bindery-cli 0.46.0__tar.gz → 0.47.0__tar.gz

{bindery_cli-0.46.0 → bindery_cli-0.47.0}/.github/workflows/ci.yml

@@ -44,9 +44,10 @@ jobs:
         # The install floor (decision 63): vir-tui/cquarry are marker-gated
         # to 3.14 there, so these legs install the core (tqdm only) and run
         # exactly the test modules that never touch the gated stack: the
-        # repair core, the plugin under calibre stubs, validate, and the
-        # version/grammar pins. audit/library/CLI-stack tests stay on the
-        # 3.14 leg above.
+        # repair core, the plugin under calibre stubs, validate, the
+        # version/grammar pins, the RSC-016 comment-hyphen fix, the
+        # RepairFlags wiring pins, and the flags-driven real-core run.
+        # audit/library/CLI-stack tests stay on the 3.14 leg above.
         python-version: ["3.12", "3.13"]
     steps:
       - uses: actions/checkout@v5
@@ -59,6 +60,8 @@ jobs:
           tests.test_transforms tests.test_epub tests.test_pagination
           tests.test_watermark tests.test_reserialize tests.test_version
           tests.test_plugin tests.test_validate
+          tests.test_comment_hyphens tests.test_flags_wiring
+          tests.test_flags_real_core
   plugin-compat:
     name: Parse-check the plugin zip under the interpreters Calibre embeds
     runs-on: ubuntu-latest

{bindery_cli-0.46.0 → bindery_cli-0.47.0}/CLAUDE.md

@@ -30,15 +30,27 @@ Born from the 2026 library audit (see the user memory `calibre-library-epubcheck
   `xmlns`, bare `&`, named entities, void self-closing), the NCX pipeline, and the
   mimetype fix (added/normalized/first-stored, epub.py's archive rewrite); every core
   fix must render identically to the author's intent: never add, remove, or reorder
-  visible content. The canonical opt-in inventory lives in spec.md: sixteen structural
-  repairs, three lossy strips, four safe opt-ins, plus `--reserialize`.
+  visible content. The canonical opt-in inventory lives in spec.md: twenty-one structural
+  repairs, four lossy strips, four safe opt-ins, plus `--reserialize`.
   * **Structural repairs** (`--fix-empty-body`, `--fix-missing-title`, `--fix-id-colons`,
     `--fix-page-map`, `--strip-epub3-attrs`, `--downgrade-epub3-tags`,
     `--unwrap-block-in-inline`, `--strip-invalid-value`, `--unwrap-illegal-tags`,
     `--prune-missing-resources`, `--strip-broken-anchors`, `--encode-url-spaces`,
     `--fix-container`, `--fix-media-types`, `--fix-cover`,
-    `--fix-comment-double-hyphen`; transforms.py, threaded
-    through epub.py). `--fix-comment-double-hyphen` (v0.46.0, the RSC-016 fatal)
+    `--fix-comment-double-hyphen`, `--fix-svg-dup-ids`, `--fix-cdata-terminator`,
+    `--fix-misnested-inline`, `--fix-stray-close`, `--fix-unterminated-attr`;
+    transforms.py, threaded through epub.py). The v0.47.0 five (the 2026-10-06
+    issue batch): `--fix-svg-dup-ids` renames duplicate glyph ids inside
+    standalone `.svg` entries (first keeps its name, references resolve to the
+    first definition so they are never rewritten); `--fix-cdata-terminator`
+    completes a truncated `/*]]>` CDATA-terminator comment inside inline
+    `<style>` blocks (CSS-008); `--fix-misnested-inline` normalizes the
+    Mobipocket drop-cap mis-nest `<i><b>X</i></b>` (punctuation between the
+    reversed closers survives outside both spans; a letter run is refused);
+    `--fix-stray-close` removes an end tag with no open start frame (a merely
+    mis-nested pair is never touched); `--fix-unterminated-attr` closes an
+    attribute value open to the tag's own `>` (bare-word values, no quote
+    before the next `<`); all normal `gate`. `--fix-comment-double-hyphen` (v0.46.0, the RSC-016 fatal)
     replaces `--` inside XML comments with an en-dash: comment bodies only (text
     nodes and CDATA never touched, the one deliberate exception to the
     never-rewrite-comments invariant), in content documents, the NCX, and the OPF
@@ -71,7 +83,10 @@ Born from the 2026 library audit (see the user memory `calibre-library-epubcheck
     in `RepairReport.watermark_refusals`, and surfaced as a `manual_watermark_repair`
     decision by `run phase1` on both the read-only and apply paths. `--strip-stub-docs`
     (2026-09-15) mirrors emptytext's placeholder signals: identical short text across
-    >= 3 spine docs and >= 30% of the spine; the whole-spine-stubs book is refused
+    >= 3 spine docs and >= 30% of the spine; since v0.47.0 the pool is the `<body>`
+    span and image-carrier pages (empty body text, an `<img>` present) are exempt, so
+    identical head `<title>`s can never weld illustrated divider pages into a fake
+    stub class (the Great Change misfire); the whole-spine-stubs book is refused
     (EMPTY, re-source), and the drop cascades to archive entries, manifest, spine,
     NCX navPoints, and nav toc entries.
   Do not let any NEW fix touch content without its own flag; if a candidate repair cannot
@@ -80,11 +95,13 @@ Born from the 2026 library audit (see the user memory `calibre-library-epubcheck
 - **The gate is the safety contract.** Never apply a repair epubcheck has not accepted.
   Respect the two-mode logic in `validate.gate` (fatal-fixing tolerates error unmasking;
   error-cleanup does not). The lossy strips (`--strip-pagination`, `--strip-broken-tags`,
-  `--strip-watermarks`) are accepted by `validate.no_worse` instead (their gain is
-  invisible to epubcheck, so it only forbids a regression, never demands a measured
-  improvement). The structural opt-ins go through the normal `gate`: their gain IS
-  visible (they clear errors), so a run with no measurable improvement is a noop and
-  nothing is applied. Changing either bar means re-running the library dry run.
+  `--strip-watermarks`, `--strip-stub-docs`) are accepted by `validate.no_worse` instead
+  (their gain is invisible to epubcheck, so it only forbids a regression, never demands
+  a measured improvement). The structural opt-ins go through the normal `gate` (their
+  gain IS visible, so a run with no measurable improvement is a noop and nothing is
+  applied) except the two whose gains sit on axes epubcheck does not measure:
+  `--fix-cover` and `--encode-url-spaces`, accepted under the same `no_worse` bar with
+  the partial rule intact. Changing either bar means re-running the library dry run.
 - **Library writes are sacred.** Replacement must stay atomic (temp in same dir, then
   `os.replace`), touch only the `.epub`, preserve mode, and be dry-run by default.
   Calibre format installation goes through cquarry's write module (`--install-to-calibre`;
@@ -93,7 +110,7 @@ Born from the 2026 library audit (see the user memory `calibre-library-epubcheck
 
 ## Layout
 
-- `src/bindery/transforms.py`: pure `str -> (str, int)` text transforms (including `strip_broken_tags`). Since v0.32.0: `strip_attrs_in_start_tags` anchors attribute edits on the quote-aware start-tag matcher, `_outside_protected` forwards arguments (context-carrying transforms can be decorated), `fix_id_colons` matches only the bare `id` attribute and only internal fragments (`fix_ncx_src_fragments` carries renames into the NCX), `fix_ncx_playorder` is anchored to `<navPoint>` start tags, and the css selector boundary covers namespaced and functional selector forms. HAZARD (the 0.41.0 spin): the quote-aware tag-prefix alternation (double-quoted | single-quoted | [^>]) OVERLAPS ([^>] matches quotes too), so a plain `*` re-partitions exponentially on failing candidates; every such greedy loop must carry the possessive `*+` (re 3.11+, the plugin's minimum Calibre). Since v0.42.0 the census is complete and pinned by tests/test_matcher_hardening.py: every greedy loop of this family is possessive (the three 0.41.1 rewrites plus _START_TAG_RE, _COVER_META_RE, _GUIDE_REF_RE, prune_dangling_edges' spine|item|itemref matcher, _ITEM_TAG_RE, _SPINE_ITEM_RE -- possessive is byte-identical on success because the loop is always followed by a literal `>` the [^>] branch cannot consume), and the five LAZY siblings (_VOID_RE, the transforms <img> matcher, and epub's link/a/img matchers) use the disjoint catch-all `[^>"']` instead: mutually exclusive branches cannot re-partition, and an unterminated quote inside a tag now fails to match rather than pairing across markup. NEW tag matchers: quote-aware possessive for greedy loops, the disjoint catch-all for lazy ones, and the tempered-quote idiom (["'])((?:(?!\1).)*)\1 for attribute values. Since v0.37.0: `fix_ncx_playorder` strips quotes before its already-correct comparison (regex groups carry quote characters; comparing quoted values against bare numbers counted phantom fixes on every sequential NCX and rewrote single-quoted correct attributes to double quotes).
+- `src/bindery/transforms.py`: pure `str -> (str, int)` text transforms (including `strip_broken_tags`). Since v0.32.0: `strip_attrs_in_start_tags` anchors attribute edits on the quote-aware start-tag matcher, `_outside_protected` forwards arguments (context-carrying transforms can be decorated), `fix_id_colons` matches only the bare `id` attribute and only internal fragments (`fix_ncx_src_fragments` carries renames into the NCX), `fix_ncx_playorder` is anchored to `<navPoint>` start tags, and the css selector boundary covers namespaced and functional selector forms. HAZARD (the 0.41.0 spin): the quote-aware tag-prefix alternation (double-quoted | single-quoted | [^>]) OVERLAPS ([^>] matches quotes too), so a plain `*` re-partitions exponentially on failing candidates; every such greedy loop must carry the possessive `*+` (re 3.11+, the plugin's minimum Calibre). Since v0.42.0 the census is complete and pinned by tests/test_matcher_hardening.py: every greedy loop of this family is possessive (the three 0.41.1 rewrites plus _START_TAG_RE, _COVER_META_RE, _GUIDE_REF_RE, prune_dangling_edges' spine|item|itemref matcher, _ITEM_TAG_RE, _SPINE_ITEM_RE; possessive is byte-identical on success because the loop is always followed by a literal `>` the [^>] branch cannot consume), and the five LAZY siblings (_VOID_RE, the transforms <img> matcher, and epub's link/a/img matchers) use the disjoint catch-all `[^>"']` instead: mutually exclusive branches cannot re-partition, and an unterminated quote inside a tag now fails to match rather than pairing across markup. NEW tag matchers: quote-aware possessive for greedy loops, the disjoint catch-all for lazy ones, and the tempered-quote idiom (["'])((?:(?!\1).)*)\1 for attribute values. Since v0.37.0: `fix_ncx_playorder` strips quotes before its already-correct comparison (regex groups carry quote characters; comparing quoted values against bare numbers counted phantom fixes on every sequential NCX and rewrote single-quoted correct attributes to double quotes).
 - `src/bindery/pagination.py`: the opt-in lossy page-number strip (runhead detection, page-layer decision, block-centric removal/merge, safety nets).
 - `src/bindery/watermark.py`: the opt-in lossy watermark strip (anchored and anchorless signature removal).
 - `src/bindery/reserialize.py`: structural repair via `html5lib`.
@@ -102,8 +119,8 @@ Born from the 2026 library audit (see the user memory `calibre-library-epubcheck
 - `src/bindery/validate.py`: epubcheck wrapper, the `gate` (improvement) and `no_worse` (no-regression, for the lossy strips) acceptance bars.
 - `src/bindery/library.py`: Calibre walk, atomic replace, backups, and native format installation. Since v0.19.0 `CalibreIdResolver` resolves the book id from `metadata.db` through cquarry (one lazy path→id map per run; since v0.20.0 the map comes from `CalibreDB.format_path_index()`, re-normalized with `resolve().lower()` for the resolver's symlinked-directory and case-insensitive matching); the `(id)` directory-name regex is only the no-catalog fallback. Since v0.24.0 `install_format()` places the repaired file atomically and updates the `data` row through `cquarry.write.WritableCalibreDB` (since v0.31.0 via `set_format`, cquarry 1.17's sanctioned remove+add in one transaction); the external `calibredb` CLI is gone. Since v0.32.0 a guessed id drives a row update only when metadata.db corroborates it (`_verify_guess`: books row exists, file in the book's own directory, catalogued `data.name` when a row exists); anything else saves in place and leaves the catalog untouched: never reintroduce directory-name guessing as the primary source: renamed/mismatched directories would replace the wrong book. Since 2026-09-15 `make_backup` takes the opt-in `--backup-keep N` ring: at most N backup files per book, the author original `.bak` never deleted, newest content at the highest name; default remains an unbounded rotation.
 - `src/bindery/cli.py`: `repair` and `library` subcommands, including `--all` and `--install-to-calibre`; `repair --json` emits the library per-book record vocabulary (2026-09-15; since v0.45.0 every per-book record, the phase1 payload included, also carries the structured fix breakdown: the `fixes` dict plus `ncx_uid_synced`/`watermark_refusals` off the `Outcome`, and `_phase1_decisions` reads that data instead of substring-matching the rendered `summary`, which stays for human eyes; CalibreQuarry's lossy-consent mirror is the waiting consumer, floor `bindery>=0.45.0`); since v0.29.0 also the `run` verbs: `phase1` (pre-import vetting over loose files: the audit battery composed with one gated repair sweep; read-only until `--apply-lossy`, which IS the lossy-strip consent) and `phase3` (the scoped post-import apply step: `library --id --sweep --only all --apply --all --install-to-calibre` plus a pre/post summary; mechanically refuses unscoped sweeps with exit 2; since 2026-09-15 the summary sums applied/equal states only and carries rejected projections as `rejected_projection`). Plus `bindery doctor` (2026-09-15): the environment self-check that imports none of the VirInvictus stack, never raises, and always exits 0. Both run verbs take `--non-interactive` and surface open questions as `decisions_needed` in their JSON, never prompts. They drive the shipped `run_library` through the real parser, so no flag can drift between the verb and the subcommand it wraps. Since v0.46.0 `cli._flags_from_args(args)` is the one place the argparse names map onto `RepairFlags` fields (each `or --all`), shared by `run_library` and `run_repair`; the old per-verb ~25-kwarg blocks are gone. `library --workers N` (v0.30.0) parallelizes only the sweep's candidate pass, in windows of N consumed in order so `--limit` stays lazy; the repair phase is never parallel (shared workdir, atomic-replacement contract). Since v0.40.0 `main()` guards the VirInvictus stack: a `ModuleNotFoundError` naming `vir_tui`/`cquarry` (marker-gated to 3.14+ on the 3.12 floor) becomes one stderr line and trouble exit 2 instead of a traceback; anything else missing still raises.
-- `plugin/__init__.py` + `scripts/build_plugin.py`: the Bindery Repair Calibre plugin (v0.37.0; identity `Bindery Repair` / `bindery_repair`, approved and built per the scoped spec in roadmap.md Phase 3). The builder vendors `transforms.py`, `epub.py`, `pagination.py`, `watermark.py`, `reserialize.py` byte-identically into the zip (the zip root is a package, so their relative imports resolve unchanged; the suite's drift test pins the equality) with the version tuple substituted from the single-source VERSION; `publish.yml` attaches `BinderyRepair-v<VERSION>.zip` to each release. The plugin runs ONLY the default pass (epubcheck cannot gate inside Calibre), never raises, is byte-idempotent (zero fixes returns the original path), testzip()-verifies its output, logs one line per book, and takes JSON config via `site_customization` (`log`, `log_path`, `max_size_mb` default 150, `epubcheck_path` experimental validation default OFF). Load it in tests through the `tests/test_plugin.py` stubs: the calibre namespace is installed in sys.modules and the package loads with a two-dir `__path__` (plugin dir first, then src/bindery) exactly like calibre's zipplugin loader shape. **Plugin/interpreter compatibility (v0.40.0):** v0.39.0 shipped PEP 758 unparenthesized except tuples, and the plugin died with a bare SyntaxError on every released Calibre (7.x and the whole 8.x series embed Python 3.11, 9.0+ embeds 3.14; from calibre's bypy sources.json, checked against the upstream clone). The core is now grammar-pinned: parenthesized except tuples everywhere, `ruff`'s `target-version = "py312"` pin (inference from requires-python would rewrite the parens back off), `tests/test_version.py`'s parse guard, and CI's `plugin-compat` job byte-compiling the built zip under 3.11/3.12/3.13/3.14. `minimum_calibre_version` is (7, 0, 0): the oldest series whose interpreter CI covers, replacing the hollow (2, 0, 0); `supported_platforms` stays `["linux"]` (developed and tested on Linux only). Any new syntax in the vendored modules must clear the guard before it ships.
-- `tests/`: transforms, end-to-end repair, atomic replace, pagination, watermarks, the audit analyzers (content/pagenumbers/emptytext/ocr/monolithic/completeness, archive integrity, spine classification), the library sweep and `CalibreIdResolver`, validate, reserialize, CLI wiring, the version pin and the syntax-floor grammar guard, and the Calibre plugin (build drift + run behavior under calibre stubs).
+- `plugin/__init__.py` + `scripts/build_plugin.py`: the Bindery Repair Calibre plugin (v0.37.0; identity `Bindery Repair` / `bindery_repair`, approved and built per the scoped spec in roadmap.md Phase 3). The builder vendors `transforms.py`, `epub.py`, `pagination.py`, `watermark.py`, `reserialize.py` byte-identically into the zip (the zip root is a package, so their relative imports resolve unchanged; the suite's drift test pins the equality) with the version tuple substituted from the single-source VERSION; `publish.yml` attaches `BinderyRepair-v<VERSION>.zip` to each release. The plugin runs ONLY the default pass (epubcheck cannot gate inside Calibre), never raises, is byte-idempotent (zero fixes returns the original path), testzip()-verifies its output, logs one line per book, and takes JSON config via `site_customization` (`log`, `log_path`, `max_size_mb` default 150, `max_log_mb` rotation cap default 2, `epubcheck_path` experimental validation default OFF). Load it in tests through the `tests/test_plugin.py` stubs: the calibre namespace is installed in sys.modules and the package loads with a two-dir `__path__` (plugin dir first, then src/bindery) exactly like calibre's zipplugin loader shape. **Plugin/interpreter compatibility (v0.40.0):** v0.39.0 shipped PEP 758 unparenthesized except tuples, and the plugin died with a bare SyntaxError on every released Calibre (7.x and the whole 8.x series embed Python 3.11, 9.0+ embeds 3.14; from calibre's bypy sources.json, checked against the upstream clone). The core is now grammar-pinned: parenthesized except tuples everywhere, `ruff`'s `target-version = "py312"` pin (inference from requires-python would rewrite the parens back off), `tests/test_version.py`'s parse guard, and CI's `plugin-compat` job byte-compiling the built zip under 3.11/3.12/3.13/3.14. `minimum_calibre_version` is (7, 0, 0): the oldest series whose interpreter CI covers, replacing the hollow (2, 0, 0); `supported_platforms` stays `["linux"]` (developed and tested on Linux only). Any new syntax in the vendored modules must clear the guard before it ships.
+- `tests/`: transforms, end-to-end repair, atomic replace, pagination, watermarks, the audit analyzers (content/pagenumbers/emptytext/ocr/monolithic/completeness, archive integrity, spine classification), the library sweep and `CalibreIdResolver`, validate, reserialize, CLI wiring, the version pin and the syntax-floor grammar guard, and the Calibre plugin (build drift + run behavior under calibre stubs). Since the v0.46.0 follow-ups: `tests/test_flags_wiring.py` reflectively pins the RepairFlags inventory to the argparse dests and `_flags_from_args` mapping (a flag added without a field, a field without a flag, or a remapped line fails there), the all-off defaults, and the version gate's caller-immutability contract (test_epub's `TestVersionGateKeepsFlagsIntact`); `tests/test_comment_hyphens.py` carries the RSC-016 fix; `tests/test_flags_real_core.py` drives every repair flag from argv through the unmocked core (--no-validate; the gate itself stays pinned by the mocked-oracle tests and runs for real wherever epubcheck exists). All three ride CI's core-compat 3.12/3.13 leg; the hand-listed module list there must gain an entry for every new stack-free test module.
 
 ## Conventions
 

{bindery_cli-0.46.0 → bindery_cli-0.47.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: bindery-cli
-Version: 0.46.0
+Version: 0.47.0
 Summary: Audit EPUB body text and repair broken EPUBs with deterministic, epubcheck-gated fixes; atomic in-place replacement in a Calibre library.
 Project-URL: Homepage, https://github.com/VirInvictus/bindery-cli
 Project-URL: Repository, https://github.com/VirInvictus/bindery-cli
@@ -51,7 +51,7 @@ bindery-cli makes accidentally broken markup well-formed again. It does not rewr
 
 Everything below is off until its flag is passed (or all at once via `--all`). They go further than well-formedness (altering structure or adding minimal content), so they never run by default:
 
-- **`--fix-ids`**: rewrite ids that are not valid XML names (start with a digit, contain a colon) in the OPF manifest, updating every reference to them (spine, fallback, media-overlay, the EPUB 2 cover meta), and in the NCX (where old conversions stamp navPoint ids from UUIDs). Touches the OPF, so it is off by default; the dc: metadata is never altered.
+- **`--fix-ids`**: rewrite ids that are not valid XML names (start with a digit, contain a colon, or carry any character outside the XML NameChar set, like the apostrophes calibre copies from filenames) in the OPF manifest, updating every reference to them (spine, fallback, media-overlay, the EPUB 2 cover meta), and in the NCX (where old conversions stamp navPoint ids from UUIDs). Touches the OPF, so it is off by default; the dc: metadata is never altered.
 - **`--add-img-alt`**: add `alt=""` to `<img>` elements missing the required attribute. Renders identically, but it adds markup the author never wrote, and an empty alt tells a screen reader the image is decorative; hence opt-in.
 - **`--reserialize`**: rebuild content documents that are still malformed by re-parsing them with html5lib and re-emitting XHTML, closing unclosed `<p>`/`<div>`/`<span>`/`<blockquote>` that the regex transforms cannot. Runs only on documents that are not already well-formed, so good files are untouched.
 - **`--strip-bad-attrs`**: drop attributes that are invalid XML (a name starting with a digit, or a namespaced name whose prefix is never declared, like Office VML `v:shapes`). Surgical and a no-op on well-formed files.
@@ -67,11 +67,16 @@ Everything below is off until its flag is passed (or all at once via `--all`). T
 - **`--unwrap-illegal-tags`**: strips completely invalid or deprecated HTML tags that break EPUB3 validation (`<st>`, `<sentence>`, `<o>`, `<w>`, `<pagebreak>`) while retaining their inner text. Any of those names styled as an *element selector* by an EPUB stylesheet (`.css` entries and inline `<style>` blocks alike; class/id selectors like `.st`/`#w` don't count) is protected for the whole book, guaranteeing format preservation.
 - **`--prune-missing-resources`**: removes references to files the archive does not contain (RSC-007/PKG-010): dead `<link>` elements, anchors' `href` to absent files (anchor text preserved), absent `<img>` sources (replaced by their alt text when they carry one), and orphaned non-spine OPF manifest items (with every package edge that pointed at them rewritten: `spine@toc`, `item@media-overlay`, `item@fallback`, and the EPUB2 cover meta). Spine documents are never pruned: a missing spine document is reported as a damaged fragment, not silently dropped.
 - **`--strip-broken-anchors`**: strips href attributes that cannot resolve, keeping the anchor text byte-for-byte: a `#fragment` the target document does not define (RSC-020/RSC-012; NCX navTargets fall back to the document target, so chapter navigation survives) and unresolvable URI schemes (`kindle:`, `file:`).
-- **`--encode-url-spaces`**: percent-encodes raw spaces in `src`/`href` attribute values across the package (OPF manifest, NCX, content documents). A URL with a literal space is not a valid URL (RSC-020) and unresolvable on strict readers; the encoded form denotes the same file.
+- **`--encode-url-spaces`**: percent-encodes raw spaces in `src`/`href` attribute values across the package (OPF manifest, NCX, content documents). A URL with a literal space is not a valid URL (RSC-020) and unresolvable on strict readers; the encoded form denotes the same file. Since v0.44.0 the flag also renames archive entries whose names carry raw spaces to their underscore spellings and rewrites every reference to them (OPF, NCX, content, CSS `url()`), clearing the PKG-010 space-in-filename advisories: underscore, never percent-encoding, because epubcheck decodes references before entry lookup.
 - **`--fix-container`**: generates `META-INF/container.xml` at the located OPF when the container is missing or names a file the archive does not contain. This is the gateway defect: epubcheck stays fatal while the OPF is unfindable, so no repair can be gate-accepted on such a book until the container exists.
 - **`--fix-media-types`**: normalizes wrong manifest media-type declarations (OPF-029, e.g. a jpg stamped `image/png` by an aggregator). Attribute-only and quote-preserving; it fires only when the file's magic bytes confirm the extension, so a misnamed-but-consistent file (a PNG renamed .jpg) is never made worse.
 - **`--fix-cover`**: repairs dangling EPUB2 cover wiring. A `<meta name="cover">` whose content names no manifest item is re-pointed when the OPF guide's own cover reference names an existing item, and removed when nothing identifies it; EPUB3 `properties="cover-image"` is deliberately audit-only (guessing which image is the cover is not deterministic). Cover wiring is invisible to epubcheck, so cover-only repairs are accepted under the same no-worse bar the lossy strips use.
 - **`--fix-comment-double-hyphen`**: replaces `--` sequences inside XML comments with en-dashes. `--` is illegal inside an XML comment (epubcheck fatal RSC-016: the book opens in lenient readers but never passes epubcheck). The edit stays inside the comment bodies: text nodes and CDATA sections are never touched, and each comment's terminator is left intact.
+- **`--fix-svg-dup-ids`**: renames duplicate `id` values inside standalone `.svg` entries (old calibre SVG page renders repeat glyph ids within one document; the intake wave that found this carried 445 to 1,339 'Duplicate "glNNNN"' errors per book). The first occurrence keeps its name, later ones gain `_2`/`_3`, and internal references are deliberately untouched: they already resolve to the first definition under every reader's first-match lookup.
+- **`--fix-cdata-terminator`**: completes a truncated CDATA-terminator comment in an inline `<style>` block (`/*]]>` becomes `/*]]>*/`). A comment-wrapped CDATA block whose closing `*/` the converter dropped leaves the CSS parser inside an unterminated comment: one CSS-008 'Premature end of file' per affected document. Only the malformed-terminator token class is rewritten; invalid CSS stays unfixed, and stylesheet files are not touched.
+- **`--fix-misnested-inline`**: rewrites the Mobipocket drop-cap mis-nest `<i><b>X</i></b>` to `<i><b>X</b></i>` (em/strong variants and either tag order). Only the both-open-then-reversed-close shape is touched; a punctuation run caught between the reversed closers (`<i><b>xile</i>,</b>`) survives outside both spans, and a letter run there is refused (which span keeps the letters has no deterministic answer).
+- **`--fix-stray-close`**: removes an end tag whose element has no open start tag anywhere above it (the stray extra `</div>` on a shell page, which cascades into 'body must be terminated by the matching end-tag'). Merely mis-nested pairs are never touched: a close is removed only when nothing by that name is open.
+- **`--fix-unterminated-attr`**: closes an attribute value left open to the tag's own `>` (`<p class="footnote>` becomes `<p class="footnote">`). The parser swallows prose into the value until the next quote and dies on the first `<` it meets. Guards keep it off legal `>`-bearing values: bare-word values only, and no quote may close before the next `<`.
 
 
 Four opt-in fixes are **lossy** and stand apart from the semantics-preserving rest:
@@ -79,7 +84,7 @@ Four opt-in fixes are **lossy** and stand apart from the semantics-preserving re
 - **`--strip-pagination`**: remove print page numbers and running headers that a PDF/OCR conversion baked into the body text as literal paragraphs (so they reflow into the middle of a sentence: "where the hay cart **16** was taking him"). It removes only that injected furniture, never the author's prose: where a number split a sentence it rejoins the two paragraphs (closing up a word split like `compli-` / `mentary`), and it preserves roman chapter numbers, page-list nav anchors, and years. A roman numeral counts as a page number only when it is well-formed and under 100, so ordinary words (`mid`, `mix`, `civil`) are never mistaken for pages. A book is only treated as paginated when it has both a dense run of arabic numbers and several confident mid-sentence interrupts, so a merely chapter-numbered book is left alone. Three safety nets guard every edit (character conservation, tag balance, and an epubcheck no-regression check); any failure leaves the document untouched.
 - **`--strip-broken-tags`**: remove leaked HTML closing tags missing their open brackets (e.g. `</p>`) that render as raw text.
 - **`--strip-watermarks`**: remove known producer and distributor watermarks (e.g. OceanofPDF, ABC Amber LIT Converter) and stray marker files. It locates the stamp and deletes the outermost wrapper whose entire visible text is the watermark, ensuring prose that merely mentions the URL is preserved. An inline stamp link is deleted only when it holds nothing but the stamp; a match too large to be safe (an unclosed stamp anchor that swallowed prose) is refused and reported for manual repair rather than deleted.
-- **`--strip-stub-docs`**: drop spine documents whose entire visible text is one identical short placeholder repeated across the spine: the Bookmate-style export whose chapters are all the same "content unavailable" notice, passing epubcheck because the placeholder is valid XHTML. The drop cascades (archive entries, manifest items, spine order, NCX navPoints, nav toc entries), so the book opens straight into its real chapters. The identity rule is deliberately conservative (same text, at least 3 docs, at least 30% of the spine, each doc short but not blank) and a book whose *every* spine doc is the stub is refused outright: that book is empty and needs a re-source, never a repair.
+- **`--strip-stub-docs`**: drop spine documents whose entire visible body text is one identical short placeholder repeated across the spine: the Bookmate-style export whose chapters are all the same "content unavailable" notice, passing epubcheck because the placeholder is valid XHTML. The drop cascades (archive entries, manifest items, spine order, NCX navPoints, nav toc entries), so the book opens straight into its real chapters. The identity rule is deliberately conservative (same body text, at least 3 docs, at least 30% of the spine, each doc short but not blank; the pool is the `<body>` span, and image-carrier pages with no body text are exempt, so illustrated part-divider pages are never candidates) and a book whose *every* spine doc is the stub is refused outright: that book is empty and needs a re-source, never a repair.
 
 All lossy edits are invisible to epubcheck, so they are accepted when the result is *no worse* rather than measurably better.
 
@@ -94,7 +99,7 @@ Introducing a net-new fatal is always rejected. If epubcheck itself fails to run
 
 Under the hood, the oracle can run two ways: the usual `epubcheck --json -` subprocess, or a persistent daemon that writes `FastDaemon.java` into a temp directory at runtime, compiles it in memory with the running JVM's own compiler (Java's single-file source launcher: no javac, no javac/java version skew), and drives the warm JVM over a pipe. The daemon reports the counts it reads from the JSON document that epubcheck's own `CheckingReport` serializes, so it measures exactly what the subprocess measures, just ~15x faster. It is a pool bounded by `--workers` (serial runs use one), every roundtrip is bounded by the caller's timeout, and any failure tears it down and falls back to the subprocess for good.
 
-The lossy modes (`--strip-pagination`, `--strip-broken-tags`, and `--strip-watermarks`) are the exception to the "must improve" rule. Since they remove visible markup rather than correcting XML schema violations, epubcheck counts often remain unchanged. They are accepted when the result is **no worse** (no net-new fatals or errors), relying on strict programmatic safety nets instead.
+The lossy modes (`--strip-pagination`, `--strip-broken-tags`, `--strip-watermarks`, and `--strip-stub-docs`) are the exception to the "must improve" rule. Since they remove converter-injected content rather than correcting XML schema violations, epubcheck counts often remain unchanged. They are accepted when the result is **no worse** (no net-new fatals or errors), relying on strict programmatic safety nets instead.
 
 ## Install
 

{bindery_cli-0.46.0 → bindery_cli-0.47.0}/README.md

@@ -24,7 +24,7 @@ bindery-cli makes accidentally broken markup well-formed again. It does not rewr
 
 Everything below is off until its flag is passed (or all at once via `--all`). They go further than well-formedness (altering structure or adding minimal content), so they never run by default:
 
-- **`--fix-ids`**: rewrite ids that are not valid XML names (start with a digit, contain a colon) in the OPF manifest, updating every reference to them (spine, fallback, media-overlay, the EPUB 2 cover meta), and in the NCX (where old conversions stamp navPoint ids from UUIDs). Touches the OPF, so it is off by default; the dc: metadata is never altered.
+- **`--fix-ids`**: rewrite ids that are not valid XML names (start with a digit, contain a colon, or carry any character outside the XML NameChar set, like the apostrophes calibre copies from filenames) in the OPF manifest, updating every reference to them (spine, fallback, media-overlay, the EPUB 2 cover meta), and in the NCX (where old conversions stamp navPoint ids from UUIDs). Touches the OPF, so it is off by default; the dc: metadata is never altered.
 - **`--add-img-alt`**: add `alt=""` to `<img>` elements missing the required attribute. Renders identically, but it adds markup the author never wrote, and an empty alt tells a screen reader the image is decorative; hence opt-in.
 - **`--reserialize`**: rebuild content documents that are still malformed by re-parsing them with html5lib and re-emitting XHTML, closing unclosed `<p>`/`<div>`/`<span>`/`<blockquote>` that the regex transforms cannot. Runs only on documents that are not already well-formed, so good files are untouched.
 - **`--strip-bad-attrs`**: drop attributes that are invalid XML (a name starting with a digit, or a namespaced name whose prefix is never declared, like Office VML `v:shapes`). Surgical and a no-op on well-formed files.
@@ -40,11 +40,16 @@ Everything below is off until its flag is passed (or all at once via `--all`). T
 - **`--unwrap-illegal-tags`**: strips completely invalid or deprecated HTML tags that break EPUB3 validation (`<st>`, `<sentence>`, `<o>`, `<w>`, `<pagebreak>`) while retaining their inner text. Any of those names styled as an *element selector* by an EPUB stylesheet (`.css` entries and inline `<style>` blocks alike; class/id selectors like `.st`/`#w` don't count) is protected for the whole book, guaranteeing format preservation.
 - **`--prune-missing-resources`**: removes references to files the archive does not contain (RSC-007/PKG-010): dead `<link>` elements, anchors' `href` to absent files (anchor text preserved), absent `<img>` sources (replaced by their alt text when they carry one), and orphaned non-spine OPF manifest items (with every package edge that pointed at them rewritten: `spine@toc`, `item@media-overlay`, `item@fallback`, and the EPUB2 cover meta). Spine documents are never pruned: a missing spine document is reported as a damaged fragment, not silently dropped.
 - **`--strip-broken-anchors`**: strips href attributes that cannot resolve, keeping the anchor text byte-for-byte: a `#fragment` the target document does not define (RSC-020/RSC-012; NCX navTargets fall back to the document target, so chapter navigation survives) and unresolvable URI schemes (`kindle:`, `file:`).
-- **`--encode-url-spaces`**: percent-encodes raw spaces in `src`/`href` attribute values across the package (OPF manifest, NCX, content documents). A URL with a literal space is not a valid URL (RSC-020) and unresolvable on strict readers; the encoded form denotes the same file.
+- **`--encode-url-spaces`**: percent-encodes raw spaces in `src`/`href` attribute values across the package (OPF manifest, NCX, content documents). A URL with a literal space is not a valid URL (RSC-020) and unresolvable on strict readers; the encoded form denotes the same file. Since v0.44.0 the flag also renames archive entries whose names carry raw spaces to their underscore spellings and rewrites every reference to them (OPF, NCX, content, CSS `url()`), clearing the PKG-010 space-in-filename advisories: underscore, never percent-encoding, because epubcheck decodes references before entry lookup.
 - **`--fix-container`**: generates `META-INF/container.xml` at the located OPF when the container is missing or names a file the archive does not contain. This is the gateway defect: epubcheck stays fatal while the OPF is unfindable, so no repair can be gate-accepted on such a book until the container exists.
 - **`--fix-media-types`**: normalizes wrong manifest media-type declarations (OPF-029, e.g. a jpg stamped `image/png` by an aggregator). Attribute-only and quote-preserving; it fires only when the file's magic bytes confirm the extension, so a misnamed-but-consistent file (a PNG renamed .jpg) is never made worse.
 - **`--fix-cover`**: repairs dangling EPUB2 cover wiring. A `<meta name="cover">` whose content names no manifest item is re-pointed when the OPF guide's own cover reference names an existing item, and removed when nothing identifies it; EPUB3 `properties="cover-image"` is deliberately audit-only (guessing which image is the cover is not deterministic). Cover wiring is invisible to epubcheck, so cover-only repairs are accepted under the same no-worse bar the lossy strips use.
 - **`--fix-comment-double-hyphen`**: replaces `--` sequences inside XML comments with en-dashes. `--` is illegal inside an XML comment (epubcheck fatal RSC-016: the book opens in lenient readers but never passes epubcheck). The edit stays inside the comment bodies: text nodes and CDATA sections are never touched, and each comment's terminator is left intact.
+- **`--fix-svg-dup-ids`**: renames duplicate `id` values inside standalone `.svg` entries (old calibre SVG page renders repeat glyph ids within one document; the intake wave that found this carried 445 to 1,339 'Duplicate "glNNNN"' errors per book). The first occurrence keeps its name, later ones gain `_2`/`_3`, and internal references are deliberately untouched: they already resolve to the first definition under every reader's first-match lookup.
+- **`--fix-cdata-terminator`**: completes a truncated CDATA-terminator comment in an inline `<style>` block (`/*]]>` becomes `/*]]>*/`). A comment-wrapped CDATA block whose closing `*/` the converter dropped leaves the CSS parser inside an unterminated comment: one CSS-008 'Premature end of file' per affected document. Only the malformed-terminator token class is rewritten; invalid CSS stays unfixed, and stylesheet files are not touched.
+- **`--fix-misnested-inline`**: rewrites the Mobipocket drop-cap mis-nest `<i><b>X</i></b>` to `<i><b>X</b></i>` (em/strong variants and either tag order). Only the both-open-then-reversed-close shape is touched; a punctuation run caught between the reversed closers (`<i><b>xile</i>,</b>`) survives outside both spans, and a letter run there is refused (which span keeps the letters has no deterministic answer).
+- **`--fix-stray-close`**: removes an end tag whose element has no open start tag anywhere above it (the stray extra `</div>` on a shell page, which cascades into 'body must be terminated by the matching end-tag'). Merely mis-nested pairs are never touched: a close is removed only when nothing by that name is open.
+- **`--fix-unterminated-attr`**: closes an attribute value left open to the tag's own `>` (`<p class="footnote>` becomes `<p class="footnote">`). The parser swallows prose into the value until the next quote and dies on the first `<` it meets. Guards keep it off legal `>`-bearing values: bare-word values only, and no quote may close before the next `<`.
 
 
 Four opt-in fixes are **lossy** and stand apart from the semantics-preserving rest:
@@ -52,7 +57,7 @@ Four opt-in fixes are **lossy** and stand apart from the semantics-preserving re
 - **`--strip-pagination`**: remove print page numbers and running headers that a PDF/OCR conversion baked into the body text as literal paragraphs (so they reflow into the middle of a sentence: "where the hay cart **16** was taking him"). It removes only that injected furniture, never the author's prose: where a number split a sentence it rejoins the two paragraphs (closing up a word split like `compli-` / `mentary`), and it preserves roman chapter numbers, page-list nav anchors, and years. A roman numeral counts as a page number only when it is well-formed and under 100, so ordinary words (`mid`, `mix`, `civil`) are never mistaken for pages. A book is only treated as paginated when it has both a dense run of arabic numbers and several confident mid-sentence interrupts, so a merely chapter-numbered book is left alone. Three safety nets guard every edit (character conservation, tag balance, and an epubcheck no-regression check); any failure leaves the document untouched.
 - **`--strip-broken-tags`**: remove leaked HTML closing tags missing their open brackets (e.g. `</p>`) that render as raw text.
 - **`--strip-watermarks`**: remove known producer and distributor watermarks (e.g. OceanofPDF, ABC Amber LIT Converter) and stray marker files. It locates the stamp and deletes the outermost wrapper whose entire visible text is the watermark, ensuring prose that merely mentions the URL is preserved. An inline stamp link is deleted only when it holds nothing but the stamp; a match too large to be safe (an unclosed stamp anchor that swallowed prose) is refused and reported for manual repair rather than deleted.
-- **`--strip-stub-docs`**: drop spine documents whose entire visible text is one identical short placeholder repeated across the spine: the Bookmate-style export whose chapters are all the same "content unavailable" notice, passing epubcheck because the placeholder is valid XHTML. The drop cascades (archive entries, manifest items, spine order, NCX navPoints, nav toc entries), so the book opens straight into its real chapters. The identity rule is deliberately conservative (same text, at least 3 docs, at least 30% of the spine, each doc short but not blank) and a book whose *every* spine doc is the stub is refused outright: that book is empty and needs a re-source, never a repair.
+- **`--strip-stub-docs`**: drop spine documents whose entire visible body text is one identical short placeholder repeated across the spine: the Bookmate-style export whose chapters are all the same "content unavailable" notice, passing epubcheck because the placeholder is valid XHTML. The drop cascades (archive entries, manifest items, spine order, NCX navPoints, nav toc entries), so the book opens straight into its real chapters. The identity rule is deliberately conservative (same body text, at least 3 docs, at least 30% of the spine, each doc short but not blank; the pool is the `<body>` span, and image-carrier pages with no body text are exempt, so illustrated part-divider pages are never candidates) and a book whose *every* spine doc is the stub is refused outright: that book is empty and needs a re-source, never a repair.
 
 All lossy edits are invisible to epubcheck, so they are accepted when the result is *no worse* rather than measurably better.
 
@@ -67,7 +72,7 @@ Introducing a net-new fatal is always rejected. If epubcheck itself fails to run
 
 Under the hood, the oracle can run two ways: the usual `epubcheck --json -` subprocess, or a persistent daemon that writes `FastDaemon.java` into a temp directory at runtime, compiles it in memory with the running JVM's own compiler (Java's single-file source launcher: no javac, no javac/java version skew), and drives the warm JVM over a pipe. The daemon reports the counts it reads from the JSON document that epubcheck's own `CheckingReport` serializes, so it measures exactly what the subprocess measures, just ~15x faster. It is a pool bounded by `--workers` (serial runs use one), every roundtrip is bounded by the caller's timeout, and any failure tears it down and falls back to the subprocess for good.
 
-The lossy modes (`--strip-pagination`, `--strip-broken-tags`, and `--strip-watermarks`) are the exception to the "must improve" rule. Since they remove visible markup rather than correcting XML schema violations, epubcheck counts often remain unchanged. They are accepted when the result is **no worse** (no net-new fatals or errors), relying on strict programmatic safety nets instead.
+The lossy modes (`--strip-pagination`, `--strip-broken-tags`, `--strip-watermarks`, and `--strip-stub-docs`) are the exception to the "must improve" rule. Since they remove converter-injected content rather than correcting XML schema violations, epubcheck counts often remain unchanged. They are accepted when the result is **no worse** (no net-new fatals or errors), relying on strict programmatic safety nets instead.
 
 ## Install
 

{bindery_cli-0.46.0 → bindery_cli-0.47.0}/patchnotes.md

@@ -1,4 +1,99 @@
 # bindery-cli Patch Notes
+
+## v0.47.0 (2026-10-06)
+
+### The intake-wave batch: five repair surfaces from five open issues
+
+The 2026-10-05/06 acquisition wave filed five issues against real books; all
+five land here. Two are classifier and validity corrections to existing
+surfaces, three are new opt-in structural repairs, and each was verified
+epubcheck-gated against its own fixture before shipping.
+
+* The stub classifier pools BODY text and exempts image carriers (issue #1).
+  The four Kobo part-divider pages of the Great Change carry only a
+  full-bleed `<img>` each, and their identical head `<title>` welded them,
+  with the title page, into a fake five-doc stub class: emptytext answered
+  PARTIAL and `--strip-stub-docs` offered a strip that deleted real
+  illustrations (the apply-time gate caught the deletion, but the offer was
+  wrong). Both the analyzer and the repair twin now pool the `<body>` span,
+  and a doc with no body text that carries an image is furniture, never a
+  candidate; identical titles alone are never a stub class.
+* `--fix-ids` sees every invalid character, not just colons and digit leads
+  (issue #3). The Lisey's Story OPF carries ids like
+  `Stephen_King_-_Lisey's_story22` (apostrophes copied from filenames),
+  which passed the original empty/colon/leading-char NCName check
+  wholesale, so a full `--all` sweep left the RSC-005 id errors standing
+  (epubcheck's message says "without colons"; the defect was never colons).
+  The check now enforces the XML NameChar set at every position, and the
+  rename sanitizes every invalid character to `_`, references in sync as
+  before. Fixture: 22 ids renamed, the OPF id error class cleared; the
+  errors that remain on that book are unrelated classes (deprecated body
+  attributes, stray GeoXML elements).
+* `--fix-svg-dup-ids` (issue #2): old calibre 0.8.x SVG page renders repeat
+  glyph ids within one document. The first occurrence keeps its name, later
+  ones gain `_2`/`_3`, and references are deliberately untouched: they
+  resolve to the first definition under every reader's first-match lookup,
+  so first-keeps-id preserves exactly what rendered. Standalone `.svg`
+  entries only; content-document ids are styled and anchor-targeted and
+  stay out of scope. Fixture: 346 ids rewritten, 80 aggregated errors to
+  zero.
+* `--fix-cdata-terminator` (issue #4): completes the truncated `/*]]>`
+  CDATA-terminator comment inside inline `<style>` blocks (the 2002-era
+  conversion lineage whose 19 content docs each died with CSS-008
+  "Premature end of file"; the aggregated gate saw one message). Only the
+  malformed-terminator token class: invalid CSS stays unfixed, stylesheet
+  files are not touched, and an intact closer (even space-separated) is
+  left alone.
+* The markup-fatal trio (issue #5): `--fix-misnested-inline` normalizes the
+  Mobipocket drop-cap mis-nest `<i><b>X</i></b>` (em/strong variants and
+  either order; punctuation between the reversed closers survives outside
+  both spans; a letter run is refused as ambiguous),
+  `--fix-stray-close` removes an end tag whose element has no open start
+  frame (a merely mis-nested pair is never touched; script/style text is
+  skipped to its end tag and an unclosed comment stops the walk), and
+  `--fix-unterminated-attr` closes an
+  attribute value open to the tag's own `>` (`<p class="footnote>`;
+  bare-word values only, and only when no quote closes before the next
+  `<`). Life (Keith Richards): 3 fatals to zero, gate-accepted with the
+  unterminated-attr repair running before the tag-shape readers an open
+  value would otherwise lie to.
+
+Tests: 615 (up from 575), including a real-core row per new flag, the
+wiring-table pins for the five new dests, classifier regressions for both
+sides of the stub fix, and a version-sync pin for uv.lock's root entry
+(the pre-release audit found v0.46.1 shipped with the lock one patch
+stale, a class the pyproject pin cannot see). The plugin ships the default pass only; it
+vendors the changed modules byte-identically and CI byte-compiles it under
+3.11 through 3.14.
+
+## v0.46.1 (2026-10-02)
+
+### The closing sweep: documentation truth after the 0.46 line
+
+The 2026-10-02 verification of the four-lane run left four documentation
+findings against the v0.46.0 tree; all four land here as text plus one
+plugin string. No surface change beyond the plugin's help text.
+
+* The safety contract's lossy enumeration names all four modes: the
+  paragraph listed three and omitted `--strip-stub-docs` (shipped
+  v0.41.0), while the mode list twenty lines up, spec.md, and CLAUDE.md
+  all carry four. Its rationale sentence now says converter-injected
+  content rather than visible markup, true of the document-dropping stub
+  mode too.
+* CLAUDE.md's tests bullet loses its em-dash: the one straggler in the
+  v0.45.0..HEAD markdown delta, recast with a semicolon.
+* ci.yml's core-compat comment describes every module its list runs: the
+  v0.46.0 follow-ups added comment-hyphens, flags-wiring, and
+  flags-real-core to the module list without touching the enumeration
+  above it, so a reader reconciling comment against list found three
+  orphans.
+* The plugin's `customization_help` names `max_log_mb`: the rotation cap
+  is documented in the README and honored by the log writer, but the
+  in-Calibre customization dialog never mentioned the key. Rides the
+  release-attached plugin zip.
+* Suite: 544 -> 575 tests since v0.46.0 (the two follow-up wiring-test
+  commits, unpinned in that entry's count), all green on this text-only
+  sweep.
 ## v0.46.0 (2026-10-02)
 
 ### The last open repair class closes: RSC-016, plus the RepairFlags refactor
@@ -26,7 +121,7 @@
 
 * **`--encode-url-spaces` gained its rename half**, the biggest real-prevalence repair class left (PKG-010: 8,237 occurrences in 284 books at the 2026-09-17 refresh). Archive entries whose names carry raw spaces are renamed to their underscore spellings and every reference is rewritten (OPF, NCX, content `src`/`href`, stylesheet `url()`). The design was decided by experiment, not by the roadmap box's plan: epubcheck 5.3 DECODES references before entry lookup, so percent-encoded entries are unfindable by the encoded references pointing at them (all four raw/percent/underscore combinations built and checked); underscore on both sides resolves and clears PKG-010. Ambiguous renames are refused per entry (a target that already exists, two names converging, a duplicated archive name), a book with any non-UTF-8 document refuses all of its renames, a renamed OPF regenerates a stale container, and the acceptance rides `no_worse` because the gain sits on the warning axis the improvement gate does not measure (same bar as `--fix-cover`).
 * **The matcher fix that explains Theaetetus** (the 2026-09-16 hand-rescue): the encode pass's value class `[^"']` stopped at apostrophes inside double-quoted hrefs, so the 270 raw-space OPF hrefs never matched and the fatal survived. The matcher now takes the tempered-quote idiom (possessive, house style), and an apostrophe in a double-quoted filename is ordinary.
-* **`bindery audit cover`** (advisory, the EPUB3 half of the ruled hybrid): dangling EPUB2 `<meta name="cover">` (the `--fix-cover` class), the EPUB3 `properties~="cover-image"` declaration, and whether the named cover file exists. **`bindery audit tocdrift`**: the structural diff between the NCX navMap and the EPUB3 nav toc, keyed on the full target including fragments -- entries each side is missing plus label mismatches. Both never flag a book and never move the exit code; ToC synthesis stays out of the repair charter permanently. The loader exposes the parsed OPF, the NCX text, and the nav HTML on `Book`.
+* **`bindery audit cover`** (advisory, the EPUB3 half of the ruled hybrid): dangling EPUB2 `<meta name="cover">` (the `--fix-cover` class), the EPUB3 `properties~="cover-image"` declaration, and whether the named cover file exists. **`bindery audit tocdrift`**: the structural diff between the NCX navMap and the EPUB3 nav toc, keyed on the full target including fragments: entries each side is missing plus label mismatches. Both never flag a book and never move the exit code; ToC synthesis stays out of the repair charter permanently. The loader exposes the parsed OPF, the NCX text, and the nav HTML on `Book`.
 * **The audit-refresh deep-dive ran** (read-only FastSweep extract over the full 5,228-book library; raw report in the facility): RSC-005 164,302 -> 13,782 occurrences, RSC-007 3,219 -> 14,162 in 604 books (the classics-wave imports), RSC-020 9,953 in 337 books, PKG-010 8,237 in 284 books (population flat: the rename class was still live, hence this ship).
 * **The roman grammar is one copy again**: pagination.py is the canonical home (it vendors into the plugin; audit does not), audit.py imports `ROMAN_RE`/`ROMAN_MAX`/`roman_value`, and the two byte-identical copies cannot drift. The RepairFlags dataclass refactor stays boxed with its reason (pure churn on the signatures this release touched).
 * Suite: 497 -> 521 tests (tests/test_space_renames.py, tests/test_cover_tocdrift.py, and the updated `--encode-url-spaces` end-to-end pins).
@@ -35,13 +130,13 @@
 
 ### The stdlib XML hardening: untrusted metadata is bounded and entity-refused, no dependency
 
-* **`--audit` and every `load_book` caller no longer decompress blind.** Two protections, both stdlib-only (the defusedxml dependency stays declined): single entries whose declared size exceeds `_MAX_ENTRY_BYTES` (32 MB) are refused before decompression -- a bomb entry in a tiny zip cannot be downloaded into memory -- and the untrusted metadata files (container.xml, OPF, encryption.xml) parse through `_safe_xml_parse`, which refuses DOCTYPE/ENTITY declarations before xml.etree sees them. A refused or unparseable metadata file returns the shell book (the archive verdict is the book's whole story), never a hang or a traceback.
+* **`--audit` and every `load_book` caller no longer decompress blind.** Two protections, both stdlib-only (the defusedxml dependency stays declined): single entries whose declared size exceeds `_MAX_ENTRY_BYTES` (32 MB) are refused before decompression (a bomb entry in a tiny zip cannot be downloaded into memory), and the untrusted metadata files (container.xml, OPF, encryption.xml) parse through `_safe_xml_parse`, which refuses DOCTYPE/ENTITY declarations before xml.etree sees them. A refused or unparseable metadata file returns the shell book (the archive verdict is the book's whole story), never a hang or a traceback.
 * Pinned by `tests/test_xml_hardening.py` (entity-bomb container returns the shell book; clean XML parses; the entry cap is sane). The defusedxml dependency ask is closed as declined with the stdlib shape delivered.
 
 ## v0.42.0 (2026-09-16)
 
 ### The matcher audit executed: every quote-aware tag matcher hardened
-The follow-up box the 0.41.1 hotfix opened is closed with a full per-matcher census, and the audit found the box's own enumeration incomplete. Six GREEDY patterns of the exponential overlap class were unhardened -- `_START_TAG_RE` (the quote-aware start-tag matcher anchoring `--strip-epub3-attrs` and `--strip-bad-attrs`, running under `repair --all` on every content doc; not even named in the box), `_COVER_META_RE`, `_GUIDE_REF_RE`, `prune_dangling_edges`' spine|item|itemref matcher, `_ITEM_TAG_RE`, and `_SPINE_ITEM_RE`. All six are possessive-bounded now: byte-identical on every successful match (the loop is always followed by a literal `>` that the `[^>]` branch cannot consume, so backtracking only ever happened on failure) and linear on failure. Five LAZY patterns of the quadratic class (`_VOID_RE`, the transforms `<img>` matcher, and epub's link/anchor/img matchers) took the disjoint catch-all `[^>"']` instead of a blind possessive swap: mutually exclusive branches cannot re-partition, so any quantifier is linear, and an unterminated quote inside a tag now fails to match instead of pairing with a distant quote and matching across markup. Every well-formed shape is byte-identical. The epub `_IMG_TAG_RE` is renamed `_PRUNE_IMG_TAG_RE` (it collided with the transforms matcher of the same name). Everything else in the package audited safe by construction: tempered-quote value matchers, two-branch quote alternations, `[^>]*` single-class prefixes (linear, not quote-aware -- accepted, documented), lazy paired-delimiter spans, and the analyzer tokenizers.
+The follow-up box the 0.41.1 hotfix opened is closed with a full per-matcher census, and the audit found the box's own enumeration incomplete. Six GREEDY patterns of the exponential overlap class were unhardened: `_START_TAG_RE` (the quote-aware start-tag matcher anchoring `--strip-epub3-attrs` and `--strip-bad-attrs`, running under `repair --all` on every content doc; not even named in the box), `_COVER_META_RE`, `_GUIDE_REF_RE`, `prune_dangling_edges`' spine|item|itemref matcher, `_ITEM_TAG_RE`, and `_SPINE_ITEM_RE`. All six are possessive-bounded now: byte-identical on every successful match (the loop is always followed by a literal `>` that the `[^>]` branch cannot consume, so backtracking only ever happened on failure) and linear on failure. Five LAZY patterns of the quadratic class (`_VOID_RE`, the transforms `<img>` matcher, and epub's link/anchor/img matchers) took the disjoint catch-all `[^>"']` instead of a blind possessive swap: mutually exclusive branches cannot re-partition, so any quantifier is linear, and an unterminated quote inside a tag now fails to match instead of pairing with a distant quote and matching across markup. Every well-formed shape is byte-identical. The epub `_IMG_TAG_RE` is renamed `_PRUNE_IMG_TAG_RE` (it collided with the transforms matcher of the same name). Everything else in the package audited safe by construction: tempered-quote value matchers, two-branch quote alternations, `[^>]*` single-class prefixes (linear, not quote-aware; accepted, documented), lazy paired-delimiter spans, and the analyzer tokenizers.
 
 Pinned by `tests/test_matcher_hardening.py` (15 tests): alarm-guarded spin shapes per greedy site (OPF flavor, the truncated-OPF prune shape), unterminated-quote skips, and well-formed byte-identity. `CLAUDE.md`'s hazard note now carries the census and the rule for new matchers. The full suite is 491 tests.
 

{bindery_cli-0.46.0 → bindery_cli-0.47.0}/plugin/__init__.py

@@ -270,7 +270,9 @@ class BinderyRepair(FileTypePlugin):
     def customization_help(self, gui=False):
         return (
             'Enter a JSON object: {"log": true, "log_path": ..., '
-            '"max_size_mb": 150, "epubcheck_path": null}. '
-            "epubcheck_path enables the experimental post-repair validation "
-            "with the epubcheck binary (slow; off by default)."
+            '"max_size_mb": 150, "max_log_mb": 2, "epubcheck_path": null}. '
+            "max_log_mb is the log rotation cap in MB (default 2, one .old "
+            "generation kept, 0 disables). epubcheck_path enables the "
+            "experimental post-repair validation with the epubcheck binary "
+            "(slow; off by default)."
         )

{bindery_cli-0.46.0 → bindery_cli-0.47.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "bindery-cli"
-version = "0.46.0"
+version = "0.47.0"
 description = "Audit EPUB body text and repair broken EPUBs with deterministic, epubcheck-gated fixes; atomic in-place replacement in a Calibre library."
 readme = "README.md"
 classifiers = [

{bindery_cli-0.46.0 → bindery_cli-0.47.0}/roadmap.md

@@ -40,6 +40,7 @@ The plugin vendors the core and never gates (the epubcheck latency ruling).
 | Waves 15 + blitz | The plugin PEP-758 HIGH + the 3.11-3.14 interpreter matrix, analyzer robustness, the per-matcher census (possessive bounding; cited by tests/test_matcher_hardening.py), stdlib XML hardening, FastDaemon v2 (reconcile-not-retire), `--strip-stub-docs`, `bindery doctor` + `repair --json`, `--encode-url-spaces` entry renames, the cover + tocdrift analyzers, the privacy/GitHub/docs/comment/hygiene batches, requires-python 3.12 with markers, the em-dash sweep | v0.40.0-v0.44.0 |
 | 0.45 | Structured fix records (`fixes`, `ncx_uid_synced`, `watermark_refusals`) in every JSON payload; CalibreQuarry consumes them with the 0.45.0 floor | v0.45.0 |
 | 0.46 | RSC-016 `--fix-comment-double-hyphen` (the last open repair class) and the RepairFlags dataclass | v0.46.0 |
+| 0.47 | The intake-wave batch (five open issues): the stub classifier's body-pool + image-carrier exemption, the NCName-complete `--fix-ids`, and `--fix-svg-dup-ids`, `--fix-cdata-terminator`, `--fix-misnested-inline`, `--fix-stray-close`, `--fix-unterminated-attr` | v0.47.0 |
 
 ## Open work
 
@@ -89,6 +90,12 @@ The plugin vendors the core and never gates (the epubcheck latency ruling).
       docs/screenshots/library-sweep.png against a synthetic testing_facility
       library (invented titles, real dry-run output); the current PNG shows 16
       real rows. Asked 2026-09-15, unanswered.
+- [x] **The 2026-10-02 closing sweep** (the four-lane verification's doc
+      findings against the v0.46.0 tree): the safety contract's lossy
+      enumeration names all four modes, the core-compat em-dash is gone,
+      ci.yml's module comment covers all eleven stack-free modules, and the
+      plugin's `customization_help` names `max_log_mb`. *(SHIPPED v0.46.1:
+      text plus the one plugin string; no surface change.)*
 
 ## The repair domain in the ecosystem parity program (recorded 2026-09-29)
 
@@ -107,7 +114,7 @@ metadata.db registration and the Calibre plugin. Dispositions recorded under the
   embed/subset fonts, jacket, smarten punctuation, remove-unused-css, compress-images,
   upgrade-book, download-external-resources, the --opf rewrite, --cover. They are
   content-improving rather than repair, and dc: metadata editing is out of charter
-  (spec.md:583). Recorded here so the parity ledger's bindery row names them declined,
+  (spec.md:597). Recorded here so the parity ledger's bindery row names them declined,
   not missing.
 - **The `data`-row waiver note**: bindery has updated `data` rows through
   `WritableCalibreDB.set_format` since v0.24.0 (the sanctioned install path);
@@ -134,7 +141,7 @@ metadata.db registration and the Calibre plugin. Dispositions recorded under the
   also fires on `add_format`, so the plugin is byte-idempotent. The vendor slice
   (transforms/epub/pagination/watermark/reserialize + a plugin `__init__`) is
   generated at release from the tagged tree with a byte-equality drift test. Active
-  set = the gate-safe default pass only: every structural repair and the three
+  set = the gate-safe default pass only: every structural repair and the four
   lossy strips stay CLI-only, because epubcheck cannot gate inside Calibre (no jar/JVM
   and seconds-per-book latency). Config via `site_customization` JSON (`log`,
   `log_path`, `max_size_mb`, recorded default 150MB); `publish.yml` attaches

{bindery_cli-0.46.0 → bindery_cli-0.47.0}/spec.md

@@ -13,15 +13,17 @@ restyle, re-compress, or restructure content, and it does not attempt to fix arb
 schema (RSC-005) violations, which are usually harmless to readers and not safely
 mechanizable.
 
-The deliberate exceptions to "semantics-preserving" come in opt-in groups. The sixteen
+The deliberate exceptions to "semantics-preserving" come in opt-in groups. The twenty-one
 **structural repairs** (`--fix-empty-body`, `--fix-missing-title`, `--fix-id-colons`,
 `--fix-page-map`, `--strip-epub3-attrs`, `--downgrade-epub3-tags`,
 `--unwrap-block-in-inline`, `--strip-invalid-value`, `--unwrap-illegal-tags`,
 `--prune-missing-resources`, `--strip-broken-anchors`, `--encode-url-spaces`,
-`--fix-container`, `--fix-media-types`, `--fix-cover`,
-`--fix-comment-double-hyphen`) alter
-markup structure or fabricate minimal content; the three
-**lossy modes** (`--strip-pagination`, `--strip-broken-tags`, `--strip-watermarks`) remove
+`--fix-container`, `--fix-media-types`, `--fix-cover`, `--fix-comment-double-hyphen`,
+`--fix-svg-dup-ids`, `--fix-cdata-terminator`, `--fix-misnested-inline`,
+`--fix-stray-close`, `--fix-unterminated-attr`) alter
+markup structure or fabricate minimal content; the four
+**lossy modes** (`--strip-pagination`, `--strip-broken-tags`, `--strip-watermarks`,
+`--strip-stub-docs`) remove
 content a converter injected rather than content the author wrote. A fourth group of
 **safe opt-ins** (`--fix-ids`, `--add-img-alt`, `--strip-bad-attrs`,
 `--escape-unknown-entities`) repairs without altering visible markup, and
@@ -123,7 +125,7 @@ byte-for-byte. The normal gate applies.
 
 ### Opt-in: structural repairs
 
-Sixteen repairs go past well-formedness and therefore require their own flag; none is ever
+Twenty-one repairs go past well-formedness and therefore require their own flag; none is ever
 part of the default pipeline:
 
 - **`--fix-empty-body`**: `&nbsp;` inside a strictly empty `<body></body>` ("body
@@ -232,12 +234,60 @@ use, with the `partial` rule intact.
   never touched, and each comment's `-->` terminator stays intact. An unclosed comment is a
   different fatal class and is left alone. It applies to content documents, the NCX, and
   the OPF alike, and every replaced sequence is counted.
-
-Fifteen of the sixteen are evaluated by the normal `gate`: unlike the lossy strips,
+- **`--fix-svg-dup-ids`**: rename duplicate `id` values inside standalone `.svg`
+  entries (old calibre 0.8.x SVG page renders repeat glyph ids within one document;
+  four books from one intake wave carried 445 to 1,339 'Duplicate "glNNNN"' RSC-005s
+  each). The first occurrence keeps its name; every later one gains `_2`, `_3`, ... in
+  document order (prefixed with `_` until the name is free). Internal
+  `xlink:href="#x"` / `url(#x)` references are deliberately never rewritten: they
+  already resolve to the first definition under every reader's first-match lookup, so
+  first-keeps-id preserves exactly what rendered. Scope is standalone `.svg` entries
+  only: content-document ids are styled by CSS selectors and targeted by anchors, and
+  renaming those needs a reference graph this repair does not build.
+- **`--fix-cdata-terminator`**: complete a truncated CDATA-terminator comment in an
+  inline `<style>` block: `/*]]>` becomes `/*]]>*/` (a comment-wrapped CDATA block
+  whose closing `*/` the converter dropped; epubcheck's CSS parser reads the dangling
+  `/*` as an unterminated comment and answers CSS-008 'Premature end of file' once per
+  affected document, 19 identical errors on the 2002-era fixture). Deliberately narrow:
+  only the malformed-terminator token class, inside `<style>` blocks only; invalid CSS
+  properties and selectors stay unfixed (renderers degrade gracefully), stylesheet FILE
+  entries are not touched, and a closer that already sits intact (possibly after
+  whitespace) is left alone.
+- **`--fix-misnested-inline`**: rewrite the Mobipocket drop-cap mis-nest
+  `<i><b>X</i></b>` to `<i><b>X</b></i>` (em/strong variants and either tag order
+  included). Only the both-open-then-reversed-close shape is rewritten: any match is
+  XML fatal, and swapping the closers is the unique well-formed form carrying the same
+  two spans over the same text. The Exile variant, where a short punctuation run sits
+  between the reversed closers (`<i><b>xile</i>,</b>`), is covered: the punctuation
+  survives outside both spans, byte-preserved. A letter run between the closers
+  (`<i><b>S</i>ome</b>`, where `<b>` spans `Some` and `<i>` only `S`) is refused: which
+  span keeps the letters has no deterministic answer. Same-tag pairs, attribute-bearing
+  starts, and markup crossing the run are untouched.
+- **`--fix-stray-close`**: remove an end tag whose element has no open start tag
+  anywhere above it (the stray extra `</div>` on a shell page, which cascades into
+  'body must be terminated by the matching end-tag'). The rule is deliberately narrower
+  than a nesting repair: a close is removed only when NOTHING by that name is open, so
+  a merely mis-nested pair (`<div><span></div></span>`) is left exactly as written
+  (its `</span>` still has an open frame and is never misread as stray). Self-closed
+  and void starts open no frame; raw-text elements (`script`, `style`) are skipped to
+  their end tag so JS/CSS text is never read as markup; the walk stops at an unclosed
+  comment or CDATA opener, where the parser stops reading markup too.
+- **`--fix-unterminated-attr`**: close an attribute value left open to the tag's own
+  `>` (`<p class="footnote>` -> `<p class="footnote">`; the parser swallows prose into
+  the value until the next quote, meets a `<`, and dies with the RSC-016 fatal "the
+  value of attribute ... must not contain the '<' character"). Three guards keep it off
+  legal `>`-bearing values: the value is one bare word, the `>` must be followed by no
+  quote before the next `<` (a real multi-line value closes quote-first), and only
+  double-quoted values are judged. It runs before the other tag-shape readers, which an
+  open value otherwise lies to.
+
+Nineteen of the twenty-one are evaluated by the normal `gate`: unlike the lossy strips,
 their benefit is visible to epubcheck (they clear errors), so a run with no measurable
-improvement is a noop and nothing is applied. The exception is `--fix-cover`, whose
-gain is invisible to epubcheck: it is accepted under the same `no_worse` bar the
-lossy strips use, with the partial rule intact. CDATA sections are never rewritten
+improvement is a noop and nothing is applied. The two exceptions are `--fix-cover` and
+`--encode-url-spaces`, whose gains sit on axes epubcheck does not measure (cover
+wiring is not a schema finding; the rename half's PKG-010 gain sits on the warning
+axis): both are accepted under the same `no_worse` bar the lossy strips use, with
+the partial rule intact. CDATA sections are never rewritten
 anywhere; comment bodies only by the double-hyphen fix above, and nothing else inside
 them.
 
@@ -264,7 +314,7 @@ Since this removes visible text from the reading experience, it is lossy by desi
 Strips known producer and redistributor watermarks out of EPUBs (e.g. OceanofPDF.com, ABC Amber LIT Converter). The removal is a balanced-element surgery rather than regex slicing: it locates the stamp and deletes the outermost wrapper whose *entire visible text* is the watermark, ensuring prose that merely mentions the URL is preserved. Also drops known zero-byte marker files. Like other lossy operations, this is verified via `no_worse`.
 
 ### Opt-in, lossy: stub-document strip (`--strip-stub-docs`)
-Drops spine documents whose entire visible text is one identical short placeholder repeated across the spine: the Bookmate/DRM-sample export whose chapters are all the same "content unavailable" notice, valid XHTML that epubcheck passes. The identity rule mirrors the emptytext analyzer's placeholder signals and is deliberately conservative: a candidate doc's visible text is 12-600 characters (short, not blank) and IDENTICAL to the text of at least two other spine docs, and the class covers at least 30% of the spine; only the single most common repeated class qualifies.
+Drops spine documents whose entire visible BODY text is one identical short placeholder repeated across the spine: the Bookmate/DRM-sample export whose chapters are all the same "content unavailable" notice, valid XHTML that epubcheck passes. The identity rule mirrors the emptytext analyzer's placeholder signals and is deliberately conservative: a candidate doc's visible body text is 12-600 characters (short, not blank) and IDENTICAL to the text of at least two other spine docs, and the class covers at least 30% of the spine; only the single most common repeated class qualifies. The pool is the `<body>` span, not the whole document (whole document when a doc has none; a converter stamps the same head `<title>` on every shell page, which used to weld unrelated docs into a fake stub class), and an image-carrier page (empty body text, an `<img>`/`<image>` present) is exempt outright: those are illustrated part-divider pages, furniture the strip must never delete (the Great Change misfire, 2026-10-05).
 
 The drop cascades fully: the placeholder archive entries, their manifest items and spine itemrefs (with every package edge rewritten), their NCX navPoints (the always-on playOrder resequencing heals the sequence afterward), and their nav toc `<li>` entries. Refusals, each reported as a no-op: no repeated class, the class under the fraction bar, and the class covering the whole spine (a book whose every spine doc is the same stub is EMPTY; it needs a re-source, never a repair). Accepted under the `no_worse` bar with the partial rule intact.
 
@@ -445,7 +495,7 @@ Two more advisory analyzers ship in v0.44.0. `cover` reads the package's own cov
 wiring (the EPUB3 half of the ruled hybrid): a dangling EPUB2 `<meta name="cover">`
 (the `--fix-cover` repair class), an EPUB3 `properties~="cover-image"` declaration,
 and whether the named cover file exists. `tocdrift` diffs the two tables of contents
-an EPUB 3 book may carry -- the NCX navMap against the nav document -- reporting
+an EPUB 3 book may carry (the NCX navMap against the nav document), reporting
 entries each side is missing and labels that disagree; it feeds decisions, never a
 rewrite (ToC synthesis is out of the repair charter permanently). Both never flag a
 book and never move the exit code.

{bindery_cli-0.46.0 → bindery_cli-0.47.0}/src/bindery/__init__.py

@@ -11,7 +11,7 @@ Calibre plugin built from this core.
 Single source of truth for the version. pyproject.toml must match VERSION below.
 """
 
-VERSION = "0.46.0"
+VERSION = "0.47.0"
 __version__ = VERSION
 
 from .epub import RepairReport, repair_epub  # noqa: E402