Prediction Accuracy

How often the published odds have been right: hit rates by predicted category, calibration by odds band, and the per-run vintages, scored against actual FCC dispositions.

What counts as a hit: a granted prediction was right when the application was actually granted, and a dismissed one when it was actually dismissed or withdrawn. A prediction becomes scorable only after its expected process date has passed and the FCC's files show a terminal disposition.

The calibration table groups each scored choice by the decile band its predicted grant odds fell into and compares the share actually granted against the band. Unknown-callsign rows — the grant could not be attributed to the requested callsign — count in the band's Scored and Unknown Callsign columns but never in Granted or in the observed rate.

Applications still awaiting a disposition sit in their own pending bucket and pad neither side of any rate. The date panel is the signed gap in days between the actual and predicted process dates — negative means the actual disposition came earlier — and undated rows are excluded rather than substituted.

The over-time table reads as vintages: each row is one prediction run scored on its own, so an improvement to the rules shows as a step between adjacent vintages.

Rates over fewer than 20 decided predictions show as too few to judge: below that floor a single prediction moves the rate by more than 5 percentage points, so a percentage would imply precision the cohort does not have. Counts always render.

Score run #130 — finished 2026-09-24T00:50:06.587792Z (applied) — process date 2026-09-23 — 113 scored runs

Accuracy by Category (All Runs) (4)
CategoryScoredCorrectIncorrectUnknown CallsignHit Rate
Granted Certain9900too few to judge
Granted Unknown0000too few to judge
Dismissed Likely0000too few to judge
Dismissed Certain1058718083%

Each row is one prediction-run vintage — the report as it stood after that run — so an improvement to the rules shows as a step between adjacent vintages rather than smeared into the lifetime average above.

Accuracy Over Time (13)
RunGranted CertainGranted UnknownDismissed LikelyDismissed Certain
Run #67 — 2026-09-130too few to judge0too few to judge0too few to judge2too few to judge
Run #66 — 2026-09-130too few to judge0too few to judge0too few to judge2too few to judge
Run #65 — 2026-09-130too few to judge0too few to judge0too few to judge2too few to judge
Run #64 — 2026-09-130too few to judge0too few to judge0too few to judge2too few to judge
Run #63 — 2026-09-131too few to judge0too few to judge0too few to judge5too few to judge
Run #62 — 2026-09-131too few to judge0too few to judge0too few to judge5too few to judge
Run #61 — 2026-09-131too few to judge0too few to judge0too few to judge5too few to judge
Run #60 — 2026-09-131too few to judge0too few to judge0too few to judge5too few to judge
Run #59 — 2026-09-131too few to judge0too few to judge0too few to judge5too few to judge
Run #58 — 2026-09-121too few to judge0too few to judge0too few to judge5too few to judge
Run #57 — 2026-09-121too few to judge0too few to judge0too few to judge5too few to judge
Run #56 — 2026-09-121too few to judge0too few to judge0too few to judge5too few to judge
Run #55 — 2026-09-121too few to judge0too few to judge0too few to judge5too few to judge
← PrevNext →
Calibration (10)
Probability BandScoredGrantedUnknown CallsignObserved Grant Rate
0%–10%105000%
10%–20%000too few to judge
20%–30%000too few to judge
30%–40%000too few to judge
40%–50%000too few to judge
50%–60%000too few to judge
60%–70%000too few to judge
70%–80%000too few to judge
80%–90%000too few to judge
90%–100%900too few to judge
Still Pending

These 4859 applications have not been scored yet — they sit in their own bucket and pad neither side of any rate above. 4859 are overdue (their predicted process date has passed); 0 are not yet due.

113 Granted Certain0 Granted Unknown226 Dismissed Likely4520 Dismissed Certain

Date Accuracy

114 scored rows carry a date error — the signed gap in days between the actual and predicted process dates, negative meaning the FCC acted earlier than predicted. Mean signed error: -7.631578947368421 days; mean absolute error: 7.631578947368421 days. 0 scored rows have no actual disposition date and are excluded rather than substituted.

← Full prediction summary

FCC Data Freshness
Derived2026-09-24T00:20:07.046885Z
license-weeklyl_amat.zip — applied 2026-09-20T16:18:23.306716Z
license-dailyl_am_tue.zip — applied 2026-09-23T18:01:16.692874Z
application-weeklya_amat.zip — applied 2026-09-20T16:18:23.410853Z
applied with 2 quarantined USI(s) (HD:1, VC:1) — e.g. 13141547, 13743541 — those records retain their previous values and are known stale
application-dailya_am_tue.zip — applied 2026-09-23T18:01:16.698065Z