Track Record

What is recorded, and how it gets judged

Each verdict row stores the observed bonus, the tracked average and high, the CP Offer Score with its factor breakdown, a confidence tier derived from how many observations back it, and a reference to the exact scraper snapshot it was computed from. Cards without current live data are skipped and counted — never guessed.

A verdict is judged against what actually happened next: after 30 days we compare the offers observed in that window to the offer at verdict time. "Consider waiting" is correct if a materially better offer (≥10% higher) appeared; "apply now" and "neutral" are correct if none did. Too little data in the window is recorded as inconclusive, not guessed.

The scoring formula is versioned, so a future formula change can never silently mix differently-computed scores in one record. Full details in the methodology.

What this record can and cannot prove yet

It cannot prove accuracy yet. 0 outcomes have matured so far; we do not publish hit-rates or calibration percentages until at least 30 verdicts have completed their evaluation window, because a percentage computed from a handful of cases is noise. The first verdicts reach their horizon on September 11, 2026.

There is no backfill. Rows exist only from the real start date forward — a track record recorded after the fact would not be one. If we ever publish reconstructions from archived data, they will be a separate, explicitly labeled dataset.