Sources

The Sources tab — Sources — the audit trail. — is where the scan shows its working. Every item in the loaded snapshot is traced to the document behind it, and the page lays out what was searched, how citations were matched, what was verified, and how the scan was configured.

It exists for the moment trust is questioned: before you forward a briefing to a regulator-facing committee, when a number looks surprising, or when someone asks “how do we know this?”. The hero’s meta block sets the tone immediately — the engine the scan ran on, and the citation-audit verdict: CITATION AUDIT ● COMPLETE · 41/44 EDITORIAL CLAIMS CITED.

On this tab the ticker strip swaps to trust statistics: ITEMS · JURISDICTIONS · CLAIMS CITED · VERIFIED. Segments the snapshot cannot support are omitted rather than shown as zero — an absent number is honest, a fabricated 0 is not, and the page holds that line throughout.

The Trust Scorecard — The trust numbers.

The scorecard is the one-panel summary, with an audit-status dot in its header (green Citation audit complete; amber for partial, failed or timed out; muted when the audit was not run). Its cells:

Cells hide individually when a snapshot lacks that data — older snapshots may show a shorter card.

Source provenance — Where each item comes from.

The main list groups items by jurisdiction. Each item row carries:

• The title — click it to open the item in the shared detail panel.
• A sourcing-confidence chip, assessed at scan time:
  • Primary regulator source — sourced from a primary regulator document.
  • Supranational body — from a supranational standards body.
  • Secondary commentary — from commentary rather than the regulator itself; treat with more care.
• The source URL, with a cross-reference badge:
  • ✓ verified — this exact URL was found in the corpus of pages the scan actually retrieved.
  • ≈ domain — a page on the same domain was retrieved, but not this exact URL.
  • ? unverified — neither the URL nor its domain appears in the retrieved corpus.
• Or, plainly, a no source URL line when the scan could not attach one.
• Status, impact, likelihood and topic pills.

The two signals are deliberately independent: the chip grades what kind of source, the badge grades whether the scan can prove it retrieved it. A Primary regulator source that is ? unverified is a different risk from Secondary commentary that is ✓ verified.

Analysis — the working

Beneath provenance sit the audit’s collapsible worksheets:

• Search queries & sources — the actual queries issued, turn by turn, with each turn’s retrieved source URLs expandable underneath.
• Citation matching — x/y ITEMS ATTRIBUTED, with per-item scored source matches. When items could not be attributed above threshold, an amber band says so rather than hiding it.
• Source cross-reference — the counts behind the ✓/≈/? badges: exact matches, domain matches, unverified, and no-source.
• Relevance filtering — appears only on older snapshots that used a retrieval-filtering step; current scans do not produce it.

Scan configuration — How this scan was run.

The configuration panel records the run itself: jurisdictions, topic areas, the engine, the time horizon, entity types and any custom topics, followed by one row per analysis stage that ran — including whether the verification and citation-audit stages were enabled for this scan, and which translation languages were requested. The foot line stamps SCANNED date and time, DURATION, SEARCHES, and the scan’s cost (plus a running total of any export costs for this snapshot).

On sample data this panel says “Run a live scan to see configuration details here.”

Methodology — the key

The final panel, Methodology — how to read the trust signals, is the in-product key to everything above: the ✓/≈/? badge semantics, the three confidence chips (including “tunnelling” — the silent upgrade of an aggregator link to the primary regulator, with the stated reason in the chip’s tooltip), and the amber audit chip.

Its most important sentence is worth restating here: the [N] citation markers are an attributor, not a fact-checker. The audit selects which retrieved source backs each editorial sentence; it never asserts the source is true. An amber Citation audit partial chip means some markers may be missing or inactive — not that the scan failed, and not that any finding is wrong.

Worked example: standing up a number

The Briefing claims a licensing figure you are about to quote to your board.

1. On the Briefing, tap the sentence’s [7] marker — it names an ESMA page. Note the item it belongs to.
2. Open More → Sources. The ticker reads CLAIMS CITED 41/44 · VERIFIED 19/24 — the audit ran and mostly verified.
3. Find the item under its jurisdiction. Chip: Primary regulator source. Badge: ✓ verified — the exact URL is in the retrieved corpus. Open it and check the figure in the regulator’s own text.
4. For belt and braces, expand Citation matching and confirm the item was attributed with a strong score.

Now you can quote the figure with its provenance attached — regulator document, retrieved by this scan, attributed by the audit.

A non-obvious tip

The scorecard’s Tunnel substitutions cell is a quiet quality signal worth reading. Each substitution means the scan found a claim via an aggregator but upgraded the citation to the primary regulator document. A handful of substitutions on a news-heavy topic is the system doing its job; if you see none on a topic dominated by commentary, expect more Secondary commentary chips below — and budget verification time accordingly.

Where next

• Overview — the confidence chips and claim counts resurface on every detail card.
• Briefing — where the citation markers live.
• Scans and snapshots — how a scan is scoped and recorded.