The Observation Theory EncyclopediaFrom TSKAboutBy kindBy chapterBy Lean fileLedgerProvenance

rerank instrument

DefinitionReordering a candidate list with a second scorer. Its recall is at most the candidate coverage. Chapter 10 section 10.5.
ExampleReranking 100 candidates that hold 70 of the true neighbours can reach recall 0.7 and no more.
BookData Mining as Observation, draft 0.2, commit f3914f0; entry id rerank, kind instrument.
Statusno ledger row names this entry. Corrections: none recorded.
Defining equationnone
Assumptions and scope
  • Reordering a candidate list with a second scorer. A rerank cannot return a row the list does not hold, so its recall is at most the candidate coverage, an oracle rerank reaches the coverage exactly, and a deeper list can only raise the coverage.
  • Reranking the compressed candidates read 0.6627, identical to the unreranked recall, because candidate coverage was the wall, and the depth curve from 51 to 501 moved coverage from 0.696 to 0.926.
Prior artnone recorded
Evidenceturboquant-pro/docs/RESULTS_strata_phase23_gates.md:45-110, lean/DataMiningAsObservation/Rerank.lean
Reviewednot yet reviewed; generated 2026-09-10 from records at the commits on the provenance page.
recall is at most the candidate coverage
A second scorer reorders the candidate list and cannot add to it.

Equation

none

Conditions

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

Ledger

none

First stated

Chapter 10 section 10.5 of Data Mining as Observation, with the candidate-coverage finding in turboquant-pro/docs/RESULTS_strata_phase23_gates.md.

Measurements

Where the book states it Numbers, as the book’s sources table records them Source
chapter 10 section 10.5 rerank identical 0.6627, candidate coverage, depth curve 51 to 501, coverage 0.696 to 0.926, fidelity 0.663 to 0.702, originals rerank 0.923, 4096 bytes on 148 turboquant-pro/docs/RESULTS_strata_phase23_gates.md:45-110

Failures and corrections

none

Invariance envelope

none declared

Machine checked

lean/DataMiningAsObservation/Rerank.lean, theorems hits_le_coverage, recall_le_coverage, oracle_rerank, coverage_mono, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

Used in

Data Mining as Observation chapters 4, 10, 11, 12.

Related

recall at k; anti-hub recall; inverted file; rank certificate; retrieval-augmented pipeline.

See also

Book equations stated beside the entry’s terms, not defining it: 11.3, 10.7, 12.3.

Ledger rows that cite the entry’s records without naming it: NEG-14, GO-B-Llama, GO-B-Llama-rematch.

Sources-table rows that share a record with the entry without naming it: chapter 4 section 4.5, chapter 10 section 10.4, chapter 10 section 10.5, chapter 11 section 11.4, chapter 12 section 12.4.

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.

← replicaresidual →