stratification instrument

Definition	Splitting rows into strata, by difficulty or by group, and scoring each separately, so that the report can take the minimum over strata. Chapter 10 and chapter 12. Also stratif.
Example	Fourteen strata by difficulty, seven eligible, two abstained for having 2 rows and 1.
Book	Data Mining as Observation, draft 0.2, commit `f3914f0`; entry id `stratification`, kind instrument.
Status	no ledger row names this entry. Corrections: none recorded.
Defining equation	Book equation 10.7.
Assumptions and scope	Splitting rows into strata, by difficulty or by group, and scoring each separately, so that the report can take the minimum over strata and count the strata too thin to score. A weighted aggregate over strata is at most the pass bar when one stratum fails by enough, and the folds of a stratified split test every row once. The first strata run scored seven of fourteen eligible strata and abstained on those with two rows and one, and the second prediction inverted, which the record keeps.
Prior art	none recorded
Evidence	`turboquant-pro/docs/RESULTS_strata_phase23_gates.md:1-45`, `lean/DataMiningAsObservation/Certificate.lean`, `lean/DataMiningAsObservation/CrossValidation.lean`
Reviewed	not yet reviewed; generated 2026-09-10 from records at the commits on the provenance page.

Rows scored in strata, so the minimum can be reported.

Equation

Book equation 10.7.

\[\text{verdict}=\min_{i:\ n_i\ge n_{\min},\ |Q_i|\ge q_{\min}}\ \mathrm{score}_i,\qquad \text{ABSTAIN otherwise}.\]

Conditions

Splitting rows into strata, by difficulty or by group, and scoring each separately, so that the report can take the minimum over strata and count the strata too thin to score. A weighted aggregate over strata is at most the pass bar when one stratum fails by enough, and the folds of a stratified split test every row once.
The first strata run scored seven of fourteen eligible strata and abstained on those with two rows and one, and the second prediction inverted, which the record keeps.

Conditions are curated in entries.toml rather than read from a record.

Ledger

none

First stated

Chapter 10 section 10.4 of Data Mining as Observation, with the strata design in turboquant-pro/docs/STRATA_RFC.md:24-98.

Measurements

Where the book states it	Numbers, as the book’s sources table records them	Source
chapter 10 section 10.5	Gate A design error, A prime skew 3.970 to 3.177, max 287 to 213, Robin Hood 0.372 to 0.261, fraction 0.117 to 0.079, compressed path 0.663 vs 0.90, seven strata, 0.62 to 0.69 vs 0.76 to 0.84	`turboquant-pro/docs/RESULTS_strata_phase23_gates.md:1-45`

Failures and corrections

none

Invariance envelope

none declared

Machine checked

lean/DataMiningAsObservation/Certificate.lean, theorems falseClear_mul_coverage, coverage_empty, falseClear_mem_unit, minOverStrata_passes_iff, minOverStrata_le_weighted_mean, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

lean/DataMiningAsObservation/CrossValidation.lean, theorems sizes_sum, accuracy_weighted, accuracy_mean_of_equal, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

Used in

Data Mining as Observation primer S, chapters 0, 5, 8, 10, 11, 12.

min-over-strata; abstention; anti-hub recall; cross-validation; harness.

Status

Generated 2026-09-10 by encyclopedia/generate.py; book at observation-data-mining f3914f0; the commit of every record is listed in the encyclopedia’s provenance.

← standardizationsum of squared errors →