Fundamentals

close: schema:MonetaryAmount

One extracted financial-statement line item, as filed, for one entity and period — and for one OBSERVATION of that value. Restatements arrive as new rows and nothing is ever overwritten, so (dataco_id, period_end, period_type, statement, line_item) is NOT unique; record_ref identifies the individual vintage. Two dates matter and they answer different questions: period_end is what the number describes, available_at is from when someone could have acted on it. A backtest must filter on both.

Slots

slot	range		description
`dataco_id`	`string`	required	Certvas entity identifier — a deterministic hash of the LEI where one exists, else a ULID. Stable across rebuilds, which is what lets it be a public, citable identifier.
`period_end`	`date`	required	THE EFFECTIVE DATE — the last day of the fiscal period the value describes. Filtering a backtest on this alone leaks: the number was not knowable until it was published.
`period_type`	`PeriodTypeEnum`	required	The length of the reporting period.
`statement`	`StatementEnum`	required	Which financial statement the line item belongs to.
`line_item`	`string`	required	Canonical taxonomy code for the line item, defined in the data dictionary.
`value_reported`	`decimal`	required	The value exactly as filed, in the currency as filed. Never rescaled.
`currency_reported`	`string`	required	ISO 4217 code of the currency the value was filed in.
`value_usd`	`decimal`	optional	The value converted to USD. Null where no conversion was applied.
`conversion_ref`	`string`	optional	FX source and method used for value_usd, so the conversion is reproducible.
`extraction_confidence`	`double`	required	Extractor confidence in [0,1]. Published rows are accepted extractions only (>= 0.85); anything lower goes to a human review queue rather than to a customer.
`source_page`	`integer`	optional	Page number within the source document, so a value can be checked against the filing.
`source_doc_hash`	`string`	optional	Content hash of the source filing in the raw archive.
`record_ref`	`string`	required	Stable identifier for THIS observation (this vintage) — not for the value's natural key. Deterministic across rebuilds, so a re-push updates a row in place while a genuinely new vintage lands as a new row.
`vintage_seq`	`integer`	required	1 for the originally filed value, 2 and upward for each later restatement.
`is_as_reported`	`boolean`	required	True where vintage_seq = 1. The as-reported view filters on THIS rather than on `restated = false`: a value revised twice has two non-latest vintages, so a flag-based filter would return the intermediate revision alongside the original and duplicate the key.
`restated`	`boolean`	optional	True where a LATER vintage carries a DIFFERENT value for the same key. DERIVED by comparing vintages, never read from an upstream flag — and a re-extraction producing the same number is not a restatement, so re-running the pipeline cannot manufacture phantom revisions.
`superseding_record_ref`	`string`	optional	record_ref of the earliest later vintage that changed the value.
`announcement_date`	`date`	optional	THE AVAILABILITY DATE — when the filing became public. Sourced as a fact and NULL when unevidenced, never inferred: every candidate must fall within [period_end, period_end + 1095 days], and one outside that window is discarded rather than clamped, because a repaired date is indistinguishable from an evidenced one downstream.
`announcement_source`	`AnnouncementSourceEnum`	optional	Which evidence tier announcement_date came from.
`available_at`	`date`	required	FILTER AS-OF QUERIES ON THIS. `coalesce(announcement_date, observed_at)`, never null. The fallback direction is load-bearing: observed_at is always later than true publication for a backfilled filing, so it can only ever EXCLUDE a row from an as-of window, never admit one before it was knowable. Poor announcement-date coverage costs recall, never correctness.
`filing_lag_days`	`integer`	optional	announcement_date minus period_end, in days. NULL where unevidenced — deliberately never defaulted to 0, which would read as same-day publication.
`observed_at`	`datetime`	required	When Certvas itself observed this value. Never a proxy for publication date.
`source_ref`	`string`	required	URL or document-hash pointer into the immutable raw archive.
`license_class`	`string`	required	License Register ID governing redistribution of THIS row. Per-row rather than per-table because a single table legitimately mixes sources with different terms.
`valid_from`	`date`	required	Start of the interval over which this assertion holds.
`valid_to`	`date`	optional	End of the validity interval; NULL means current. Half-open, so `valid_from <= t < valid_to` selects without double-counting the boundary.