The Hub · Method

How records are rated and kept fresh.

Every record on this hub carries a corroboration rating, a Checked stamp and a checked link. This page explains the loop that produces them, the ladder behind the ratings, and the publication policy that decides, in advance, what gets published at all.

The loop

Scan, verify, organise, maintain.

The Hub is produced by one repeating loop, not by one-off editing.

Scan

Sweep the sector's scattered public record: schemes, regulators, programmes, datasets, investigations, events.

Verify

Corroborate each claim across independent sources and rate it. A body publishing about itself counts as one source.

Organise

Compose the verified facts into sections, profiles and feeds, every figure attributed and dated.

Maintain

Re-scan on a set cadence, re-check links, renew stamps, and log every change in the Updates feed.

Corroboration

The rating ladder

Material claims are rated by the number, diversity and reliability of the independent sources behind them. Figures are dated to the year they refer to, not just the year they were printed.

Well corroborated
Three or more independent sources of different types agree.
Corroborated
Two independent sources, or one authoritative source plus corroboration.
Single source
One source only. Read with caution.
Contested
Credible sources disagree. Both sides are published and attributed; contested counts appear as ranges, never as one number.
Freshness

Checked stamps and link health

Every record carries a "Checked" stamp: the month its link and record details were last checked. The verb is deliberate. A stamp says the Hub checked the record; it does not vouch for every claim on the far side of a link. That is what the corroboration rating is for.

Every external URL is verified with a real request before publication and again on each maintenance pass. A URL the checker has not seen reads "unchecked", never a silent ok, and nothing ships on a dead link.

The freshness targets: no record stamp older than 90 days, a fortnightly changelog through the Updates feed, and automated link health across the whole reference.

Publication policy

Nine rules, no judgement calls

Every combination of rating, source type and claim type maps to exactly one publication outcome, so no crawl or maintenance pass ends with an open editorial call. The rules, one sentence each:

P1Status chips are computed, not asserted: "active" needs the body's own channels active within 12 months and independent corroboration within 18; silence beyond 18 months reads "dormant".
P2Contested counts publish as attributed ranges, the floor source and the ceiling source both named, never one number and never an average.
P3Self-reported figures always carry "per X, unaudited" and never enter headline statistics.
P4A statistic that cannot be traced to a named primary source is dropped, not hedged.
P5Register inclusion is criteria-based: at least corroborated existence and scope, plus direct fit to a section; anything below the bar parks in the ledger rather than being silently dropped.
P6Single-source allegations naming people or firms live only in the dated Updates feed, with the counterparty's response quoted or "no response documented".
P7Legal-instrument status is computed from the official record (draft, agreed but not in force, in force, suspended, voluntary); secondary claims do not move a chip.
P8Edition-based datasets carry their current edition, reference year, cadence and expected next edition, so "overdue" renders by computation.
P9A claim resting solely on an unfetchable, unmirrored URL parks in the ledger; nothing is published with a "check this yourself" flag.
Upkeep

Maintained, in the open

The hub is re-scanned on a set cadence. New records are added, links re-checked, stamps renewed, and every refresh lands in the dated Updates feed, so the reference can be audited as it evolves.

The crawl ledger

Candidates that fail the bar are parked, not discarded. Leads, unresolved counts, fetch-blocked sources and watch items are logged in a ledger and retried on later passes, so coverage compounds instead of resetting with each crawl.

Where this is going

Planned, and stated as planned: continuous curation by agents on a standing cadence, with crowdsourced feedback from logged-in contributors folded into the same verification loop.

Report a gap or a correction

If something is missing, wrong or out of date, say so. Reports are logged when they arrive and handled in the next maintenance pass.

Send a correction
Changelog

What changed, when

Every expansion or correction of the record base gets a dated entry here.

  • 2026-07-03

    Maintain pass. Re-verified all 15 open frontier leads and the 6 parked candidates against live sources: no material change since the founding build (both LBMA incident reviews still open with the refiners still on the Good Delivery List, RGG v10 and Fairmined 3.0 still in consultation or stalled, the EU list of responsible smelters still unpublished, Gold Demand Trends Q2 2026 not yet out). Corrections and additions: the CMSI Final Consultation Report date fixed to 12 March 2026; the DMCC Rules pinned to Version 2 (2020); the RJC certified-vs-total member split added (about 1,644 certified of roughly 2,000); the GoldBod traceability procurement updated (27 bidders, technical evaluation, no vendor awarded); a Swissaid UAE gold-import figure added (316 t / CHF 27bn, Jan-Sep 2025), corrected from a mislabelled "total Swiss imports" framing. Link health re-checked: 100 URLs, all live except the known Borsa Istanbul TLS case.

  • 2026-07-02

    Founding build. Two-round source-rated crawl (10 clusters, gap sweep, adversarial verification) distilled into 100+ rated records across 8 directories, 8 flagship profiles, 9 method-tagged number spreads, a 38-entry dated feed and a 12-document library. Publication policy P1-P9 adopted; every record link liveness-checked.

The engine

About Crawl

Crawl is a DAAC product: an agentic engine that scans a sector's scattered public record, verifies findings by corroboration, organises them into a navigable hub, and maintains their freshness. This hub runs on it, alongside DAAC's other reference hubs.

Crawl at daac.ai