Version 1.1.8 · Last updated October 3, 2026

Time, evidence, quality, and trust

Argus results are useful only when a caller interprets time and evidence together. The same value can be valid for one historical query and invalid for another.

Six different times

Field	Meaning	Common mistake
data_period	Period described by the value	Treating it as publication time
filing_time	When the source filed or published the information	Treating it as immediately available to Argus
known_time	Earliest time Argus considers the information knowable	Ignoring it and leaking future data
revision_time	When a later revision became available	Replacing the earlier historical view
generated_at	When Argus assembled the package	Using it as the fact's economic time
invocation_time	When the machine invoked the tool	Using it as source timing

For an as_of=T query, a consumer must not use a fact with known_time > T. Revisions create a new view; they do not rewrite what was knowable before the revision.

Evidence chain

Each FactRecord.evidence_ids entry must resolve to a SourceEvidence entry in the same package. Evidence identifies its source document or stored object, fragment/page/field position, filing and retrieval times, parser version, confidence, credibility, and conflict state.

Use this order when validating a fact:

1. Resolve every evidence id.
2. Confirm the pointer is specific enough to reproduce the extraction.
3. Check credibility_level and extraction_method.
4. For model-assisted facts, require confidence and a non-trivial review status.
5. Inspect conflicting evidence and the related quality issue before choosing a value.

Quality does not mean truth

data_quality.quality_level summarizes checks such as missing fields, unit or currency mismatch, source conflict, outliers, duplicates, revisions, confidence, and license restrictions. It is not an accuracy guarantee. A high-quality package can still reflect an incorrect source disclosure; a low-quality package can still contain individually usable facts.

When requires_human_review is true, downstream agents should stop automatic consumption and preserve the issue list, affected field, and evidence ids for a reviewer.

Credibility and trust

Credibility records source provenance, not a directional view of a company. The levels range from regulatory originals and company announcements through authorized vendor/system-collected data to model-assisted or unverified caller input. Trust filtering may exclude facts below a requested level, but it must not silently upgrade remaining facts.

License and output policy

License controls whether the authenticated institution may use a source for a declared purpose and whether redistribution is allowed. Output policy controls what Argus may return even when the underlying data is licensed. A result can be licensed for factual lookup while still being prohibited from investment advice, target-price generation, trading signals, portfolio allocation, or automatic orders.

Preserve allowed_uses, prohibited_uses, restricted_fields, redistribution, and output_restrictions when caching, transforming, or forwarding a package.