encoder concept

Definition	A model that maps an input to an embedding. Chapter 12.
Example	LaBSE maps a sentence in any of its languages to a 768-dimensional vector.
Book	Data Mining as Observation, draft 0.2, commit `f3914f0`; entry id `encoder`, kind concept.
Status	no ledger row names this entry. Corrections: none recorded.
Defining equation	Book equation 12.3.
Assumptions and scope	A model that maps an input to an embedding. An encoder votes only when its held-out AUROC on a corpus it was not trained on clears both nulls by the preregistered margin, its authority is its reliability weight, and a strictly monotone recalibration of its score changes neither the verdict nor the weight. The rights encoder fell below the untrained baseline and was retired rather than tuned, and the legal-citation encoder’s flip is the ledger’s demonstrated row. An encoder’s held-out AUROC depends on the orientation of its score. A score that ranks the wrong way scores 1 minus A, and the gate reads the score as the encoder produces it and not the better of the two orientations.
Prior art	none recorded
Evidence	`lean/DataMiningAsObservation/CrossCorpusGate.lean`
Reviewed	semantic review 2026-09-06; generated 2026-09-10 from records at the commits on the provenance page.

A model that maps an input to an embedding.

Equation

Book equation 12.3.

\[\text{validated}\iff \mathrm{AUROC}_{\text{cross}}-\max\big(\mathrm{AUROC}_{\text{untrained}},\ \mathrm{AUROC}_{\text{BoW}}\big)\ \ge\ 0.10.\]

Book equation 14.1.

\[S=\frac{\sum_i w_i\,s_i}{\sum_i w_i},\qquad w_i=\max\big(0,\ 2\cdot\mathrm{AUROC}_i-1\big).\]

Conditions

A model that maps an input to an embedding. An encoder votes only when its held-out AUROC on a corpus it was not trained on clears both nulls by the preregistered margin, its authority is its reliability weight, and a strictly monotone recalibration of its score changes neither the verdict nor the weight.
The rights encoder fell below the untrained baseline and was retired rather than tuned, and the legal-citation encoder’s flip is the ledger’s demonstrated row.
An encoder’s held-out AUROC depends on the orientation of its score. A score that ranks the wrong way scores 1 minus A, and the gate reads the score as the encoder produces it and not the better of the two orientations.

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

Ledger

none

First stated

Chapter 12 section 12.5 of Data Mining as Observation, with the program’s encoders in xbse and the calibrated authority of gtc-prototype/docs/CALIBRATED_AUTHORITY.md:19-63.

Measurements

Where the book states it	Numbers, as the book’s sources table records them	Source
chapter 8 section 8.6	rights encoder 0.467 below untrained baseline, the AUROC lesson, cross-corpus same-sign fix	`xbse/README.md:160-172`; `xbse/experiments/rights_r6_summary.json`

Failures and corrections

none

Invariance envelope

none declared

Machine checked

lean/DataMiningAsObservation/CrossCorpusGate.lean, theorems clears_bow, clears_untrained, not_validated_of_saturated, margin_example, validated_comp, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

Used in

Data Mining as Observation chapters 0, 7, 8, 9, 10, 11, 12, 14.

embedding; cross-corpus gate; reliability weight; retrieval-augmented pipeline.

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.

← embeddingensemble →