Economic evidence and accounting scope

Economic analysis starts from dated observations, then applies deterministic calculations, freezes a snapshot and optionally asks a model to interpret that exact evidence. Model prose cannot create an unavailable financial value or establish a causal relationship by itself.

Observations and revisions

Raw source responses are content-addressed. The cache tracks when a response was checked and when that revision was first observed. A derived snapshot references those revisions and passes an integrity check when reopened. Publication time, observation period, announcement date and system observation time have separate meanings.

Current public sources can revise older values. The system therefore marks these snapshots as ineligible for historical point-in-time experiments. Shifting current GDP values backward, or using a report-period date as a release date, would introduce future information. The existing Alpha PIT import and availability rules remain the route for qualified historical research.

Financial periods

Cumulative income and cash-flow statements measure a year-to-date interval. Balance sheets measure a point in time. Subtracting Q1 cumulative income from half-year cumulative income gives Q2 income only if both belong to the same year and accounting scope. Missing intervening periods prevent that calculation.

TTM sums four consecutive individual quarters, rather than four cumulative reports. A missing quarter invalidates the result. Year-on-year percentages require a positive comparable base; a turnaround from a loss needs the original values and an explicit state rather than an ordinary growth percentage.

Cash-flow quality ratios use consolidated net profit when available. Consolidated net profit can be derived from total profit less income-tax expense only when both inputs are available in the same statement period. Reported weighted cumulative ROE and reported gross margin retain the source definition; the application does not silently annualize ROE.

Absent borrowing fields are unknown. A sum of known borrowing and bond components is labeled as a partial subtotal and is not a statement of total interest-bearing debt. Banks, insurers and other financial institutions require different economics and do not receive generic industrial-company judgments.

Comparability and interpretation

Comparisons require the same company identity, reporting endpoint, currency and metric basis. Selected peers must share the industry and period. A small user-selected sample does not establish an industry percentile. A current quote cannot create a historical valuation distribution.

The selected endpoint bounds financial trend and report evidence even if the raw current response includes later reports. Peer identifiers are normalized before deduplication. Simulated provenance propagates to the combined comparison and model evidence. Legacy stock reports cannot silently attach today's revisions to a historical research date; an explicit frozen snapshot must also pass company and availability checks.

GDP cumulative growth, price growth, PMI levels, monthly financing increments and annualized interest rates have different units and frequencies. They use separate axes and explicit definitions. Stale observations stay visible. Policy and industry transmission remain interpretations unless supported by separate evidence.

The report workflow preserves ordinary text generation, json_repair parsing and textual ReAct. Each substantive section cites actual supplied evidence. Reports preserve gaps, do not produce unsupported forecasts and do not automatically change strategy scores, portfolio weights or orders.