Model confidence
Training and probability calibration require additional point-in-time observations. Four quarters per stock are for the research pilot. These preliminary research estimates have not yet been statistically calibrated; model confidence cannot be estimated reliably.
Model accuracy · ongoing tracker
Actual historical earnings ≠ model forecasts. We only count predictions that were timestamped and locked before earnings.
Data coverage
Sources: issuer releases and SEC filings. One historical PLTR consensus pair is included as a separately sourced reference.
Methodology and source standards
EPS is adjusted/non-GAAP diluted EPS as published by each issuer. Revenue is reported quarterly sales in USD billions. SMCI uses fiscal quarters ending September 30, December 31, March 31 and June 30; AMD and PLTR use calendar quarters. Historical beat rates are not calculated without matched pre-announcement analyst consensus. Company guidance is never mislabeled as analyst consensus.
Forecast gauges display current Hybrid research snapshots for every stock when the required inputs qualify; otherwise the stock shows a pending state. These are preliminary, statistically uncalibrated research estimates—not validated probabilities. The displayed percentages have not yet been shown to correspond to their stated frequencies. Historical heuristic factor adjustments are illustrative rather than fitted model contributions. Confirm point-in-time consensus and validate out-of-sample before marketing these as predictive probabilities.