Loading…
Every Monday the desk cultivates one index from the live cross-venue board and publishes it here with the reasoning that produced it — and, beside it, the recognition of how the previous week's index actually did. The recognition is written whether the week went well or badly; a ritual that only publishes its wins is marketing, and the only thing worth building here is a record.
The weekly index is always a slate, never a benchmark. A benchmark carries versioned methodology and a 14-day consultation window before any rule change, so one cannot be minted weekly without destroying what makes that class dependable. See the index board for both classes.
10 elections markets carrying the "nominee" axis across gemini, polymarket, every one resolving inside two weeks, tracked as one composition.
Inception lock pending. This index has been published but its entry prices have not been struck yet, so it has no score. The price the desk saw while composing is deliberately not used in its place — it is a different price, observed at a different moment.
Several specialist scorers, each with one narrow published lens, combined by a gate that weights them and filtered by a judge. It is a mixture-of-experts pattern, not a neural one, and nothing here learns yet — the gate weights are fixed and published, and become adaptive only once the accuracy scorecard carries enough graded outcomes to justify it.
| Lens | Weight | Score |
|---|---|---|
| Catalyst All 10 dated members resolve between 2026-08-18 and 2026-08-25 — a 7-day window a median 5 days out. No member parses into a subject this engine can name, so what is measured here is a shared RESOLUTION WINDOW and not a shared catalyst — venues write annual contracts onto one expiry, and a date span alone cannot tell that calendar artefact apart from a single dated event. The reading is held at 0.5 until 50% of the composition shares a subject. | 0.29 | 0.50 |
| Liquidity 10 of 10 members clear the 250-unit volume floor and the basket's median volume is 31586, giving a depth factor of 1. | 0.21 | 1.00 |
| Novelty No published index — 33 of them, benchmarks included — shares this composition's vocabulary or would admit its members under its own component rule. | 0.21 | 1.00 |
| Coherence 10 of 10 members sit in `elections` and 90% share the term "nominee" — the basket names itself. | 0.29 | 0.95 |
| Divergence abstained: None of the 10 members parses into a structural claim, so no cross-venue comparison was attempted. identity.ts recognises price thresholds, price ranges and rate decisions and deliberately refuses everything else; a score of 0 here would report venue agreement that was never tested. | — | — |
| Calibration abstained: 0 graded outcomes match this composition's market family, against a floor of 30. The whole graded set is 212 observations and 2 of its 10 decile buckets clear the 30-outcome bar, so there is no calibration signal to report here — pooled or per family. This lens abstains rather than manufacturing one from a handful of points. | — | — |
The refusals are published because they are the evidence the method is doing work. A ranked list nobody can interrogate is an oracle, not a method.
hypothetical · mid marks · before fees, spread and slippage
No score published — no inception lock was ever struck for this release, so there is no basis to measure from. The composition-time mid is not a substitute: it is a different price, observed at a different moment.
Primary — Elections, inside two weeks cannot be scored for this window — no inception lock was ever struck for this release, so there is no basis to measure from. The composition-time mid is not a substitute: it is a different price, observed at a different moment.
These are the contracts that repriced. W.E.T. records prices, not reasons: no explanation of why any market moved is offered here, and none should be inferred from the ordering.
No week has been scored yet. Rates appear at 12 scored weeks; until then there is nothing to average.
Claimed: All 8 dated members resolve between 2026-08-11 and 2026-08-18 — a 8-day window a median 7 days out. No member parses into a subject this engine can name, so what is measured here is a shared RESOLUTION WINDOW and not a shared catalyst — venues write annual contracts onto one expiry, and a date span alone cannot tell that calendar artefact apart from a single dated event. The reading is held at 0.5 until 50% of the composition shares a subject.
The index never locked an inception price, so no lens can be graded against a return.
Claimed: 8 of 8 members clear the 250-unit volume floor and the basket's median volume is 43698, giving a depth factor of 1.
The index never locked an inception price, so no lens can be graded against a return.
Claimed: Closest published index is WELW (W.E.T. Election Week Index, slate) at 2% overlap by thematic vocabulary overlap, measured across all 33 published indices.
The index never locked an inception price, so no lens can be graded against a return.
Claimed: 8 of 8 members sit in `elections` and 100% share the term "primary" — the basket names itself.
The index never locked an inception price, so no lens can be graded against a return.
Claimed: abstained: None of the 8 members parses into a structural claim, so no cross-venue comparison was attempted. identity.ts recognises price thresholds, price ranges and rate decisions and deliberately refuses everything else; a score of 0 here would report venue agreement that was never tested.
The index never locked an inception price, so no lens can be graded against a return.
Claimed: abstained: 0 graded outcomes match this composition's market family, against a floor of 30. The whole graded set is 85 observations and 0 of its 10 decile buckets clear the 30-outcome bar, so there is no calibration signal to report here — pooled or per family. This lens abstains rather than manufacturing one from a handful of points.
The index never locked an inception price, so no lens can be graded against a return.
Half the panel is marked not gradable on purpose. Whether a catalyst landed, whether a cross-venue divergence was the one that mattered — those are claims about the world, and this engine records prices. Marking them anyway would be inventing the grade.
| Week | Index | Score |
|---|---|---|
| 2026-08-10 | Primary — Elections, inside two weeks | unscored |
| 2026-08-03 | Seat — Elections, inside six months | unscored |
No week has been scored yet. Rates appear at 12 scored weeks; until then there is nothing to average.
The score is the percentage move of an equal-weight composition, marked at two-sided mid prices and oriented to each member's declared side. It is size-blind by construction — identical whether a reader follows it with a little or a lot — which is what makes one week comparable to every other week, and what keeps it from being a profit-and-loss figure. There are no currency amounts on this page and no field that could carry one.
A member that stops quoting is never marked to zero — it is excluded and the coverage figure says so. A member the venue voided is not a loss: the question was never graded, so it leaves the basket. Below a 60% coverage floor no score is published at all, because an index scored on the half of its members that stayed quotable is a different index.
W.E.T. is not an exchange and accepts no orders. Every member trades individually on its own venue, this combination is a published view rather than an offered instrument, and nothing here is advice. hypothetical · mid marks · before fees, spread and slippage.