Public methodology
How a disclosure becomes a defensible data point
ASI is a reference layer over company disclosures. It preserves the source, reported wording, extraction method, quality checks and competing values behind the answer it serves.
1. Source acquisition
Sources are company filings, annual reports, integrated reports, sustainability reports and recognized disclosure programs. Duplicate wrappers and interim documents are excluded from the primary ledger. Document identity and year remain attached to every extracted candidate.
2. Extraction and normalization
Values are read from structured filing fields or source documents. The reported figure is retained alongside a normalized value and unit. Turkish number formats, unit conversions, bounds and entity attribution pass deterministic guards before a candidate can enter the resolved record.
High: direct or strongly corroborated evidence.
Medium: defensible extraction with review context.
Low: usable only with the warning and evidence visible.
Confidence describes extraction reliability. It is not data completeness, classification confidence or an ESG score.
3. Resolution and dissent
Multiple clean candidates for the same company, metric and year are ranked using source authority, evidence quality, extraction method, confidence and recency. One golden value is served; alternate bases and genuine disagreement are never deleted. Corroboration can raise confidence. Dissent lowers it and excludes an unresolved metric from percentile calculations.
The live corpus currently carries 2,311 resolved metric-years with visible dissent.
4. Evidence and coverage
A page citation requires a named document and a numbered page. Structured filing evidence has no PDF page, so its link opens the submitted filing section instead. The current page-citation rate is 59.7% across 42,610 resolved data points.
The reviewed sources do not report the metric.
A source exists, but the metric is not yet defensibly extracted.
5. Known limitations
- — Historical depth varies by company and sector; no missing year is interpolated.
- — A disclosed claim records what the company said, not independent proof that the claim is effective.
- — Peer percentiles depend on comparable metric definitions, reporting years and production boundaries.
- — CBAM/ETS outputs are screening and pricing intelligence, never declaration-grade calculations or legal advice.
- — Confidence remains heuristic until the independently adjudicated 300-record calibration is published.