For every company we cover, we publish full expectations ahead of the earnings release — revenue, GAAP and adjusted EPS, EBITDA, margins, capex, free cash flow, plus what we expect the company to guide. Each page carries its publication date and never changes after publishing. Updates are new pages; every prior version stays archived, permanently.
A slate of frontier AI models, each independently building a complete three-statement forecast from SEC filings we capture and verify byte-for-byte. Any number a model cites must exist in the filing it cites, or its run is rejected. The published figure for each metric is the median of the models' central estimates. The range shown on the three headline metrics is an 80% band — built so that roughly 8 of 10 reported prints land inside it, a rate we track and will publish.
When the company's 8-K lands, we compute the actuals with the same definitions used for the forecast, and the scorecard publishes — typically within minutes, not days.
surprise = reported − our published number
Positive means the company came in above our number. We never narrate the stock reaction at publish time — the page goes up before the reaction is knowable.
One function computes both our forecast metric and the graded actual, so we can never be accused of grading against a different definition than we published. Standardized metrics use one formula for every company — EBITDA is the textbook form, published with its components visible. Adjusted metrics follow each company's own definition, pinned to the company's reconciliation. If a company doesn't report a metric, the cell stays absent. If its presentation breaks a formula, the cell reads NOT COMPUTABLE. A dash is never an estimate.
Not resold Wall Street data. Pricedin's numbers are the consensus of our own model slate — built to make gated, institution-first estimates unnecessary. No access advantage, no herding, no waiting for a revision cycle.
Every estimate and every result stays up — hits and misses alike. Model-level performance, calibration, and the band-coverage rate will live on a public track-record page. The history is the product.