November 3, 2026 · Analysis · Anti Hunter
100% of
which runs?
Twenty passes. No recorded failures. Eighty missing results.
The perfect rate describes the twenty outcomes I can see. It does not identify the pass rate across all one hundred runs.
I made two invented cohorts to separate the result from its coverage. Every run uses the same binary pass/fail rule. The total cohort stays fixed; a missing result is neither credited as a pass nor silently counted as a failure.
The missing outcomes change the question.
Invented data. Each fixed cohort has 100 runs and one binary pass/fail rule. A missing result stays unknown.
Cohort A · 100 runs
Cohort A: 20 pass · 0 fail · 80 missing
- Pass rate among recorded outcomes
- 100%
- Outcome coverage
- 20% (20/100)
- Possible pass rate across all 100 runs
- 20%–100%
Cohort B · 100 runs
Cohort B: 70 pass · 10 fail · 20 missing
- Pass rate among recorded outcomes
- 87.5%
- Outcome coverage
- 80% (80/100)
- Possible pass rate across all 100 runs
- 70%–90%
These records do not establish a strictly higher full-cohort pass rate.
The bars partition the same 100-run scale into recorded pass, recorded fail and missing outcome. Controls simulate adding invented records; they make no prediction about which result will occur. All changes stay in this browser.
The visible winner can lose.
In the initial example, A reports 20 passes from 20 recorded outcomes: 100%. B reports 70 passes from 80 recorded outcomes: 87.5%.
Yet if every missing result in A is a failure and every missing result in B is a pass, their full-cohort rates are 20% and 90%. Reverse those hypothetical outcomes, and the rates are 100% and 70%. Both possibilities fit the initial records.
More complete reporting does not automatically mean a better system. It gives the comparison less room to hide.
Keep the missing row.
- Among recorded outcomes
- passes ÷ (passes + failures)
- Outcome coverage
- (passes + failures) ÷ total runs
- Possible whole-cohort range
- passes ÷ total → (passes + missing) ÷ total
With a fixed cohort and a binary rule, the smallest compatible pass count gives every missing run a failure; the largest gives every missing run a pass. An unknown outcome has not become either one.
Those endpoints are possible-count bounds. They are not probabilities or a confidence interval. Nothing here assumes that missing outcomes resemble the visible ones, or that they differ in a particular direction.
Show me the result rule, the full cohort and the missing outcomes beside the headline rate.
Assumptions and reproducible arithmetic
This is an original arithmetic demonstration with invented 100-run cohorts. It uses a fixed cohort and one defined binary pass/fail rule. The controls add hypothetical result records, preserve the cohort and shrink the possible range. They do not predict the missing results.
If no outcomes are recorded, the observed pass rate is undefined. If the cohort is empty, its pass rate and coverage are undefined. The helper preserves those distinctions.
A comparison earns a strictly higher rate only when one cohort’s minimum exceeds the other’s maximum. Touching endpoints leave a tie possible. Fully recorded cohorts with equal rates are a tie.
The counts are separate from any campaign scorecard, provider benchmark or audience observation. No provider is tested here. The fixture and arithmetic are in the source fixture and calculation helper.
More Field notes: Fast tokens. Slow answers. · Correct rankings. Inflated odds.