Public dataset — quarterly snapshot
A full download of the Trustgent verified-provider corpus, published every quarter as CSV and JSON. Schema v0.1.0, first published 2026-07-07. No login. No API key. Anyone can cite it, replicate it, and challenge it.
What this is
Every quarter Trustgent publishes a full snapshot of the verified-provider corpus — one row per provider × capability, one file per quarter, free to download and redistribute under a Creative Commons attribution license with a no-misrepresentation clause. This is the public projection of the moat: the same verification levels, decay states, and record counts that power the directory, exposed as a queryable file an outside analyst can reproduce our aggregate claims from without access to any internal system.
There is no live API. Snapshots only. That is a deliberate choice under the Attestation Protocol — the snapshot cadence keeps citations reproducible (the same file exists at the same URL forever), keeps the ops budget honest, and forces every downstream analysis to declare the snapshot it was computed on.
Download the latest snapshot
Latest published quarter: 2026-Q3. Schema version v0.1.0. Next snapshot expected 2026-Q4 (published on or shortly after 2026-10-01).
The latest-providers.* aliases always point to the newest cut and are the right link to embed in a long-lived reference. The quarter-suffixed files (2026-Q3-providers.csv) are stable and appropriate for reproducible citations.
Schema — field by field
Every row in the CSV and JSON snapshot carries the ten columns below. A version bump in the schema is announced in the changelog and stamped into the methodology_version field of the JSON header, so consumers on a prior schema can detect the change deterministically.
| Column | Type | Notes |
|---|---|---|
| provider_id | string | Stable public token. Currently the provider slug (already public on trustgent.com). Will migrate to a salted-hash token once the ledger volume trigger fires — see the day-180 gate in the strategy plan. Cross-quarter joins remain stable. |
| name | string | Public display name — the same string that renders on the profile page. |
| capability | string | Canonical capability slug. One row per (provider × capability); a provider offering N capabilities emits N rows. See /how-we-analyze for the capability taxonomy. |
| country | string | ISO-3166-1 alpha-2 HQ country code, lowercase (e.g. nl, de, gb, us). |
| l_tier | string | One of L0, L1, L2, L3, L4, L5. Meaning of each level is defined on /how-we-verify at the methodology version stamped in the JSON header. |
| attested_at | string (ISO-8601) | null | Timestamp of the most recent attestation that put the provider at this level. Null at L0 by design — L0 is the reach layer and carries no attestation. |
| decay_state | string | fresh | stale | expired. Computed from attested_at against the L-tier freshness horizon (documented in the snapshot script). Providers with an expired attestation appear in the snapshot so consumers can see the honest state of the corpus. |
| record_count | number | Total count of independently verified engagement records at this provider's l_tier or above. A rank input under the moat charter's frozen allowlist. |
| claimed | boolean | True if the provider has claimed the listing. Uninformative for ranking; useful for coverage analysis (what fraction of the corpus is owner-managed). |
| founded_year | number | null | Provider-declared founding year. Already public on the profile. |
What is not in the snapshot
The snapshot exposes only the fields listed above. Under the moat charter (see the Attestation Protocol), the following are structurally excluded:
- Plan tier, subscription status, or any payment signal.
- Contact emails, phone numbers, or any owner PII beyond what is already public on the profile.
- Evidence bundle contents (contracts, testimonials, interview transcripts).
- Ledger inbound records — those live under /standard/ledger under their own attestation regime.
- Buyer telemetry (brief starts, deal history, referring queries).
- Any signal outside the moat charter's frozen ranking allowlist.
A physical data-separation CI check gates every pull request against ranking code importing from the publication schema, so this exclusion is enforced in code, not by convention.
Cadence
Quarterly. The snapshot cron runs at 06:00 UTC on the 1st of January, April, July, and October — the same cadence as the moat audit. Between snapshots the files do not change. Any consumer that needs a fresher cut should either wait for the next quarterly rollover or reproduce the query against the public methodology and explain the deviation.
Snapshot cadence SLO: ±7 days from the quarter boundary. Miss two cycles = freeze publication and reassess (see the strategy plan's Flywheel section).
License
The dataset is licensed under Creative Commons Attribution 4.0 International with a Trustgent-specific no-misrepresentation clause. Full text: /standard/data/LICENSE.md (draft — marked for counsel review).
In plain English: republish, remix, and analyze the dataset freely, including commercially. Attribute Trustgent and link to /standard/data. Do not publish a modified version under branding that would mislead a reader into thinking it is Trustgent's own assessment.
Reserved: survey aggregates (not yet published)
The schema reserves a separate quarterly file, YYYY-QN-survey-aggregates.json, for aggregate answers from the optional research questionnaires (buyers and verified builders). It publishes only once two conditions hold per question: the research data product has passed counsel review, and the question has at least 30 responses. Until then the file is absent by design — we publish a gap, not a thin number.
Two properties are fixed in advance. Survey answers never appear as columns on provider rows and never at respondent level: only buckets of three or more, exactly as the consent notice next to each questionnaire promises. And survey data never touches ranking — it is research output, not a verification signal. Shape and activation detail: the schema changelog.
Changelog
Schema history: docs/standard/data-CHANGELOG.md. Semver applies — minor bumps add columns without breaking existing consumers; major bumps redefine or remove columns.
Related surfaces
- /how-we-verify — what each L-tier means and what evidence produces it.
- /how-we-rank — the plan-blind ranking function.
- /protocol — the versioned Attestation Protocol specification.