MERIDIANGlossary — guided explanations
What Meridian's labels, badges and states mean, and why values are sometimes deliberately absent. This page explains how this build works; it contains explanatory prose only, no data values.
Value classes — how every displayed value is labelled
Every value in the interface is one of these six classes; an inference is never dressed up as a fact.
| Badge | Class | Meaning |
|---|---|---|
| SF | Source Fact | The value exactly as published by the named source — e.g. an ISIN, listing segment or auditor from the SIX issuer directory, or an address from GLEIF. Republished unchanged, with the retrieval trail attached. |
| CD | Company Disclosure | Stated by the company itself — e.g. the next general-meeting date an issuer communicates via the exchange. It is the company's own claim, faithfully reproduced; Meridian does not adopt it as its own conclusion. |
| ORF | Official Regulatory Fact | Taken from an official legal or regulatory source (Fedlex, FINMA feeds, EU official publications, sanctions lists): titles, publication dates, measures — text as published. |
| CM | Calculated Metric | Computed by Meridian from sourced inputs and labelled with its formula and version — e.g. SMA 20/50, 52-week range position, the Geneva-link classification, and the impact scores (impact-1.1), whose full factor breakdown is shown on each match page. Inputs are always sourced values. |
| EBI | Evidence-Based Inference | A candidate conclusion derived from evidence — e.g. a rule-generated company-regulation match. Always shown together with its evidence and never presented as a disclosed fact. |
| UNV | Unverified | No verified source, or not enough evidence. Shown honestly as unavailable or unverified — never replaced with a guess. |
Coverage states — what the source badges mean
Every connected source carries exactly one of 14 coverage states, shown as badges on Source Health; when anything is failing, unverified or awaiting credentials, a banner appears on every page — gaps are shown, never concealed. Seven states are assigned automatically by the current ingestion runner (confirmed current, current within declared delay, polling delayed, authentication required, source failure, parser failure, and the coverage-unverified default); the other seven are defined vocabulary reserved for manual assignment and planned checks.
| Badge | Meaning |
|---|---|
| Confirmed current | The last live fetch succeeded; the source is considered current. |
| Current within declared delay | Loaded from a recorded capture at most 48 hours old — current within the source's declared publication delay. |
| Polling delayed | Data is older than expected (stale sweep, or replay of an older capture); a refresh is due. |
| Provider delayed | The provider itself delivers delayed data (e.g. delayed exchange quotes); per-quote delay notes and freshness labels carry this meanwhile. |
| Backfill in progress | A known gap is being filled; history is incomplete meanwhile. |
| Possible publication gap | The source may have published items Meridian did not capture (e.g. a finite feed window was exceeded). |
| Authentication required | Credential-gated adapter without configured credentials — implemented, inactive pending configured credentials (see .env.example). |
| Quota exhausted | The provider's quota or rate budget is used up. |
| Source failure | A live fetch failed (network/HTTP error); a coverage incident is open. |
| Parser failure | The payload was fetched but parsed to zero records or the parser raised; a coverage incident is open. |
| Schema change detected | The response shape no longer matches what the parser expects. |
| Reconciliation failed | A cross-source consistency check failed. |
| Coverage unverified | Never successfully run (the default state at seed time). |
| Source retired | Deliberately decommissioned (e.g. a superseded list edition). |
Legal stage
Every regulatory record carries a legal stage — where the text stands in its lifecycle. The feeds connected in this build are publication feeds (Fedlex SPARQL, FINMA announcements, EU official-journal notices), so every ingested instrument is currently recorded at stage published: the source published it as an official text. The vocabulary matters because a proposal or consultation draft is not binding law: Meridian never describes a proposal as current binding law, and never infers a stage the source did not state. Structured lifecycle dates (entry into force, application) are extracted to the Calendar when a source publishes them.
Why a score can read “Insufficient evidence to calculate”
Every impact score is a weighted average of named factors (score = Σ weight × value ÷ Σ weight; formula version impact-1.1), and every factor is stored and displayed. A score whose inputs do not exist is stored as None and rendered as “Insufficient evidence to calculate” — never as a guessed number. Concretely: urgency needs a real publication date; exposure and materiality need company-disclosed segment or geographic figures, which are not yet ingested (filing parsing is a later phase), so they are None on every current assessment; market sensitivity is reserved. Each missing material input also lowers confidence through an explicit missing-data penalty (0.25 per missing item) — gaps reduce confidence, they are never filled in. Details: Methodology.
What a candidate match is
Every company–regulation link Meridian generates is a candidate: produced by a named rule (e.g. R-FINMA-BANK), assigned a linkage layer, stored with review status needs_review, and accompanied by quoted evidence rows plus the full score-contribution table. A candidate announces itself with an informational alert and waits for human review — nothing auto-promotes to a confirmed relationship. Name similarity alone can never establish identity: the sanctions name-screen rule carries a fixed requires-human-review factor for exactly that reason, and counter-evidence (e.g. a non-Swiss domicile against a Swiss instrument) is recorded rather than discarded.
What the provenance chip (ⓘ) shows
The small ⓘ chip next to a value summarises that row's provenance: source key, retrieval timestamp (UTC), publication timestamp when the source provides one, and parser version — hover to read it. The underlying database row additionally stores the canonical source URL, a reference to the immutable raw artefact captured at ingestion, and its SHA-256 content hash, so every external fact keeps its full journey from source to screen. A value without a verified source is shown as an explicit unavailable state with the reason — never silently blank, never invented.