thejeffadrian-lab/ds-9f4c2e

Research data archive

Access-gated working data for an in-progress quantitative research project. Access is granted individually; request access if you have been directed here by one of the authors. Not intended for redistribution. Schema and layout are being migrated; pin a commit when running an analysis.

Bulk standardization in progress

Markets contains 20,798,330 outcome/reference-instrument rows across 18,709,041 reconciled native market IDs, including exclusions and review holds. 5,402,808 markets meet the current population rules; that does not establish verified measurement coverage. Kalshi uses one YES-reference instrument per contract; Polymarket preserves every reported condition outcome, including legacy multi-outcome conditions. This is not a complete venue census.

Categories use eight shared groups, with explicit Ambiguous/Unclassified labels and native labels retained. Native metadata and resolution evidence, source hashes and update recipes are preserved in evidence packs, indexed by their original source paths. Reported lifecycle timestamps retain source-specific meanings; they do not replace final-resolution evidence.

Quotes and provider trade candles are being populated from the full archived Kalshi one-minute candle family, using the verified exact-copy removal map. These archived bid/ask closes are not verified synchronous midpoints; missing bid/ask OHLC remains null. Provider trade candles are separate from summaries computed from execution observations.

Hourly bid/ask closes and hourly provider trade candles contain 19,103,664 distinct candle observations and 136,863 empty-market bookkeeping rows from all 582 archived Kalshi 60-minute source files. The 433,086 identical repeated measurements are counted once; original metadata variants and copy lineage remain preserved alongside the tables, and prior source objects remain recoverable through dataset history. Existing partition paths are unchanged. Join companion tables on (source_ref, source_row).

The compatibility reader uses the exact bound shared market metadata, includes only markets meeting the population rules without conflicting evidence, and restricts observations to July 19, 2026, 23:59:59 UTC. Held, excluded and post-cutoff observations remain in the archive. The interval is explicitly 60 minutes. Bid/ask closes are not verified synchronous midpoints, and provider trade candles are not execution-level observations. The bulk manifest records the completed reconciliation separately from the original conversion audit.

Polymarket execution observations preserve all 406,747,336 source rows in our existing execution archive. Component decisions classify 404,539,960 distinct CLOB logs and preserve the 40 identical source copies in lineage. These are source observations and counting evidence, not a verified economic-trade table; summary logs must not be added to their maker components. Identity, clock and amount evidence is retained under execution evidence. Historical CLOB emitter labels still rely on collector provenance rather than independent native-log verification. No Polymarket execution or one-minute trade rows are promoted to verified analysis by this publication.

The bulk manifest records the number of published partitions and rows for each family. The release is incomplete and not ready for verified analyses. Market eligibility, identity, measurement verification, and source coverage are separate requirements. Polymarket scalar price history is not execution or bid/ask data. Candidate-filtered Polymarket execution recovery is not full-corpus execution coverage.

Volume measures must remain separate on both venues:

• Quantity: contracts/shares, counted once per reconciled execution.
• Payout notional: quantity × verified face payout (for example, 100 $1 contracts = $100); not price × quantity.
• Reference cash volume: fee-exclusive reference price × quantity. Use YES when explicitly named, otherwise outcome 0, for complementary binary markets; preserve separate outcome series for 3+ outcomes. Current standardized notional and volume_notional columns with *_outcome_reference units mean this measure, not payout notional.
• Source-reported cash and fees: retain separately, with source currency, units and counting rules. Traded-token cash need not equal reference cash. Do not assume collateral currency is USD.

Polymarket reported market totals preserve all 1,486,642 rows from poly_pull/onchain_market_volume.parquet, with every original value checked. reported_cash_volume, reported_fill_count, and reported_token_count retain source definitions; they are not share quantity, payout notional, harmonized reference cash, or verified execution counts. Existing analysis readers restore the old column aliases only at the compatibility boundary. The original file remains available in prior commits.

Polymarket reported_cash_volume and reported_fill_count are unverified source aggregates, not harmonized reference cash or execution counts. They lack share quantity. Older V2 aggregation also counts maker fills together with aggregate taker summaries; the full historical impact is under review. Kalshi provider reported_volume remains separate from execution-derived measures. Neither candle close × provider volume nor scalar price history can recover missing execution cash. Missing coverage is not zero volume.

Verified replacements remove old paths only from the active revision. Original files remain accessible in the pinned source revision and prior commits. History and underlying recovery objects are not purged.

Coverage and evidence

The corpus catalog accounts for 7,845,001 native metadata records, 34,977,944 additional trade-only identifiers, and 25,233,462 additional daily-report-only identifiers. Unmapped identifiers remain held; they are not automatically eligible markets. Its source ledger refers to the pinned archive, while the bulk manifest tracks the ongoing replacements.

The reviewed metadata bundle is retained as supporting evidence for 21,980 Kalshi and 62 international Polymarket markets (22,202 outcome rows). It is not the bulk table or a representative sample. Its empty measurement tables are not the new bulk measurement tables.

Base code; metadata update recipes are also saved in the evidence packs.

Metadata scope v8

The current metadata admits 4,364,447 Kalshi and 1,038,361 Polymarket markets; 44,847 and 89,834 respectively remain held. Verified own crypto-token market-cap/FDV targets, including relative token valuations, are excluded. Aggregate RWA capitalization, dominance and unresolved evidence remain held. No observation data or duplicate selections changed. Existing coverage receipts retain their original metadata pins; denominators for the new admissions have not been recomputed in this metadata publication.

Use reader code at 7374967 or a later compatible revision. Older readers can retain the previous dataset pin. Current shared category mappings are embedded in the market table; the bounded reviewed_metadata category map remains historical.