Qipudata
San Francisco · v0.7.1
qipu data›San Francisco›Datasets›claim_reliability

Claim reliability

claim_reliabilityrelease v0.7.1CC0 1.0

About

How well each claim is backed (qipu_data/reliability.py, phase e). A claim says what a record states, not that it is true; the tier says how firmly the claim is tied to its record and how well its reading was measured. verified = checked by a person against the scan, or stated by 2+ independent documents; quoted = its quote is on the cited page and its run measured at or above 0.9; unverified = everything else (the reason says which). Not the same as claims.reliability, which only says whether the run that read the claim was scored (measured / unmeasured). From v0.6.2 also in the DuckDB.

How far to trust it
The release passed its automatic checks before it was published (what is checked). A hand check by the editor of qipu data is under way: 36 of 38 items checked so far were right, on votes, money, places and the map. A row says what the record states; the scanned page is the authority. Details
Licence
CC0 1.0: use it for anything, no permission needed. The scans and other sources keep their own rights.
Version
v0.7.1, published 21 October 2026. A published release is never changed; corrections come as a new release.
Technical details
Table
claim_reliability
Rows
8,367,105
Columns
12
Parquet
claim_reliability.parquet, 132.2 MB
Version
v0.7.1, 21 October 2026
Licence
CC0 1.0. Sources keep their own rights.

Rows

Rows by quote_check

ValueRowsShare
found5,300,05763.3%
no_quote2,541,19730.4%
not_found409,1914.9%
no_ocr116,6601.4%

Rows by run_accuracy_against

ValueRowsShare
model_key4,428,85252.9%
(empty)3,923,43046.9%
human14,8230.2%

Rows by tier

ValueRowsShare
unverified5,403,52164.6%
quoted2,962,69535.4%
verified889< 0.1%

Method and accuracy

No accuracy measure for this table yet.

The release passed its automatic checks before it was published (what is checked). A hand check by the editor of qipu data is under way: 36 of 38 items checked so far were right, on votes, money, places and the map. A row says what the record states; the scanned page is the authority.

Methodology · Accuracy

Columns 12

ColumnTypeCan be emptyPrivacyPublic copyDescription
claim_idVARCHARnononeas isThe claim (see claims).
Values e.g. c:c4ca4238a0b923820dcc509a6f75849b
Can be empty no · Privacy none · Public copy as is
quote_checkVARCHARnononeas isIs the claim's quote on the cited page? found (its first words in order and 80% of its words are in the page's OCR text, or the next page's), not_found, no_ocr (no page text to look in yet), no_quote.
Values found, not_found, no_ocr, no_quote
Can be empty no · Privacy none · Public copy as is
run_accuracyDOUBLEyesnoneas isThe measured accuracy of the run that read the claim (its headline score in field_scores). Empty: the run is unmeasured.
Values share 0-1; 0..1
Can be empty yes · Privacy none · Public copy as is
run_accuracy_againstVARCHARyesnoneas isWhat run_accuracy was measured against: human (a hand-checked key) or model_key (a key written by a stronger model, not checked by a person). Shown separately: they are not the same evidence.
Values human, model_key
Can be empty yes · Privacy none · Public copy as is
run_accuracy_nBIGINTyesnoneas isHow many items the score was measured on.
Values 0..
Can be empty yes · Privacy none · Public copy as is
support_countBIGINTnononeas isHow many extracted rows read this same claim (claim_support); several readings of one page are not independent.
Values 1..
Can be empty no · Privacy none · Public copy as is
independent_sourcesBIGINTnononeas isHow many different documents state the same subject, property, value and date.
Values 1..
Can be empty no · Privacy none · Public copy as is
checks_failedVARCHAR[]yesnoneas isConsistency checks this claim failed: tally_count / tally_outcome (the printed vote tally disagrees with the extracted votes or outcome), tally_ocr_count, date_outside_volume (the date is outside the years of the cited volume).
Can be empty yes · Privacy none · Public copy as is
human_verdictVARCHARyesnoneas isThe hand check of this claim or its decision, when a person checked it against the scan (the worst field verdict).
Values right, partly, wrong
Can be empty yes · Privacy none · Public copy as is
flag_statusVARCHARnononeas isA reader's flag on this claim: pending (waiting for review), corrected, rejected (reviewed, the claim stands).
Values none, pending, corrected, rejected
Can be empty no · Privacy none · Public copy as is
tierVARCHARnononeas isverified / quoted / unverified (see the table description). Not a truth score.
Values verified, quoted, unverified
Can be empty no · Privacy none · Public copy as is
tier_reasonVARCHARnononeas isWhy the claim has its tier, in plain words.
Can be empty no · Privacy none · Public copy as is

Privacy: what kind of text the column holds. Public copy: what the published file does with it; the embargo covers records after 1951.

Preview

claim_idquote_checkrun_accuracyrun_accuracy_againstrun_accuracy_nsupport_countindependent_sourceschecks_failed
c:8a90d2048232a1f08a88042a760787bfno_quote0.95model_key11511—
c:2bb984d63d9ac7d79c2519df46934fd6found0.95model_key11511—
c:a7cbe5d643e757ea41a45f6d13849dd1found0.95model_key11511—
c:256de558ea85f5ab209fd6ea19c64900found———11—
c:bd108e1d07020e5c6ca30055deace399found———11—
c:40ee5d284546a9492622efd2e3d52da5found———11—
c:a59f64671c58e376b24f64359c8e1d8cfound———11['date_outside_volume']
c:1be0e761fc61c08cfcea4a5fc7d67f38found———11—

8 of 12 columns shown. Eight rows sampled at build time. Hover a shortened value for all of it.

Download

FormatFileSize
Parquetclaim_reliability.parquet ↗132.2 MB
SQLite (all tables)qipu_sf.db ↗6.44 GB

SQL

SELECT * FROM claim_reliability LIMIT 100

Run in the SQL console (DuckDB in the browser, over the Parquet file).

Python

import duckdb duckdb.sql("SELECT * FROM 'https://tiles.qipu.org/data/releases/v0.7.1/parquet/claim_reliability.parquet' LIMIT 5").show()