caiotheodoro/titer-edgar-officers

titer · EDGAR officer corpus

4,206,080 attested person–company–role–date tuples from SEC Forms 3/4/5, published as pointers rather than records, alongside the frozen pre-registrations that were hash-published before any measurement ran.

edgar_officers.parquet: 4.2M rows, 230,405 distinct people, 20,266 issuers, 2006q1–2026q2.

Column	Meaning
accession	SEC accession number, the pointer that reconstructs the row
person_cik	Reporting-owner CIK. Never recycled by the SEC.
issuer_cik, issuer_ticker	The filing issuer
role_class	DIRECTOR / OFFICER / TEN_PERCENT_OWNER / OTHER, parsed from the filing, never rewritten
title_class	9-class closed taxonomy, regex-only, no model
period, filed	The attested event date and the date it became public
collision_degree, collision_band, strict_degree	Distinct CIKs sharing a name, under two normalizations
name_sha256	Salted hash. No names are published.

Nothing else on Hugging Face packages this: the closest is ~34,882 rows, about 0.5% of a single year.

Why pointers and not records

Every row reconstructs from the SEC's own free quarterly files using the accession number. We distribute the pointers; the SEC distributes the records.

No contact fields appear: no email addresses, telephone numbers, postal addresses, social profiles or dates of birth, not even where the field is technically public. A publish-time guard refuses any column whose name suggests one. Aggregating scattered public facts about real people into one convenient file is itself a privacy harm, and it is the specific harm the parent project exists to measure.

The limitation that matters

person_cik is unique per filer registration, not guaranteed one per human. Different CIK proves different registration, which is the direction false-merge detection depends on. The reverse is presumptive: one person can hold two CIKs and the SEC publishes no merge table.

Ground truth is a filing rather than a model's judgement. Section 16 of the Securities Exchange Act compels an officer or director to state, by name, that they held a role at an issuer on a date, and the SEC assigns the filer a Central Index Key it never recycles.

The two studies

	Population	Question	Frozen
Employment	SEC Section 16 officers and directors	How often does a people index resolve the wrong human, and how stale is it?	2026-09-02
Expertise	Scholarly authors (OpenAlex / ORCID / Crossref)	How often does a provider affirm expertise that is not attested?	2026-09-03

The second study exists because of what the first one found about itself. Two of its designs could not answer their own questions: random sampling put 272 of 299 observations into a single elapsed-time bin, and 251 of 299 into a single name-collision band. Both were caught by looking at where observations landed rather than at a summary statistic. Simple random sampling is prohibited in the expertise pre-registration as a direct result.

The expertise population is stronger in one specific way: a false claim is constructible. A person's attested topics come from their authored works, so a false expertise claim is a topic with zero attested works, drawn from an adjacent field by a mechanical rule. The employment corpus could never manufacture a verifiable non-employment.

Why the pre-registrations are published

titer measures how often a commercial index confidently affirms something false about a person, and what it is worth paying to find out. That answer is easy to shape after the fact. The budget is small, so the achievable sample size is small, and a small sample plus a motivated author is how a disappointing result becomes a ranking. Each pre-registration fixes, in advance, the hypotheses with their falsification conditions, the estimators and interval methods, a power rule stating what may be claimed at each achievable n, and a commitment to publish every falsified hypothesis and exclusion rate.

Both are byte-identical to their first publication. Commit dates are locally forgeable; these hashes are not. Verify with sha256sum PRE-REGISTRATION.md and compare.

What the instrument found: five of its six headline numbers were artefacts of the harness, not findings about any vendor. Two formal retractions are filed. Results, decisions and retractions: https://github.com/caiotheodoro/titer

Contents

File	What
edgar_officers.parquet	The corpus described above.
PRE-REGISTRATION.md	The frozen employment analysis plan. Never edited; superseded only by dated decision entries carrying a reversal clause.
PRE-REGISTRATION.sha256	Its hash, as locked in the working repository.
PRE-REGISTRATION-EXPERTISE.md	The frozen expertise analysis plan.
PRE-REGISTRATION-EXPERTISE.sha256	Its hash.
CONTRACTS.md	The frozen definitions the pre-registrations depend on: attestation tiers, the attested tuple, closed taxonomies, outcome classes, cost profiles, reward, split rules.
CONTRACTS.sha256	Its hash.

Intended use

Evaluating identity resolution and index freshness. Not for locating or contacting individuals.

Rectification

Anyone appearing in the pointer set may request removal. Requests are honoured within 7 days, without requiring justification, and recorded as a count, never as an identity.