How much a filed number moved.
For every issuer figure reported more than once, this measures the restatement magnitude — (latest − first) / |first| — across Certvas's point-in-time vintages. Large or one-sided revisions are a recognised earnings-quality read, and this is uniquely possible because AF-FUND keeps every vintage, never overwriting.
Grain: issuer × period × line item (revised only) · point-in-time by observed_at · deterministic · on the licence-clean AF-FUND spine · 4 provenance guarantees
Panel dictionary.
| Field | Type | Description |
|---|---|---|
dataco_id | string | Certvas ID of the issuer. |
period_end / period_type | date / enum | The reporting period that was restated (FY / H1 / Qn). |
line_item | string | The taxonomy line that was revised (e.g. net_income). |
first_value / latest_value | number | Earliest- and latest-observed reported values (by observed_at). |
revision_pct | number | (latest − first) / |first| × 100 — the restatement magnitude and direction. |
n_vintages | integer | How many distinct times the figure was observed. |
first_observed_at / latest_observed_at | date | When each value became knowable. |
Honest coverage: a row exists only where a figure was observed at ≥ 2 distinct vintages (an actual restatement), so this signal accrues from ingest forward and is thin (often empty) until prior periods are re-observed — then it grows automatically as the daily AF-FUND extractor re-covers issuers. A revision is not a judgement of error; it can be a legitimate audit adjustment. We publish exactly what has been re-observed rather than overclaim.