[docxology]

Closes #88.

Derive the fep-050 heat/work bounds from normalized finite-path erasure laws, reverse support, local detailed balance and a reset marginal. The biased-prior bound gives the fair-bit corner; six constructive controls include a reversible equality case. The constitutive-law bridge keeps its explicit entropy-balance assumption and conditional meaning. The maturity disposition remains unchanged.

Validate the five current successor statements and exact current metadata before copied historical comparisons. Existing sealed review fixtures remain unchanged. Retain independent science, software, test-partition and native/publication observations.

Move 182 source-only cases into Python feedback. This marker-only transition preserves its 4,469 collected identities and every test function body; all 37 compiler functions in the audited files remain explicitly serial. Frozen/uncertain owners stay excluded. Use the active interpreter for the artifact-proof freshness check, and document packaging preparation before strict test collection.

Validation:

• Ruff, formatting, strict mypy, source inventory, links, Markdown hygiene, cross-references and whitespace passed.
• Strict family/core build passed; 102 focused checks passed, including the maintained finite-path controls and exact axiom acceptance.
• A fresh once-only unfiltered native API producer compiled all 168 topics without warnings or sorry and independently validated its source-bound receipt.
• Exact split Python acceptance passed: 3,992 accounted cases, 3,982 passes, ten permitted skips, no failures/errors; both live Chrome controls passed. All measured package coverage floors passed; aggregate line coverage was 92.44%.
• Fresh local PDF/HTML rendering passed with the pinned template, 390 PDF pages, zero TeX errors and zero missing glyphs. Its dirty-source checkpoint is explicitly disclosed.

The first local split run passed all executed tests but correctly failed its final integrity join when wheel building first created six packaging metadata files. That failed record is retained. Wheel metadata was prepared before a new collection and full execution; no source-drift waiver or relaxed acceptance was used.

Normal hosted PR and merged-main CI are still required. No release or tag is proposed. H3 export/refactor and broader Gaussian issue acceptance remain in their own scopes.

Integration update: normally merged paired main 0a3feec63b4561202d168548a14f66de70af2b8d into this branch. The only conflict was the changelog; both entries are retained. All generated-owner freshness checks passed. Earlier local native/PDF receipts and original-head hosted artifacts remain scoped to their original checkpoints. Fresh full combined-head hosted qualification is required before main integration.

Combined-head qualification completed at 4d7d1d6e00975a5c3859a814f9ab4172db3dd054: all 22 hosted contexts accepted (21 success, expected documentation skip). Independent retained-artifact audit verified all five ZIP digests, 4,001 accounted Python cases (3,982 passes and 19 permitted skips), every package coverage floor, 168 clean native topic results, the 2,060-declaration audit, and current source-bound 392-page PDF payload. This is PR-head acceptance; actual merged-main qualification remains separate.