Backtest // walk-forward, no lookahead

Does our confidence label mean anything?

Every projected earnings date on this site carries a label: high, medium or low confidence. A label like that is worthless unless the higher tiers are actually more accurate, so this page checks. A confidence label that does not track accuracy is decoration, and decoration on a number people plan around is worse than no label at all.

4,781 scored projections · 135 companies · reports from 2005-09-14 to 2026-11-17
LabelScoredCompaniesMedian missWithin 3 daysMore than 3 days offWithin 7 daysMore than 7 days off
high4,6681351 d69.1%67.0 to 71.1%30.9%28.9 to 33.0%91.1%89.5 to 92.6%8.9%7.4 to 10.5%
medium1131131 d59.3%50.8 to 67.9%40.7%32.1 to 49.2%88.5%82.4 to 93.8%11.5%6.2 to 17.6%
lownonenonenonenonenonenonenone

Small figures are 95% intervals: the companies, resampled with replacement 2,000 times from a fixed seed, so every build prints the same numbers, keeping the middle 95% of the results. Each draw keeps all of a company's quarters together, because a company that changes its schedule misses several quarters in a row. A tier's interval prints from 20 companies.

We cannot say the labels order correctly. On the point estimates the higher tier lands within seven days slightly more often (high 91.1% against medium 88.5%), but that held in only 80% of 2,000 company resamples, short of the 95% we require. The gap between the tiers is inside the noise.

Who is scored: every company in our earnings snapshot, 142 in all: the 81 most-watched large and mid caps on the earnings calendar and 61 small caps added for insider-buying coverage. The list was chosen for coverage, not for how well its dates project. 136 have five or more filed reports and enter the backtest; scoring starts at each one's fifth report, so 135 companies have scored points.

Read the sample sizes before the percentages. 97.6% of scored projections carry a single label, so the tiers are badly unbalanced. A label that reports the same value almost every time carries little information even when it is accurate, and that is a limitation of the engine, not a detail of this page. The low tier has zero scored points, and the reason is structural: scoring only begins at a company’s fifth report, which guarantees enough prior dates to reach a higher tier. This backtest therefore cannot test the bottom of the scale at all.

The live record: projections graded as published

The backtest above is a replay: its record runs from 2005-09-14 to 2026-11-17, over dates that were already filed. Live measurement began on 2026-08-19, when we started logging each projected window as we published it. Record from 2005-09-14, measured from 2026-08-19: a gap of 7,644 days. Everything before 2026-08-19 is simulation; this section is the part that was measured.

80.1%
inside the published window
95% interval 72.8 to 86.8%
19.9%
outside it
95% interval 13.2 to 27.2%
136
reports graded, on 135 companies

A rate prints from 20 graded reports on 20 companies. The interval: the companies, resampled with replacement 2,000 times from a fixed seed, so every build prints the same numbers, keeping the middle 95% of the results. Median days outside the window: 0.

TickerLoggedPublished windowReport filedResult
DIS2026-08-192026-11-07 to 2026-11-172026-11-17inside
HD2026-08-192026-11-12 to 2026-11-222026-11-17inside
DGII2026-08-192026-11-06 to 2026-11-162026-11-12inside
CEVA2026-08-192026-11-04 to 2026-11-142026-11-09inside
MARA2026-08-192026-10-29 to 2026-11-082026-11-091 day outside
KOP2026-08-192026-11-01 to 2026-11-112026-11-06inside
NHC2026-08-192026-11-01 to 2026-11-112026-11-06inside
REI2026-08-192026-11-01 to 2026-11-112026-11-06inside
USPH2026-08-192026-10-31 to 2026-11-102026-11-06inside
ABNB2026-08-192026-10-31 to 2026-11-102026-11-05inside
COP2026-08-192026-10-31 to 2026-11-102026-11-05inside
CRK2026-08-192026-10-28 to 2026-11-072026-11-05inside
NOG2026-08-192026-10-31 to 2026-11-102026-11-05inside
LMAT2026-08-192026-10-31 to 2026-11-102026-11-04inside
MCD2026-08-192026-10-30 to 2026-11-092026-11-04inside
MGY2026-10-122026-10-23 to 2026-11-022026-11-042 days outside
MTRN2026-08-192026-10-23 to 2026-11-022026-11-042 days outside
MWA2026-08-192026-10-31 to 2026-11-102026-11-04inside
PARR2026-08-192026-10-29 to 2026-11-082026-11-04inside
QCOM2026-08-192026-10-30 to 2026-11-092026-11-04inside
UBER2026-08-192026-10-29 to 2026-11-082026-11-04inside
ADUS2026-08-192026-10-28 to 2026-11-072026-11-03inside
AMD2026-08-192026-10-29 to 2026-11-082026-11-03inside
ATRC2026-08-192026-10-23 to 2026-11-022026-11-031 day outside
CRVL2026-08-192026-10-29 to 2026-11-082026-11-03inside
PFE2026-08-192026-10-29 to 2026-11-082026-11-03inside
SHOP2026-08-192026-10-29 to 2026-11-082026-11-03inside
SMCI2026-08-192026-10-29 to 2026-11-082026-11-03inside
TALO2026-09-222026-10-31 to 2026-11-102026-11-03inside
XYZ2026-08-192026-10-31 to 2026-11-102026-11-03inside
AAPL2026-08-192026-10-24 to 2026-11-032026-11-02inside
ALG2026-08-192026-10-31 to 2026-11-102026-11-02inside
NSSC2026-08-242026-10-28 to 2026-11-072026-11-02inside
PLTR2026-08-192026-10-28 to 2026-11-072026-11-02inside
TNC2026-08-192026-10-28 to 2026-11-072026-11-02inside
ABBV2026-10-122026-10-25 to 2026-11-042026-10-30inside
CVX2026-08-192026-10-25 to 2026-11-042026-10-30inside
AMZN2026-08-192026-10-24 to 2026-11-032026-10-29inside
CAT2026-08-192026-10-23 to 2026-11-022026-10-29inside
COIN2026-08-192026-10-24 to 2026-11-032026-10-29inside

How this is measured

The backtest is not a track record. It replays the engine over dates that were already filed. It is not the same claim as the max pain accuracy ledger, which records what we published as we published it and refuses backfill. The live record above is our track record for earnings dates, and it starts the day logging started.

Methodology →Every register on one rule →The parser behind confirmed dates →Max pain accuracy ledger →Earnings calendar →