← Trust
Data dictionary
Praxis · MarketGraphs · trust · document 10
The fields a subscriber receives, by depth, and what each one means. A sample of fewer than 100 open rows, each more than three months old, is sent on request to alex@market-graphs.com.
Index. Every paper, both plans
| Field | Meaning |
|---|---|
| paper_id | Our stable number for the paper. |
| title, authors, published_date, year | As published. |
| doi, arxiv_id | The paper's identifiers, where it has them. |
| link | The original, at its publisher or repository. |
| source_family | The index or repository the paper was identified from. |
| family, secondary_families | The strategy family the paper belongs to, one of 22, and any others it touches. |
| asset, asset_class | The asset the paper studies and its class, one of nine. |
| instrument_types, geography, strategy_clusters, technique_families, predictor_types | Lists of tags describing what the paper trades, where, and how. |
| in_universe_since | The day the paper entered the library. |
| max_level | The paper's depth: summary, method or audit. |
| has_extraction, audited | Whether a method record exists, whether an audit exists. |
Summary. Every paper, both plans
| Field | Meaning |
|---|---|
| claim_numbers | The numbers the abstract states: Sharpe ratio, return, t-statistic, period. |
| methods, datasets, instruments | What the abstract names as the method, the data and the instruments. |
| jel_codes, keywords, topic_tags | Classification codes and keywords, from the paper and from us. |
| sample_window, sample_windows, frequency_finest | The test period and the finest data frequency the abstract states. |
| replicability, robustness | What the abstract says about reproducing and stress-testing the result. |
| costs_mentioned, out_of_sample, multiple_testing, benchmark_named | Whether the abstract states each of these disclosures. True, false or unknown. |
| long_short, universe_size, vendors | Whether the strategy is long-short, the size of the universe, data vendors named. |
| claim_words | The words the abstract uses to characterise its result. |
| abstract_sentence | One quoted sentence of the abstract, attributed. |
| neighbour_count | How many papers sit nearest to this one in the library. |
Neighbours. Every paper, both plans
| Field | Meaning |
|---|---|
| rank, neighbour_id, neighbour_title | The nearest papers, in order. |
| reason | Why the paper is a neighbour, in words. |
Method. Papers read in full, paid plan
| Field | Meaning |
|---|---|
| methodology | How the strategy works: signal, universe, rebalancing, holding period, sizing, costs, parameters. |
| data_sourcing, data_cleaning | The data the paper used and how it prepared it. |
| sample_universe | The securities, markets and period the paper tested on. |
| results | The results the paper reports, as the paper states them. |
| validation_disclosures | What the paper discloses about its tests: out of sample, costs, multiple testing, trials. |
| limitations, regime_claims, caveats | What the paper says limits the result, and what it claims about regimes. |
| replication_confidence | What is stated and what is missing for someone to reproduce the result. |
| fields | Trading-desk measures: volatility, skewness, kurtosis, turnover, headline net of costs, microcap filter, execution lag, short side, capacity, data requirements. |
| Every value above | Carries a quote of at most 200 characters and the section it comes from, is marked sure or unsure with a confidence, and anything missing is listed. |
| missing, unsure | The fields the paper does not state, and the values we are not sure of. |
| verification | Which checks the record passed. |
| extraction_confidence | Our confidence in the record as a whole. |
| model_assisted, model_id, prompt_version | Whether a model helped, which one, and which version of our instructions. |
| extraction_schema, extracted_at | The record's format version and when it was built. |
| computed | Numbers derived from the record itself, without a quote: complexity, sample years, publication gap, minimum back-test length. |
Audit. Papers we re-implemented
| Field | Meaning |
|---|---|
| run_id, verdict, verdict_summary | Which run, the verdict in one word, and in one paragraph. |
| claimed_sharpe, measured_sharpe | What the paper claimed and what we measured on clean data. |
| deviations, flags, focal_config | Where our run differs from the paper, what to watch, and the configuration under audit. |
Provenance. Every paper, both plans
| Field | Meaning |
|---|---|
| source, source_family, file_rule | Where the paper was identified and the rule that applies to files from that source. |
| licence, licence_url, licence_max_level | The licence on the paper, where it is stated, and the depth it allows. |
| licence_evidence_url, licence_checked_on | Where we saw the licence and when. |
| obtained_via, fetched_from, fetched_at | How we obtained the file, from where, and when. |
| file_discarded | Whether the file was deleted after reading. |
| provenance_not_recorded | Always false on a served method record. |
| model_id, prompt_version, extraction_schema, extracted_at | As in the method record. |