Items, sources and citations
An item is the atom of a snapshot: one regulatory development, placed at one jurisdiction × area coordinate, described and classified. Everything the dashboard shows you is built from items — a heatmap cell is a stack of them, a timeline entry is one of them with a date, a briefing priority is a claim about several of them. This page covers what an item carries, and then the part that deserves the most careful reading: how the product tells you how well it knows what it is telling you.
The anatomy of an item
Open any item — from a heatmap cell, the timeline, or the map — and the detail card shows the same fields:
| Field | What it carries |
|---|---|
| Title | The development, named — usually the instrument or proposal itself |
| Jurisdiction · Regulator | Where it comes from, and which authority owns it |
| Area | The policy domain it is filed under (including your custom areas) |
| Status | Where the measure is in its lifecycle — see below |
| Likelihood | How firm the development is — see below |
| Impact | How much it matters — High · Medium · Low, angled by your entity type |
| Expected date | When it is expected to bite; items may also carry effective dates or compliance deadlines. Undated items show TBD rather than an invented date |
| Summary | A short editorial account, carrying numbered citation markers |
| View source → | The link to the underlying document, when the item carries one |
Two of these fields are ladders worth internalising, because the filter bar treats them as the primary lenses on every reading:
Status tracks a measure’s lifecycle: Proposed → Consultation → Final Rule → Implementation Phase → Effective. It answers “how far along is this?” — a Consultation wants a response; an Implementation Phase rule wants a project plan; an Effective rule is a live obligation.
Likelihood — Confirmed · Likely · Possible — answers a different question: “how certain is it that this lands as described?” A Confirmed item has a fixed footing (rules adopted, date set); a Possible item is direction-of-travel. Items for which no likelihood could be established are treated as Uncertain in the filters rather than being silently promoted. Status and likelihood are deliberately independent axes: a Proposed measure can be Confirmed-likely to proceed, and a Final Rule can still carry Possible timing.
Impact and likelihood combine in one visible place: the Overview heatmap. Each cell prints a pressure score out of 10 that weighs impact by likelihood, so a confirmed high-impact item heats a cell far more than a speculative one. The grid is a severity-times-certainty reading, not a raw item count — the count survives in the cell’s tooltip.
Three trust signals, kept separate
The product attaches three independent signals about sourcing to every scan, and it never blends them into a single score. The separation is the point: each signal answers a different question, fails in a different way, and is honestly absent when it was not produced (an older snapshot simply shows fewer signals — never a fabricated zero).
1. Sourcing confidence — what kind of source is this?
Assessed while the analysis is generated, shown as a chip on the item:
- Primary regulator source (high) — the item rests on a primary regulator document
- Supranational body (medium) — a supranational standards body
- Secondary commentary (low) — commentary about the regulation, not the regulator itself
Where an aggregator link was silently upgraded to the primary regulator’s own document — the product calls this tunnelling — the chip’s tooltip says why. The Sources tab counts these substitutions in the Trust Scorecard.
2. Source cross-reference — did the scan actually see this page?
After generation, each item’s source URL is cross-referenced against the corpus of pages the scan actually retrieved, and badged:
- ✓ verified — the URL was found verbatim in the retrieved corpus
- ≈ domain — a page on the same domain was retrieved, but not this exact URL
- ? unverified — neither the URL nor its domain appears in the corpus
This is a mechanical check, not a judgement — which is exactly why it is trustworthy as far as it goes, and why it is kept separate from the confidence chip. A Primary regulator source item with a ? badge is telling you something specific: the claim type is strong, but the trail to this particular page is not.
3. Citation markers — which source backs this sentence?
The numbered [N] markers on the executive narrative and summary are attached after generation by a dedicated audit pass, which selects supporting URLs from the scan’s own source pool for each editorial sentence. In the product’s own words: it is an attributor, not a fact-checker — it never verifies that a source is true, only which source backs the sentence. An unnumbered or dimmed marker means attribution was unavailable for that sentence.
The audit itself reports its status. A complete audit is silent; anything less shows an amber chip — Citation audit partial, failed, or timed out — which means some markers may be missing or inactive. Amber is a statement about the markers, not the scan: it does not mean the scan failed, nor that any finding is incorrect.
The Sources tab is where all three signals are totalled: the Trust Scorecard shows editorial claim coverage (how many editorial claims carry citations), the sourcing-confidence mix, the cross-reference mix, the tier mix of surfaced sources — Primary regulator · Supranational · Aggregator · Other source — and the tunnelling count, alongside a methodology panel explaining each signal in situ.
Reading one item end to end
Take a concrete reading. An item titled after an EU fund-liquidity proposal shows Consultation status, Likely, High impact, expected 15-10-2026, a summary whose second sentence carries [3], a Primary regulator source chip and a ✓ verified badge. What you now know: the measure is mid-lifecycle with a response window; the analysis regards it as probably landing; the claim rests on the regulator’s own document; the scan genuinely retrieved that document; and sentence-level attribution points at source 3 in this scan’s pool, which the Sources tab will show you in full. What you still do not know — and what no signal claims — is that every statement in the summary is true. The verification machinery narrows the trust gap; it does not close it.
The product ships with the same caution on every dashboard: AI-assisted analysis — always verify findings against primary sources before acting. The trust signals exist to make that verification fast — View source and the citation markers take you straight to the documents — not to substitute for it.
Why items are built this way
The design bets on a specific failure mode: in regulatory work, a wrong-but-confident summary is worse than a hedged one. Hence the separated signals (so one weak link never hides behind an average), the honest absences (TBD dates, unverified badges, dimmed markers), and the one-click path from any claim to its document. Items are also the substrate of change tracking — an item is bound to a tracked instrument, and it is instruments whose confirmed, dated events the Change Monitor reports between snapshots. Provenance discipline at the item level is what makes the change headline trustworthy one level up.
Where to go next
- The Sources tab — the trust surfaces in the product.
- What “changed” means — how item provenance becomes change evidence.
- Snapshots — the item fields on the wire for API callers.