FixPipe pre-registration v1.8 - module-level criterion, second pipeline (locked probe and freeze chain)

Frozen pre-registration for a cross-pipeline test of a module-retention criterion for LLM pipelines. Criterion: a module pays for itself only when it (i) owns an exclusive, verifiable information source and (ii) can convert it into an action on the artifact rather than into prose. FixPipe is a generate-check-repair pipeline over GitHub Issues with no vote signal. Amendment 1 (v1.1) replaced the informal neutrality band with a TOST equivalence test at margin delta = 0.015, split P4 into P4a/P4b, added a blocking retrieval-leakage guard, and specified a 100-item pilot. Amendment 2 (v1.3) corrects the operator brief's code-repository path to the existing DualMode codebase and quarantines a stale fork of the contract; the experimental design is unchanged. No corpus, adapter, pilot, or main-grid run existed at amendment time. Amendment 3 (v1.4) expands the corpus from 4 to 22 contributing GitHub repositories, because the original four yielded only 114 usable items against a requirement of n >= 1000; the expansion was outcome-blind, with no retriever, split, pilot, or grid in existence. Amendment 4 adds Rule 4 (temporal leakage: a retrieved case whose resolution postdates the target issue is a leak; missing timestamps fail closed) and Rule 5 (SWE-bench decontamination across all splits; 141 of 1213 candidates removed). Amendment 5 adds Rule 6 (locked Phase 1.5 verifiability probe: n=60, seed 42, pass iff the Wilson 95% lower bound times N_pool = 1597 is at least 1000), Rule 7 (repository-clustered bootstrap as a pre-registered sensitivity analysis) and Rule 8 (the leakage gate cannot pass on an empty denominator). Status at freeze: 1597 candidates, probe not yet run, no GPU spend, no teacher-model call, leakage gate VACUOUS_PASS_NO_CALLS.

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-07
DOI
https://doi.org/10.5281/zenodo.24034495
Primary Topic
Software Engineering Research
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

FixPipe pre-registration v1.8 - module-level criterion, second pipeline (locked probe and freeze chain)

1 links
Zenodo (CERN European Organization for Nuclear Research)
Software Engineering Research
preprint

FixPipe pre-registration v1.8 - module-level criterion, second pipeline (locked probe and freeze chain)

preprint en

Abstract

Frozen pre-registration for a cross-pipeline test of a module-retention criterion for LLM pipelines. Criterion: a module pays for itself only when it (i) owns an exclusive, verifiable information source and (ii) can convert it into an action on the artifact rather than into prose. FixPipe is a generate-check-repair pipeline over GitHub Issues with no vote signal. Amendment 1 (v1.1) replaced the informal neutrality band with a TOST equivalence test at margin delta = 0.015, split P4 into P4a/P4b, added a blocking retrieval-leakage guard, and specified a 100-item pilot. Amendment 2 (v1.3) corrects the operator brief's code-repository path to the existing DualMode codebase and quarantines a stale fork of the contract; the experimental design is unchanged. No corpus, adapter, pilot, or main-grid run existed at amendment time. Amendment 3 (v1.4) expands the corpus from 4 to 22 contributing GitHub repositories, because the original four yielded only 114 usable items against a requirement of n >= 1000; the expansion was outcome-blind, with no retriever, split, pilot, or grid in existence. Amendment 4 adds Rule 4 (temporal leakage: a retrieved case whose resolution postdates the target issue is a leak; missing timestamps fail closed) and Rule 5 (SWE-bench decontamination across all splits; 141 of 1213 candidates removed). Amendment 5 adds Rule 6 (locked Phase 1.5 verifiability probe: n=60, seed 42, pass iff the Wilson 95% lower bound times N_pool = 1597 is at least 1000), Rule 7 (repository-clustered bootstrap as a pre-registered sensitivity analysis) and Rule 8 (the leakage gate cannot pass on an empty denominator). Status at freeze: 1597 candidates, probe not yet run, no GPU spend, no teacher-model call, leakage gate VACUOUS_PASS_NO_CALLS.

Zenodo (CERN European Organization for Nuclear Research)
Software Engineering Research
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