# Polymarket 2026 US midterm scorecard: run 1

Vultax Research, Tallinn. Run date 2026-11-05. Questions and corrections: research@vultax.com.

Run 1 is the first cut the protocol set for 4 to 6 November 2026: "a first cut labelled by status, using called and unofficial-complete results where they exist; contracts without them are pending" (rules-v1.json; METHODS.md, both registered on OSF at https://osf.io/jwerm on 27 October 2026). Later dated runs add certifications, recounts and runoffs in their own folders and never rewrite this one, the checkpoint captures or the manifest. No certified result is included in this run.

## What was scored

- Manifest: the frozen v1 manifest, 13,186 contracts, SHA-256 `5b607bded8bd47c8f41ec083f54d6319d7178027c7914c64ab77f069c4a79dca` (unchanged, in the folder above).
- Prices: primary checkpoint 2 November 2026, 23:59:59 UTC, from `capture-2026-11-02T23-55-01-605Z.ndjson.gz` (finished 2026-11-02T23:59:44.833Z, 26368/26368 order-book requests answered, SHA-256 `76e65a10fa28dde053935d5ccd59be13330a88d165a1162a587e8d90070a3eab`). Secondary checkpoint 27 October 2026, 23:59:59 UTC, from `capture-2026-10-27T23-50-02-612Z.ndjson.gz` (finished 2026-10-27T23:55:11.493Z). The scorer picks both by the registered rule; `receipts/all-captures.json` lists every capture so the choice can be checked.
- Venue record (Polymarket's own payout): read from one capture taken with the frozen capture code after the vote, `receipts/capture-2026-11-05T14-51-27-229Z.receipt.json`, by `code/build-venue-outcomes.mjs`.
- Official record (state and territorial election authorities): `outcomes/race-worksheet-run-1.csv`, one row per Tier A race, read from each state's own results publication (retrieved 14:56 to 15:18 UTC on 5 November 2026), with source URL and retrieval time; turned into `outcomes/official-run-1.json` by the unchanged `code/build-race-worksheet.mjs`.
- Code: `code/score.mjs` is the frozen scorer, SHA-256 `41e938b25dcd8fbc8a166c41c9dc91730f1217a5402b3c073c78125730ecc667`, the same bytes as the copy registered before the vote, run unchanged at 2026-11-05T16:13:00.521Z. The run-1 tables below come from `code/protocol-tables-run-1.mjs`, which reads the frozen scorer's own rows and applies the registered eligibility rule (next sections).

## Status of the official record, by race (510 Tier A races)

pending 314, called 196.

- **called**: the state's unofficial count shows a leader and the Associated Press has called the race for that candidate. AP calls and their times were read at 15:23 UTC on 5 November 2026 from the New York Times election results feed (static01.nyt.com), which carries AP's race calls; a call made only by another desk does not count. An AP call is an interim status, never the official record.
- **unofficial_complete**: the state says its unofficial count is complete, or lists the winner as unopposed.
- **runoff_pending**: the state requires a majority and a runoff follows.
- **pending**: no AP call yet, a ranked-choice round still to run, the state's count leader and the AP call differ (HOU-IN-03), the state publishes no count yet, or the state's results could not be read under its own terms or past its bot protection (we do not get past bot walls). The note column gives the reason per race.

States and territories with no count in run 1 (every race pending): AK, AR, AZ, CA, GU, HI, IL, KS, KY, LA, MA, ME, MI, MO, MS, NE, NH, NJ, NM, NV, OH, OK, PA, TN, WI (25).

Chamber control (CTRL-SENATE, CTRL-HOUSE, CTRL-BOTH) is not published by any election authority; it follows from the seats.

## Which contracts run 1 scores

The registered rule: the official record decides the outcome wherever it can; "where only the venue record exists (a question no election authority answers ...), o comes from the venue record and the row is flagged venue_only"; and "a contract without an outcome at a scoring run is pending ... excluded from the means of that run" (rules-v1.json, outcomes).

Run 1's tables follow it. A contract scores on the official record. The venue record decides only for chamber control, the one Tier A question no election authority answers. Every other contract without an official result is pending in run 1, with Polymarket's payout shown beside it. Tier B contracts (margins, seat counts, turnout, county winners, ballot measures, other offices) have no official rows yet, and which of them only the venue can settle has not been classified, so all of Tier B is pending in run 1. `run-1-protocol-contracts.csv` gives every contract's basis:

- pending_no_outcome: 7,272
- pending_official_result: 2,452
- official: 1,527
- tier_c_not_scored: 1,380
- pending_tier_b_not_classified: 532
- venue_exception_chamber_control: 23

Scored under this rule: 400 contracts (venue_exception_chamber_control 8, official 392). Contracts that have an outcome but no two-sided quote at the checkpoint have no price and are not scored.

## Scores

Brier score per contract, (p − o)², computed at full precision. Race mean: each race's contracts averaged first, races weighted equally. Brackets: 90% cluster bootstrap over races, 2,000 resamples, seed 20261103, the generator and percentile positions of `score.mjs`. Log loss with p clipped to [0.01, 0.99]. Baseline: a constant 0.5 scores 0.25.

| Set | Contracts | Races | Brier, contract mean (90%) | Brier, race mean (90%) | Log loss, p clipped to [0.01, 0.99] (90%) | Share resolved Yes |
|---|---:|---:|---|---|---|---:|
| Tier A (no placeholders) | 400 | 199 | 0.0152 (0.0075 to 0.0239) | 0.0151 (0.0074 to 0.0239) | 0.0694 (0.0498 to 0.0915) | 0.4975 |
| Tier A without wide quotes | 400 | 199 | 0.0152 (0.0075 to 0.0239) | 0.0151 (0.0074 to 0.0239) | 0.0694 (0.0498 to 0.0915) | 0.4975 |
| Tier B | 0 | 0 | – | – | – | – |
| All scored | 400 | 199 | 0.0152 (0.0075 to 0.0239) | 0.0151 (0.0074 to 0.0239) | 0.0694 (0.0498 to 0.0915) | 0.4975 |

The 27 October checkpoint on matched questions (contracts priced at both checkpoints):

| Set | Matched contracts | Brier at 2 Nov (90%) | Brier at 27 Oct (90%) |
|---|---:|---|---|
| Tier A (no placeholders) | 399 | 0.0152 (0.0076 to 0.0240) | 0.0184 (0.0100 to 0.0281) |
| Tier A without wide quotes | 399 | 0.0152 (0.0076 to 0.0240) | 0.0184 (0.0100 to 0.0281) |
| Tier B | 0 | – | – |
| All scored | 399 | 0.0152 (0.0076 to 0.0240) | 0.0184 (0.0100 to 0.0281) |

Most Tier A contracts were priced near 0 or 1. When the expected side wins, such a contract scores close to 0 and pulls the mean down; when it loses, it scores close to 1. The favourites table and the calibration bins show performance by price band, and one election supplies limited independent information. Apart from the three chamber-control races, the races scored in run 1 are the ones whose state count could be read and that AP had called, which is not a random sample of races.

## Favourites, Tier A

For each Tier A race the highest-priced contract at the primary checkpoint (never a placeholder). 510 races, 199 scored: 194 won, 5 lost. Brier over scored favourites 0.0154 (0.0077 to 0.0241).

| Favourite's price | Favourites | Won | Lost | Pending |
|---|---:|---:|---:|---:|
| 0-10¢ | 0 | 0 | 0 | 0 |
| 10-20¢ | 0 | 0 | 0 | 0 |
| 20-30¢ | 0 | 0 | 0 | 0 |
| 30-40¢ | 0 | 0 | 0 | 0 |
| 40-50¢ | 0 | 0 | 0 | 0 |
| 50-60¢ | 15 | 0 | 2 | 13 |
| 60-70¢ | 23 | 1 | 1 | 21 |
| 70-80¢ | 27 | 6 | 1 | 20 |
| 80-90¢ | 44 | 12 | 1 | 31 |
| 90-100¢ | 401 | 175 | 0 | 226 |

## Calibration, Tier A

| Price bin | Contracts | Mean price | Share resolved Yes |
|---|---:|---:|---:|
| 0-10¢ | 176 | 0.023 | 0.000 |
| 10-20¢ | 15 | 0.136 | 0.067 |
| 20-30¢ | 6 | 0.268 | 0.167 |
| 30-40¢ | 2 | 0.345 | 0.500 |
| 40-50¢ | 2 | 0.465 | 1.000 |
| 50-60¢ | 2 | 0.525 | 0.000 |
| 60-70¢ | 2 | 0.655 | 0.500 |
| 70-80¢ | 7 | 0.731 | 0.857 |
| 80-90¢ | 13 | 0.863 | 0.923 |
| 90-100¢ | 175 | 0.974 | 1.000 |

## Calibration, all scored contracts

| Price bin | Contracts | Mean price | Share resolved Yes |
|---|---:|---:|---:|
| 0-10¢ | 176 | 0.023 | 0.000 |
| 10-20¢ | 15 | 0.136 | 0.067 |
| 20-30¢ | 6 | 0.268 | 0.167 |
| 30-40¢ | 2 | 0.345 | 0.500 |
| 40-50¢ | 2 | 0.465 | 1.000 |
| 50-60¢ | 2 | 0.525 | 0.000 |
| 60-70¢ | 2 | 0.655 | 0.500 |
| 70-80¢ | 7 | 0.731 | 0.857 |
| 80-90¢ | 13 | 0.863 | 0.923 |
| 90-100¢ | 175 | 0.974 | 1.000 |

## Multi-outcome events and the normalisation sensitivity

Registered as secondary tables: the multi-class Brier sum for negRisk events, and a sensitivity that divides each Yes midpoint by the event's sum when that sum lies between 0.90 and 1.10. An event qualifies only when every one of its outcomes has a price and a run-1 outcome. In run 1 no event qualifies: 1,079 events have at least one outcome without a two-sided quote (often a lettered placeholder) and 70 have every outcome priced but at least one outcome pending. Both tables are therefore empty in this run (events 0, normalised contracts 0) and are computed again in later runs.

## Markets found outside the frozen manifest (supplements s27 and s02)

The rules record markets Polymarket listed after the freeze, found by each checkpoint capture's re-sweep, as separately versioned supplements, never merged into version 1. The capture code's rule is broader: it records every market it finds outside the frozen manifest, either in an event the manifest already holds or under the re-swept `midterms` tag. `supplements/supplement-s27.csv` lists the 326 markets found by the 27 October checkpoint capture and `supplements/supplement-s02.csv` the 666 found by the 2 November one, with Polymarket's displayed price, best bid and ask as its catalogue reported them.

- By creation date: s27 created before freeze found later 38, created after freeze 288; s02 created after freeze 618, created before freeze found later 48. A market created before the freeze and found later was not shown to be listed after the freeze: when it entered the swept tags is not recorded. It stays out of version 1.
- Payout: the capture kept no resolution record for these markets, so no payout is asserted. The files give only what the 5 November capture saw: whether the market was closed and its catalogue prices (s27: closed prices binary unverified 263, not closed 63; s02: not closed 159, closed prices binary unverified 507).
- They are listed, not scored: the capture recorded each market's catalogue record but not its order book, so the registered price (the order-book midpoint) does not exist for them, and they carry no tier, which was assigned by hand before the freeze for version 1 only. Scoring them on another price would depart from the protocol, so it is not done.

## Official record and venue record

Where both records exist they agree on every Tier A contract: no row is flagged `records_differ`.

The frozen race matcher in `score.mjs` reads a party word or a surname. Polymarket labels many contracts "Name (D)" or "Name (R)", which it cannot read (the last word is the party letter). Those contracts carry an explicit outcome in `outcomes/official_by_market-run-1.json` (356 rows), the record outcomes/README.md set aside before the vote for "any Tier A contract the race matcher cannot read". Rows labelled "Other" and lettered placeholders are left to the matcher, which gives "Other" no official outcome; none of those rows has a price at the checkpoint. `code/merge-official-run-1.mjs` builds both outcome files and checks every race.

## What the frozen code computes on its own

`code/score.mjs` was run unchanged, and its own output is published unchanged: `run-1-contracts.csv`, `run-1-contracts.json.gz`, `run-1-summary.json`, `run-1-hashes.json`. It differs from the registered text in two ways, so its tables are not run 1's result:

1. It gives a contract Polymarket's payout whenever the official outcome is missing, not only for questions no election authority answers. That scores all resolved Tier B contracts and the pending Tier A races on the venue record (`venue_only` in its rows).
2. It rounds each contract's Brier score to four decimals before the means and the bootstrap. At full precision on its own population, its Tier B interval would start at 0.0463, not 0.0462; every other figure it prints is unchanged at four decimals.

Its own summary, for reference: 1,216 contracts scored.

| Set (frozen code, venue fallback) | Contracts | Races | Brier, contract mean | Brier, race mean | 90% interval (contract mean) |
|---|---:|---:|---:|---:|---|
| Tier A (no placeholders) | 909 | 452 | 0.0166 | 0.0166 | 0.0118 to 0.0220 |
| Tier B | 307 | 134 | 0.0562 | 0.0549 | 0.0462 to 0.0676 |
| All scored | 1,216 | 560 | 0.0266 | 0.0245 | 0.0216 to 0.0316 |

## Deviations and limits

- The frozen code's venue fallback and its rounding, above. Run 1's tables use the registered rule; `code/protocol-tables-run-1.mjs` was written on 5 November 2026, after the vote and after an independent check pointed out the difference.
- Tier B is pending in run 1: no official rows and no classification of venue-only questions yet.
- The supplements are listed, not scored (no order book and no resolution record was captured for them), and some of their markets were created before the freeze.
- 25 states and territories gave no count to plain scripted requests or publish none yet; their races are pending, with the reason per race. New Jersey's county clerks publish counts but the state does not before certification; county sums are not used as the state's record.
- Six pending rows have no retrieval at all (CTRL-SENATE, CTRL-HOUSE, CTRL-BOTH and the three Hawaii races): the frozen converter writes its own run time into their `retrieved_at` in `official-run-1.json`; that time is not a retrieval.
- Results are unofficial counts with AP calls; counts change until certification.

## Files

| File | Bytes | SHA-256 |
|---|---:|---|
| `METHODS.md` | 8,747 | `ffa417f044cd4dc6bbca934ee3dbd3ddb9c96868d9cbeda1d9da63c40d34f743` |
| `code/build-ap-calls.mjs` | 2,381 | `4a4996e37497456b9f31116fa30bf13db8477b380ea13e644347280db80c1766` |
| `code/build-race-worksheet.mjs` | 3,224 | `eab42074e74eeaacaf6d02d6d51edbffb3c8c320f6452135279afc5d6459fca7` |
| `code/build-venue-outcomes.mjs` | 2,427 | `576b19f80f8e9f9d91e9078328654d0dfefe173c95f93db01f30a4f4deb96a59` |
| `code/capture.mjs` | 11,287 | `a2f526d76eb7996ddfac96285229b7fa744ed4ae750a2d0429f214ce61267ef8` |
| `code/merge-official-run-1.mjs` | 20,258 | `d919d8339808f8d9d9140cb8ab7ef0abb5e5422b02cd8dc5b366fe2cf1a83d07` |
| `code/protocol-tables-run-1.mjs` | 16,921 | `f3760fa73138369200e24db59c38154ca6e8897ee32d7c0b3a159c37c1af493f` |
| `code/score.mjs` | 14,040 | `41e938b25dcd8fbc8a166c41c9dc91730f1217a5402b3c073c78125730ecc667` |
| `outcomes/README.md` | 1,540 | `bbed311b7845f2ca4bfa4da3209336d7729ff1334272035a461f37700a5e27b3` |
| `outcomes/official-run-1.json` | 108,924 | `0cbece7729d5c89152a6b7b9bf0708a44cdc34ea2c1b31ea346230a1a0c1a4c6` |
| `outcomes/official_by_market-run-1.json` | 55,688 | `93cc59d1fbf749e6213d87bbf446b7e3332b8ae8c0b5dd3d9b8004f50fda9794` |
| `outcomes/race-worksheet-run-1.csv` | 214,110 | `8c3cc8989a3592edacf6e49ff295bd3f571d24739676eb5cbeb477c35f1dd520` |
| `outcomes/venue-run-1.json` | 2,574,454 | `3290bedc62cf1761f24c305879fe31c6bb0e932bdf20cb2db22cea789f9c905b` |
| `receipts/all-captures.json` | 175,904 | `8209d181ec25da8128a63741766eb9201553e4387f48d19586cf73f9fac36aa7` |
| `receipts/capture-2026-10-27T23-50-02-612Z.receipt.json` | 657 | `f0d24deeeb14d6ff119b03660f6251ae66d347d0237d0c646362fa31c0a61af2` |
| `receipts/capture-2026-11-02T23-55-01-605Z.receipt.json` | 657 | `f8a2408795d52b53d7173368e18631007906d03382fd5e129423a932bc48211d` |
| `receipts/capture-2026-11-05T14-51-27-229Z.receipt.json` | 642 | `bd6080a1cee065fbe3f6b94b1ec8e89d2bfd7b8d13d2d89b180fc81bb86e6a3b` |
| `run-1-contracts.csv` | 2,737,202 | `c8cf604463d8f21084c0ab87732476c4bc0dc7d54ad84cc5e1f443cab958bb19` |
| `run-1-contracts.json.gz` | 2,612,177 | `1947c8309fca59e281fc2ebb22aaae6433b625acd4f924b6819429307bea4cb5` |
| `run-1-hashes.json` | 278 | `94fab64afd4805dd4551cd43340da4f51a2a052e9548d46ccb262c64c06b422c` |
| `run-1-protocol-contracts.csv` | 1,092,438 | `a7b85651ce6c857277bb746684961d2a888e08e60189745c264d1b8f2b5c2b0b` |
| `run-1-protocol-contracts.json` | 4,292,828 | `ab9e07e3ef9aeeb9c4c4ea57388402fcbf918a5dd7aaf8083e46aed1baf8669c` |
| `run-1-protocol-summary.json` | 10,210 | `6c962351814da51cf6d35c18dc8187c370d72d22c1f2c416e46e01383989bc99` |
| `run-1-summary.json` | 5,793 | `d1643beda2af71f199b15d9379017027e8571d10ca142955b2dab99ea9dd36b9` |
| `supplements/supplement-s02.csv` | 266,768 | `141ef59ba3197464fec03c4ee5143106b7c5c1971b7eec2095999aaf4bf66790` |
| `supplements/supplement-s27.csv` | 133,535 | `18302704b7c40077c6644ef2059a027004480afbd3b089860cfd77a31f7029c2` |

## Field definitions

`run-1-protocol-contracts.csv` and `run-1-protocol-contracts.json` (the same rows, one per manifest contract, all 13,186; the run-1 scoring): `market_id`, `tier`, `race_key`, `placeholder_outcome`, `p` (primary price), `p_secondary` (27 October price), `o_frozen_scorer` (the outcome the frozen code used), `outcome_source` and `official_status` and `venue_status` (as in the frozen rows), `basis` (why the contract is or is not scored in run 1, listed above), `o` (run-1 outcome, 1 or 0, empty when pending), `scored`, `brier`, `brier_secondary` and `log_loss_clipped` (full precision, p clipped to [0.01, 0.99] for log loss). `run-1-protocol-summary.json`: every run-1 table above with intervals, and the supplement counts.

`run-1-contracts.csv` and `run-1-contracts.json.gz` (the frozen scorer's rows, one per manifest contract; the JSON carries every manifest field except the rule text and event description, which are in the manifest):

- `market_id`, `event_slug`, `question`, `group_item_title`, `category`, `tier` (A, B or C), `race_key`, `placeholder_outcome`: from the frozen manifest.
- `p`: primary price, midpoint of the best bid and best ask of the Yes order book in the primary checkpoint capture; empty when the book lacks a bid or an ask.
- `best_bid`, `best_ask`, `last_trade`: from the same book record. `displayed`: Polymarket's displayed price (Gamma `outcomePrices`) in the same capture.
- `p_secondary`: the same midpoint at the 27 October checkpoint.
- `o`: the outcome the frozen code used, 1 or 0; empty when it had none.
- `outcome_source`: `official` (state or territorial election authority) or `venue_only` (Polymarket's payout, which the frozen code uses wherever the official outcome is missing).
- `official_status`: the race's status in the official record when the matcher produced an official reading for the contract; `none` when the race has no official row, or when it has one but the matcher produced no per-contract outcome (for example a row labelled "Other"). `venue_status`: open, proposed, disputed, resolved_yes, resolved_no, resolved_split or voided.
- `flags`: `wide` (spread over 10¢), `no_two_sided_quote`, `rules_changed` (rule text differs from the freeze), `records_differ`, `resolved_split`, `voided`.
- `scored`, `brier` (rounded to four decimals), `brier_secondary`, `log_loss` (price clipped to [0.01, 0.99]), as the frozen code computes them.

`run-1-summary.json`: the frozen code's own tables. `run-1-hashes.json`: the scorer's own SHA-256 of its three output files (the JSON rows uncompressed).

`outcomes/race-worksheet-run-1.csv`: `race_key`, the labels the race's contracts carry, then `winner_or_value` ("Party Name" as the state prints the name, or the word `pending`, which the scorer ignores), `status`, `source_url` (the state's results page), `retrieved_at` (UTC) and `note` (the state's count as shares, whether the state says counting is complete, the AP call and its time, and any reason a race is held). Raw state responses are kept on file with their SHA-256 and are available on request.

`outcomes/official_by_market-run-1.json`: `market_id`, `o`, `status`, `source_url`. `outcomes/venue-run-1.json`: `market_id`, `payout_yes`, `status`, `resolved_at`, `uma_statuses`, `retrieved_at`. `supplements/supplement-*.csv`: one row per market the capture found outside the manifest: its capture, ids, labels, `created_at` and `cohort` (created after the freeze, or created before it and found later), `found_in` (`manifest_event` or `tag_resweep`), catalogue prices at the checkpoint, `book_captured` (always no), `primary_price` (always empty), and on 5 November `closed_5_nov`, `outcome_prices_5_nov` and `closed_state_5_nov` (`not_closed`, `closed_prices_binary_unverified` or `closed_prices_not_binary`); `payout_verified` is always no, because no resolution record was captured.

## Reproduce

From a folder holding the manifest, the rules, the captures and these outcome files:

```
node code/score.mjs --manifest manifest-v1.json --rules rules-v1.json --captures captures --official outcomes/official-run-1.json --official-by-market outcomes/official_by_market-run-1.json --venue outcomes/venue-run-1.json --run run-1 --out results
node code/protocol-tables-run-1.mjs --rows results/run-1-contracts.json --rules rules-v1.json --captures captures --outcome-capture capture-2026-11-05T14-51-27-229Z.ndjson.gz --out results
```

The bootstrap uses a fixed seed, so the tables reproduce exactly. The checkpoint captures are kept on file with the SHA-256 in their receipts.
