Accuracy

54.7% of our estimated earnings dates landed within 3 days of the actual report, measured on 23,881 releases. This page is the method behind that number, the full current run and the ways it fails.

What we measure

The product's claim is the estimated date, so the estimated date is what we score. Every past release in the window is replayed: we rewind to five days after the quarter ended, run the estimator on only the filings visible at that moment, and compare the projected date with the actual 8-K date. Foreign private issuers are replayed against their 6-K results and included in the published number.

The run repeats weekly on the full universe. Partial runs are never published. No figure on this page is typed by hand: the backtest writes the result and this page renders it from /v1/coverage, the same endpoint anyone can read without a key.

What counts as a hit

Exact means the same calendar date. Within 1 day means one day either side, and the same for 3 and 7 days. Percentages are of all releases in the window, not of the releases we scored: a release we gave no estimate for counts as a miss. A date the company had already announced is not scored as an estimate; the test only grades projections.

The current run

Measured 2026-11-15 on 23,881 releases from 3,119 companies since 2024-11-15, each estimated 5 days after its quarter closed.

31.6%exact date
46.0%within 1 day
54.7%within 3 days
78.4%within 7 days
23,881releases scored
20,484had an estimate; the rest count as misses
3,119companies
1.0median error, days

By confidence band

Every estimate carries a confidence from 0 to 1, meant to equal the probability of landing within 1 day. This table is the check: in each band, mean confidence should track the observed within-1-day rate. A band whose hit rate runs above its mean confidence is under-confident, which is the safe side of wrong.

Confidence bandReleasesExactWithin 1 dayWithin 3 daysMean confidence
0.60 and up5,33360.5%67.4%71.4%0.67
0.45 to 0.5910,78432.9%53.9%65.7%0.49
0.30 to 0.442,70420.9%43.6%60.9%0.36
under 0.301,66312.2%24.1%31.8%0.20
no estimate3,3970.0%0.0%0.0%–

By company size

Smaller companies have shorter and less regular filing histories, so their estimates are less reliable. The confidence score reflects that, and this table shows it.

Market-cap bucketReleasesExactWithin 1 dayWithin 3 days
small ($300M to $2B)8,95329.0%43.9%53.4%
mid ($2B to $10B)8,00031.8%47.0%55.2%
large ($10B to $200B)6,37234.1%47.1%55.4%
mega ($200B and up)55639.4%52.5%61.2%

Known failure modes

The largest error class is a miss by exactly seven days: the right weekday in the adjacent week. Many of these are coin flips. A company that has reported on a Thursday in late October four years running can pick either Thursday this year, and nothing in its filings says which. We tested four alternative projections on 1,695 past releases in November 2026; every one scored lower on exact, within 1 and within 3 days, so the current projection stands.

Confidence drops when the history is thin or irregular: few past releases, a lag from quarter end that jumps around, or a stale last filing. A company with one usable release gets a confidence of 0.40 at most, and a lag spread above 3 days caps it at 0.27. When there is no usable history at all the API answers unknown rather than guessing, and that release counts as a miss here.

Runs from different dates are not comparable as a trend. The universe and the estimator both change over time, so an older published figure was measured on a different set of companies. For the current figure read this page or /v1/coverage; do not quote an old one.

An estimated date is a projection from the company's own filing history, never a company statement. The API labels it estimated with its confidence, and confirmed dates link the announcement. Method details and the live service state: docs and status.