (e.g. tariff-driven) keeping the Fed on hold. That is a genuine tail scenario but not the central case, so the probability sits in the low range. [Resolution nuance: YES requires zero cuts across the ENTIRE calendar year 2026, not a snapshot. Rate HIKES or holds do not count against YES — only cuts (25bp=1) do. Cuts made in 2025 are irrelevant. A single 25bp cut at any 2026 meeting flips this to NO, so a casual reader underrates how a full-year hold (inflation scenario) yields YES.]
Open on Kalshi ↗
95
Who will be Trump's next Attorney General?
Logic check · resolution-rules edge · Politics · edge ~73.5pp
Logic dismissed
Suggested action
BUY NO — fade the elevated YES
fair value est. ~4% vs market 78%
Signal
Resolution-rules mispricing (no price move required)
Potential ann. return (if right)
+136% · resolves in 925d
Projected ann. return (prob-weighted)
+129% · vs S&P ~10%
[Blind LLM estimate, knowledge cutoff ~Jan 2026 — VERIFY against current news before trading; large edges on fast-moving markets (appointments, Fed path) are often the model lacking recent facts, not a real mispricing.] The "current" AG as of this market is Pam Bondi, confirmed February 2025 as Trump's first-term-2 AG. For YES, Bondi must leave (or be replaced) AND the very next Senate-confirmed replacement must be specifically Todd Blanche, all before Jan 20, 2029. Blanche is currently Deputy Attorney General — a plausible successor, which is why this market exists at all, but two low-probability links must both hit: (1) an AG vacancy occurring in this window, and (2) Blanche being the specific confirmed pick over all other candidates. Cabinet AGs frequently serve full terms; even when they turn over, the successor is one of many possibilities. Blanche's DAG role gives him above-random odds if a vacancy opens, but the compound probability of vacancy × specifically-Blanche × Senate-confirmation remains low. [Resolution nuance: Acting/interim holders do NOT count — Blanche has served as Acting AG in various capacities, but the market requires him to be the next SENATE-CONFIRMED AG. A casual reader might think Blanche briefly running the department resolves YES; it does not. Also, "next" is strict: if anyone other than Blanche is confirmed as AG first, this resolves NO permanently even if Blanche later gets the job.]
Open on Kalshi ↗
20
Will the Federal Reserve Hike rates by 0bps at their July 2026 meeting?
Logic check · resolution-rules edge · Politics · edge ~18.5pp
Logic dismissed
Suggested action
BUY NO — fade the elevated YES
fair value est. ~62% vs market 80%
Signal
Resolution-rules mispricing (no price move required)
Potential ann. return (if right)
>+999% · resolves in 19d
Projected ann. return (prob-weighted)
>+999% · vs S&P ~10%
[Blind LLM estimate, knowledge cutoff ~Jan 2026 — VERIFY against current news before trading; large edges on fast-moving markets (appointments, Fed path) are often the model lacking recent facts, not a real mispricing.] This is a mutually exclusive bucket market where "Hike of 0bps" simply means the Fed holds rates unchanged (no cut, no hike). The relevant question is: at the July 2026 FOMC meeting, does the Fed hold, cut, or hike? Given my January 2026 knowledge cutoff, I cannot observe the actual rate path through mid-2026, so I must reason from base rates. Historically, the Fed holds at the large majority of individual meetings, but in an easing cycle (which the Fed appeared to be in during 2024-2025), the mix shifts toward cuts at some meetings. A hike is very unlikely absent a major inflation resurgence. The realistic contest is "hold" vs "cut," with "hold" being the modal single-meeting outcome but a cut being a meaningful minority possibility during an active cutting cycle. [Resolution nuance: The title says "Hike rates by 0bps," which a casual reader might find confusing, but the secondary rules clarify it resolves YES only if the Fed leaves rates fully unchanged — any cut OR any hike resolves NO. So this is effectively a "no change / hold" bucket. The word "hike" in the title is misleading; it does not mean an actual rate increase is required.]
Open on Kalshi ↗

Thesis

A — News overreaction: when news spikes an event market, the crowd anchors and under-corrects. Archetype: a market at 14% jumps to 35% on a false report, then sticks at 28% instead of reverting. We fade the incomplete reversion — model-free at the trigger, wins on judgment and patience, not speed.

B — Generalist logic check: some markets are mispriced on pure reasoning, no price move required — the crowd misreads resolution rules or overrates a weak case. Proven twice: the Powell governor-vs-chairman misread, and Musk v. OpenAI (obvious loser, priced like a coin flip). An LLM agent sweeps all open markets for these, skipping well-mined ones (e.g. Trump tweets) where institutional models have squeezed the edge.

Discipline: the #1 lesson from the golf desk — backtest before any money. Every stage below has a go/no-go gate. Flag-for-review only; never auto-trade in v1.

Roadmap

Stage 0 — Foundations
DONE
Events Desk page live on dashboard
Roadmap + execution flow published, phone-accessible at /events
DONE
Candlestick retention probe
FINDING (07/02): Kalshi serves full price history for open + recently-settled markets, but purges settled markets entirely ~2-3 months after settlement. A pure 12-month lookback is impossible via the API — so we archive from today and backtest the retrievable window
DONE
Kalshi archiver job
Weekly Cloud Run job harvesting Politics / World / Economics markets + candlesticks into GCS (gs://mates-events-archive) before Kalshi purges them. 14,000+ markets archived so far; every week the backtest window grows
DONE
Research brief: is the thesis documented?
Literature sweep complete — overreaction + incomplete reversion is real but CONDITIONAL on news type: false/stale news reverts, genuine news drifts (do not fade). Full brief in events_desk/research/
Stage 1 — Prove it (backtest, no money)
DONE
Historical spike scanner
DONE — 6,143 volume-backed price spikes detected across 12,900 Politics/Economics markets. First-pass: median spike gives back ~40% of its move within 24h (incomplete reversion), but 59% were directionally correct about the outcome. 21% fully revert, 28% hold — the split the news layer must explain. Dataset: events_desk/data/spikes.csv
DONE
News matching + labeling
DONE — news-volume surge for all 2,890 spikes via GDELT's BigQuery dataset (one ~2GB query; the HTTP API throttled too hard for full coverage). Every spike tagged quiet / mild / surge by news volume around it.
GO / NO-GO
Edge report (preliminary)
KEY FINDING: news-VOLUME gives no edge — reversion is flat across quiet/surge (~+0.42). But split by hindsight outcome, the signal is huge: spikes that were WRONG reverted +0.64-0.98, spikes that were RIGHT held ~+0.20. So the tradeable question is predicting wrong-vs-right AT SPIKE TIME — which needs false-vs-real news classification (an LLM reading headlines), not volume. That LLM pass is the decisive next test.
PLANNED
LLM false/none/real classification
NEXT + DECISIVE — LLM reads headlines around each surge spike, classifies false / stale / real. Test: do false/stale-news spikes revert (they should be the "wrong" spikes) while real-news spikes hold? If yes -> that IS the tradeable predictor of wrongness. If no -> shelve the overreaction fade.
Stage 2 — Live detector (only if Stage 1 clears)
PLANNED
Real-time monitor
Same detector on live markets; two-stage: alert on spike, confirm on the incomplete-reversion plateau
PLANNED
Alert queue
Ranked flagged opportunities in Cloud SQL, surfaced on this page
Stage 3 — Operate
PLANNED
Alert review workflow
Per-alert detail: price history, news, LLM reasoning summary. Human reviews + trades manually — never auto-trade in v1
PLANNED
Feedback loop
Outcomes recorded against alerts; thresholds and prompts recalibrated
Track B — Generalist logic-check scanner (LIVE)
DONE
Blind LLM market sweep
DONE (v1) — fresh evaluator per market sees ONLY title + resolution rules (no price, no position), returns an independent probability. Blind test on our own trades: it INDEPENDENTLY reproduced the Powell governor-seat-vs-Fed-chair thesis (p=5%) and the Alito base-rate fade (p=2%) with zero knowledge we held those positions. Strong validation of the method and the resolution-nuance edge.
DONE
Candidates feed the alert queue
DONE — actionable edges (>=7pp divergence) written to the queue above as logic-check alerts, human-triaged.
DONE
News-aware evaluator (v2) — PROTOTYPE VALIDATED
The blind LLM (cutoff ~Jan 2026) false-positived on news-driven markets. Fix = give the evaluator WEB SEARCH. Proven on the two worst false positives: next-AG blind 4% -> news-aware 82% (market 78%); Fed-cuts blind 15% -> news-aware 80% (market 76%). Both converged to the market, erasing the fake edge. Next: make news-aware the default Track B evaluator and re-sweep a broader, shorter-dated market set to hunt real edges.
DONE
Annualized return metrics
DONE — every alert shows Potential (if-right) and Projected (prob-weighted) ANNUALIZED return + days-to-resolution, so long-dated low-yield markets (e.g. Greenland 2029 fades annualize to only ~8-11%) are visibly filtered vs the S&P ~10% bar.
Backlog — Trump mention-market models
PLANNED
Approaches #6 resolution-rules, #1 base-rate, #2 topicality, #3 LLM forward-only
Per events_desk_plan.md — parked behind the overreaction detector; shares the candlestick + GDELT harness

Execution Flow

TRIGGER A — OVERREACTION TRIGGER B — LOGIC CHECK Kalshi price feed poll open event markets Spike detector z-score + volume anomaly News pull GDELT + LLM summary Classify news false / none / real Plateau confirm incomplete reversion sticks LLM generalist sweep all open markets, no price trigger Reasoning-edge candidates resolution misreads, weak cases ALERT QUEUE ranked, stored in Cloud SQL Human review phone / dashboard — Michael Manual trade on Kalshi never auto-trade in v1 Outcome logged recalibrate thresholds + prompts FEEDBACK