HCMA Paper 4: Governance-Native Retrieval Enables Conformal Calibration Under Temporal Distribution Shift

HCMA Paper 4: conformal abstention over a governance-native multi-version retrieval architecture (HCMA). We show that native version governance (version collapse at consolidation time) induces calibratable retrieval confidence under temporal distribution shift, where post-hoc governance and pretrained semantic rerankers fail when distractors are plausible but temporally outdated. Key results (deterministic, reproducible): • NQ plausible temporal drift: HCMA 0.973 / full conformal coverage (α=0.05) vs pretrained bi-encoder + reranker 0.432 / 0.000. • Governance ablation (causal): ungoverned 0.158 → version-collapse 0.807 → full HCMA 0.807; post-hoc collapse 0.140 (spurious margin). • Pretrained dissociation: pretrained wins on paraphrase (0.878 vs 0.711) but collapses on temporal drift (0.432 vs 0.973) — semantic similarity is orthogonal to temporal validity. License: CC-BY 4.0 (paper/docs/data); Apache-2.0 (code, SPDX headers inline). SHA-256 integrity manifest: manifest.json

Authors

Publication Details

Journal
Open Science Framework
Published
2026-10-30
DOI
https://doi.org/10.17605/osf.io/ghvmz
Primary Topic
Information Retrieval and Search Behavior
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

HCMA Paper 4: Governance-Native Retrieval Enables Conformal Calibration Under Temporal Distribution Shift

Mauro Javier Silva
3 links
Open Science Framework
Information Retrieval and Search Behavior
article

HCMA Paper 4: Governance-Native Retrieval Enables Conformal Calibration Under Temporal Distribution Shift

Mauro Javier Silva
article en

Abstract

HCMA Paper 4: conformal abstention over a governance-native multi-version retrieval architecture (HCMA). We show that native version governance (version collapse at consolidation time) induces calibratable retrieval confidence under temporal distribution shift, where post-hoc governance and pretrained semantic rerankers fail when distractors are plausible but temporally outdated. Key results (deterministic, reproducible): • NQ plausible temporal drift: HCMA 0.973 / full conformal coverage (α=0.05) vs pretrained bi-encoder + reranker 0.432 / 0.000. • Governance ablation (causal): ungoverned 0.158 → version-collapse 0.807 → full HCMA 0.807; post-hoc collapse 0.140 (spurious margin). • Pretrained dissociation: pretrained wins on paraphrase (0.878 vs 0.711) but collapses on temporal drift (0.432 vs 0.973) — semantic similarity is orthogonal to temporal validity. License: CC-BY 4.0 (paper/docs/data); Apache-2.0 (code, SPDX headers inline). SHA-256 integrity manifest: manifest.json

Open Science Framework
Openalex Percentile: Top 6%
Information Retrieval and Search Behavior
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon