[entelecheia]

Closes #449.

Section, saved-page and typed-range split outputs carried whole-source revision history. Remove hwp5_doc_history from each owned fragment clone before writing or range resource scanning, and return typed preservation evidence from SplitOutcome/RangeOutcome. The CLI merges it into the existing loss ledger before strict evaluation; MCP receives that same summary. Writers and source documents are unchanged.

Side-storage field	Decision	Basis and retained information
hwp5_doc_history	Drop by default	Whole-document revision history is outside the current fragment. Treat payloads as opaque, reporting one populated history storage plus its modeled streams per output, consistent with existing IR-level cross-format accounting.
hwp5_xml_template	Keep unchanged	Retain opaque native-form schema/binding data. Pruning it lacks a modeled dependency rule and could alter form compatibility; global information in that slot remains present.
hwpx_extra_entries	Keep unchanged	Retained opaque content can depend on master pages, scripts, extensions and original META-INF parts. Keep corresponding hwpx_opf_extra_items paired; range binary-reference scanning of these carried parts remains active.

The existing header/metadata and section/page-preview policies remain unchanged. Outputs continue to retain those global dependencies and information. This is a bounded history-exclusion policy.

History omissions use existing HwpContainerStorageRemoved/ContainerStorage and HwpContainerStreamRemoved/ContainerStream events with Removed disposition. Loss evidence is available through --loss-report, the unconditional stderr summary and MCP top-level preservation. --strict now refuses history-bearing splits before fragment publication, after saving the loss report. The frozen, binary/address-focused range-extraction-v1 JSON shape stays unchanged.

Test-only baseline 43340bc reproduced all four intended assertions: history remained in section/page/range containers and strict split incorrectly succeeded. Three compatibility controls passed. After the fix, all 7 CLI/container tests and 75 converter split/range unit tests passed, with hwp-cli/hwp-convert all-target clippy green.

Tests use owner-authored opaque sentinels, checking actual CFB storage and payload-byte absence, typed ledger counts, strict no-publication, retained schema bytes and paired HWPX parts/manifest entries. Ordinary convert/edit preserve history, and native no-op rewriting preserves the whole input byte-for-byte. Saved-page selection uses explicit fixture flags without host-font pagination assertions. These tests do not parse genuine revision-history records or establish a Hancom open/export result.

Full verification at 33114d1dead6f7c0de9b4b4ef9e3b669e74b6662: exact-head remote Linux/Rust 1.93.0, CARGO_BUILD_JOBS=2 RUST_TEST_THREADS=2 scripts/check.sh </dev/null, with verifier stdin also closed, exit 0. All 7 new CLI/container cases ran in this pass.

== check: OK (fmt/clippy/test/crate-edges/pdf-runner/structured-corpus/claims/doc-surface/release-block/readiness-selfcheck/skip-accounting/public-parity=skipped) skipped-for-missing-fixtures=49 (optional=4) ==

Missing-fixture skips are 49, including 4 optional cases; absent fixture paths remain unverified. Public PDF parity auto-skipped for the environment's Poppler version. Hosted three-OS/parity checks and native Hancom evidence remain separate.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-10T15:22:44.269295Z	33114d1	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.