Generated 2026-09-20 11:42 UTC (1.0 h ago; current daily snapshot) · CME V0.1 · methodology · raw JSON · audit log
Where linked markets disagree in the current UTC run
Track record immatureCoherence Signals answer one practical question: where do related markets disagree enough that a serious user should inspect the price before citing or trading it? The latest Coherent Markets Engine run (CME V0.1) checked 1,477 Polymarket markets against 313 linked-market rules, from a venue read this run could not show it finished, so treat the count as what was checked and not as a venue total, then emitted 1 signals after significance, cost, and resolution-risk gates. Modeled net EV for the surviving set is <$250; realized PnL still waits on settlements. This snapshot is explicitly versioned as CME V0.1. The V0.2.0 walk-forward methodology is separately tracked in manifest #288046 and should only be cited for V0.2 outputs.
Track-record caveat
Related-market consistency checks stay experimental until markets resolve. Positive realized-net rate, realized PnL, and matched controls govern any performance claim.
What it is
A structural-disagreement ledger for linked markets.
What it is not
Not a profit promise or trade recommendation.
Evidence gate
Settlements and matched controls decide stronger claims.
Next build
Depth, capacity, and control-arm reporting tighten EV realism.
Study
Use this feed to find linked markets whose prices deserve review, not blind execution.
Monitor
Researcher keeps event identities, signal deltas, and daily digest context together.
Wait for outcomes
Stronger performance claims wait for resolved signals and matched controls.
Today's coherence signals
Each row is a detected linked-market disagreement, redacted to its characterization. Event identities, the full archive, signal deltas, and trade construction are in the Researcher feed.
Type
Cme Structural
Tension
mild
Horizon
30-90d
Event identity redacted on the free tier. Upgrade to Researcher for full event ID + trade construction + day-over-day deltas.
| Signal | ||||||
|---|---|---|---|---|---|---|
| Signal #1 | mene | 3 | p<0.001 | <2% | <=90d | <$250 |
Run mechanics, venue health & provenance
Polymarket coherence-feed health (30d)
82/100· Healthy
generated 2026-09-20
25 of 150 conditions
57 signals
23 of 20 active days
251 matured
This score reflects data feed health, not market quality. It is based on coverage, volume, and activity; the resolution component is held back until our pre-registration tracking window matures. See pre-registrations.
How the run narrowed today's markets
What is on this page
- Daily three-gate signal counts + funnel summary (counts only)
- Per-signal aggregate characterizations (constraint type, tension magnitude, p-value, net EV%, est PnL) - event identities redacted
- Methodology + hash-chain audit log; V0.2 outputs are tracked under manifest #288046 with public receipt not yet verified
What the Researcher tier ($49/mo) adds
- Event identities (which Polymarket markets each signal references)
- Every signal in the run (1), not just the 8 shown here
- Trade construction (which markets to long/short, sizing, capital lockup)
- Day-over-day signal delta (which signals appeared, persisted, resolved)
- Search across the full dated archive (signals from any past day)
- Daily email digest (Resend; opt-in)
- 5-15 min delay relative to cron generation
The Coherent Markets Engine methodology is documented in full at /research/coherent-markets-engine-v1. Current snapshot version: CME V0.1. The V0.2.0 90-day walk-forward backtest is tracked in manifest #288046 (filed 2026-04-29; public receipt not yet verified; methodology frozen at commit 8616a63). The CFTC ANPRM comment on probability-axiom-feasible price coherence as a regulatory transparency standard was filed at regulations.gov on 2026-04-29 (docket CFTC-2026-0331-0001, comment tracking mok-lrhf-09it).
Audit-log row 4661b96605753205… cryptographically links the daily input artifacts (gamma metadata to the published outputs. The full hash chain is at /research/cme/audit_log.jsonl and verifies via python -m services.api.scripts.cme_v0_1.provenance verify.
This work used AI tools (Claude, GPT-4) as research aids during methodology design and pre-publication review. No claim is taken as true on the basis of an AI tool's output. Archive note 2026-08-13: the repository ships a rolling 14-day window of dated snapshots plus the latest report (cron output at apps/web/public/research/cme/); the full dated archive is retained in Convexly's R2 store, each object size-verified against git before the in-repo tail was pruned, and the hash-chain audit log remains committed. Recent results are recomputable from the shipped window with the documented seed; older snapshots are available from the archive on request.