Feature plan: Step 1 · Step 2 · Step 3 · Step 4

Debug mode: enabled

Restored inherited state: Step 3 approval, planning and Stages 1–6 commits, enabled debug, R1/P2 study history and pending bookkeeping were reconciled before Stage 7 continuation; no scope change.

Effort study: railspike-effort / v3/docs_independent_publication, standing enrollment P2; host ledger.

Execution State

Step 3 approved; branch feat/docs-independent-publication, planning-only commit f2d718a0. All eight stages complete and locally verified; review/maintainer acceptance pending. Production implementation ended at 7c255bfd; later stages add tests/evidence/operator documentation. No dependency additions or scope change.

Step 4 Entry Snapshot

UTC 2026-10-23T14:24:04Z; scope v1 clarified / R1 / P2, same stable cycle and start ddc9d2c020addbc41b19f543d0247b83259c6f69. 0.85 projected total reaffirmed before application implementation. Prior 0.75 intake/Step 1/Step 2 retained; exact R1 hashes verified at Step 3. Codex single exposed rater, prior source/score exposure and non-pristine intake disclosed. Unknown active-work measurements remain unknown; no independent/accepted outcome. Predecessor is existing infrastructure, not work forecast again.

Stage 1 - Ownership, rendering seam and frozen baseline

• Changes: Proved the canonical PlatformDocsController/RouteServices/TemplateRenderer path can render anonymously without constructing or replacing a read model. Docs storage will be a profile/destination-specific sibling of the forum static root, with its own publication lock/current pointer. Legacy retention is a bounded reference to the previous forum release, serving only canonical docs/public assets; only those resources receive an ownership exemption during forum cleanup. A prototype retained a 1,002-file docs subtree while retiring old forum HTML after examining seven entries. No bulk docs copying is required to pin that reference.

• Verification: /tmp/docs-independent-proof.php confirms isolated canonical rendering creates no database while Application::handle does; prototype cleanup retains only docs. /tmp/docs-independent-context-proof.php renders default/QDB/private profiles without database creation. Frozen source f2d718a0, 1,465 docs, digest 41302ba5246b76d17202badb6bed4de02850a2ed4861772279cf4d2207665c5e: cold publication 91.811 s, changed-tree new-record CLI import 62.267 s with 1,896 docs progress messages, unchanged duplicate 7.177 s, shared refresh 6.085 s. Changed input was rendering-source whitespace; docs inventory stays identical. README differs only by forum version marker. Evidence.

• Notes: Prototypes/fixture data are isolated temporary work; application source is unchanged. Implementation must validate reference roots and each served path, skip protected docs on resumed sweep frames as well as new traversal, and keep all forum HTML/offline cleanup active. Docs notifications require an independent endpoint and pending-version storage; legacy HTML can adapt bounded metadata on request, while explicit docs publication completes static migration. Cold static-only setup remains the documented separate publication operation. 0.85 total reaffirmed; Stage 2 may begin after this commit.

Stage 2 - Independent complete publication and CLI

• Changes: Added ./v3 docs publish with help, destination/profile selection, readable/JSON output and actionable retry errors. Docs publish to a sibling -docs root using their own lock/releases/current pointer. An anonymous DocsPageRenderer reuses the canonical docs controller/RouteServices/TemplateRenderer and explicitly rejects read-model access. DocsArtifactBuilder produces the complete library/assets/downloads/runtime files; existing integrity and streaming-copy policy validates reuse or full publication. Independent signatures/version IDs exclude forum HEAD/retraction/classification generations and include docs/render inputs. Template version endpoint/storage metadata are configurable with unchanged forum defaults.

• Verification: IndependentDocsPublicationTest plus DocsPublicationTest: 12/12 passed; final independent publication/reader run 21/21 passed. Covers cold/repeat/source-change publication, unchanged forum pointer/database/offline sentinels, record-only version stability, complete downloads/runtime retention, JSON/help/errors, failed template retry, changed inputs during rendering, unsafe storage and private-profile anonymous docs without database creation. Existing large-download bounded-memory test still passes. PHP/shell syntax and whitespace checks pass.

• Notes: Stage 1 commit 04c18547. Initial CLI error assertion checked JSON escape representation rather than its decoded next-action field; corrected the assertion and normalized JSON slash output, then reran successfully. Reader routing/version-check state are Stage 3; existing combined forum publication remains until Stage 5. No new dependencies/schema.

Stage 3 - Coherent serving, retained assets and docs identity

• Changes: Added a read-only DocsResponder before forum configuration/database startup. Native docs HTML, browse/search, safe downloads, missing recovery and immutable fingerprinted assets resolve from one independent publication, including cookie/private requests. Scoped favicon URLs avoid changing forum favicon behavior. /api/docs-version reads bounded publication metadata; dynamic fallback uses the docs version domain. Shared version checker reads an optional per-page pending-version storage key, preserving the forum default.

• Verification: Independent publication/readers plus existing docs/version checks 28/28 passed; final independent/version-state run 12/12 passed with a new JavaScript state-isolation regression. Published docs work with a nonexistent read-model path, retain exact old CSS after source asset changes, remain readable after source deletion/record commits and in private mode, and serve exact downloads. HEAD/version/missing/traversal checks pass. Pending forum updates are neither consumed nor overwritten by docs checks; a satisfied docs update clears without a reload loop.

• Notes: Stage 2 commit bd9e4a5d. Initial retained-asset test exposed a source-file fingerprint resolver being used against already-fingerprinted release files; now validated against the retained file's actual digest prefix. Existing combined forum docs temporarily retain their legacy version context until detachment in Stage 5. Legacy release adaptation/protection is Stage 4. No dependencies/schema/navigation changes.

Stage 4 - Bounded legacy retention and cleanup ownership

• Changes: Added a small relative legacy reference and safe locator fallback for prior release/direct layouts. Forum activation and invalidation preserve that reference without docs inventory/hash/copy work. Cleanup exempts only pinned docs paths/public assets, including resumed traversal frames, while forum HTML and offline retirement remain active. Existing legacy HTML receives docs version/endpoint/storage/script/favicons metadata adaptation only for the requested page; stored library bytes remain unchanged. Explicit docs publication transitions to native storage and failure retains the legacy library.

• Verification: Legacy/independent suites 9/9 passed after final fixture/preview assertions. Full retraction suite plus migration and classification-guard checks 20/20 passed. Legacy docs remain readable after forum pointer replacement, invalidation and cleanup that removes old/new forum HTML. A 1,002-file docs subtree is skipped; resumed docs traversal examines zero entries. Direct layout, readonly preview, failed/successful explicit migration, unchanged forum/read-model bytes, escaping refs and symlink rejection pass.

• Notes: Stage 3 commit 7c4f7170. Regression legacy fixtures now model old forum-version metadata without relying on the soon-to-be-detached forum builder. Reference writes are atomic bounded bookkeeping; missing/damaged docs do not cause forum publication failure. Docs metadata-write errors give independent publication guidance rather than aborting forum work. No dependencies/schema.

Stage 5 - Forum/import detachment

• Changes: Removed documentation inventory/render/reuse/input-validation paths from forum builds and shared refreshes, and removed docs evidence validation from forum activation. Forum publishers pass only bounded docs-locator context for retained access/setup guidance. Shared seeding excludes docs-owned paths; private/forum/retraction/classification behavior remains intact. Removed the temporary combined-render version override. Updated prior combined-publication regressions to the approved independent contract; full-library static regression now uses the dedicated docs publisher.

• Verification: Native/legacy/docs flow plus actual full/shared CLI smoke 15/15 passed. Cold/new-record/duplicate imports succeed with docs bulk entry points instrumented to throw, broken docs templates and a file blocking docs storage. Existing docs versions/bytes remain unchanged through actual full/shared forum commands after docs deletion and changed rendering code. Forum artifacts contain no docs directory/evidence, docs changes do not reject forum activation, and private imports skip public forum/offline artifacts. Full/shared CLI smoke now finishes in 1.39/3.08 s in this run instead of building the library.

• Notes: Stage 4 commit ccb115ce. Source selection is bounded metadata/stat work; no full-library validation/copy is substituted for rendering. Missing docs give the separate publication command as recovery guidance. Runtime reader/evidence hashes and legacy snapshots remain owned by docs publication, independent of forum activation. No dependencies/schema.

Stage 6 - Reported WSL resume and concurrent failure boundaries

• Changes: Added real CLI interrupted-import/resume cases for absent, damaged and legacy docs, then changed application/docs sources and trapped bulk docs entry points before remote-free resume. Added independent-process barriers that pause a docs publication inside its own lock while a real forum import completes, followed by both successful and failed docs continuation.

• Verification: Focused independent/legacy/normal-flow/recovery run 18/18 passed. Resume retains one imported commit, clears its journal, publishes current forum content and leaves the selected docs state untouched with zero docs bulk calls. Imports complete while docs publication is still paused; later docs success/failure does not alter the forum pointer/database, and failed docs work retains the prior docs publication. All original import/publication/recovery plus full retraction and classification guard checks 31/31 passed on this checkout, including five checkpoint failures, legacy image archives, HTTP failures and a closed output pipe.

• Notes: Stage 5 commit 7c255bfd. Tests use the actual WSL CLI and isolated destinations/archives; source changes and broken docs are deliberate failure inputs. No active user import or deployment process was changed. No production corrections or dependencies were needed in this stage. Complete-library reader/performance acceptance remains Stage 7.

Stage 7 - Complete-library operator and reader acceptance

• Changes: Recorded portable acceptance evidence for a frozen full library and the final focused current-checkout regression. Frozen docs match the Stage 1 baseline exactly: 1,465 documents, digest 41302ba5246b76d17202badb6bed4de02850a2ed4861772279cf4d2207665c5e. No production corrections were needed.

• Verification: 60/60 focused tests passed, including complete inventory/static parity and actual full/shared CLI commands. Frozen real CLI changed-tree import 2.110 s versus 62.267 s baseline; duplicate 1.784 s, full forum build 1.289 s, shared refresh 0.780 s, all with zero docs bulk calls while bulk entry points throw and the docs template is broken. Legacy library remained selected and forum output contained no docs. Actual independent JSON docs publication completed all 1,465 docs in 36.696 s, preserving the forum pointer and exact database bytes. An open legacy reader detected docs publication and reloaded to its new version; unrelated forum pending state survived. A subsequent real forum import changed only forum version and produced no docs update prompt.

• Reader/UI verification: Served dynamic/native-static 1280/390px journeys pass Tools → Docs → browse, 40-item pagination, Back, group/folder expansion, search/result/return-state, exact README download, empty results, failed search/retry and missing-document recovery; loaded/expanded/empty/failure states have no page overflow. Existing no-JS keyboard-scroll/focus/CLI-fragment suite passes 16 scenarios across default, Word97, Chicago and Midnight on both hosts. Notifications-disabled behavior passes. Representative phone/desktop/midnight screenshots visually inspected; no new navigation/copy was added. Evidence includes local screenshot paths/hashes.

• Notes: Stage 6 commit 5567310c. Source was frozen during complete-library runs to avoid self-invalidating publication inputs. The full current-checkout test run started before the final documentation edits; production source remains unchanged afterward. Timings are local single-run observations under uncontrolled load, not a speed guarantee. Full repository suite, deployment, external host and merge/push are outside this acceptance; no claim of those checks. No dependency/schema change.

Stage 8 - Operator guidance and closeout

• Changes: Updated the import, deployment and Markdown regeneration runbooks and CLI reference for separate initial setup/refresh, matching destination/profile, sibling storage, legacy retention, private/anonymous readers, independent notifications and failure/retry. Forum --resume retries forum publication; docs retry uses the separate command. Preserved the predecessor outcome and linked its follow-up. Reconciled standing study ledger and this summary; todo.txt is untouched.

• Verification: ./v3 docs publish --help, bash -n v3, PHP syntax for all 23 changed PHP files, git diff --check, local plan/runbook links and approved file/dependency inventory pass. Fresh/readable/JSON/failure-retry/private setup and legacy migration were exercised by the focused CLI regressions; the documented explicit-destination command completed a real full-library publication in Stage 7. Source/template/deployment instructions agree with the implemented command. No production code changed after the 60/60 regression and browser acceptance.

• Notes: Stage 7 commit 0eabe1a9. Full repository suite and production host/deployment remain unverified, outside this local acceptance; no merge/push or active-instance publication performed. No schema migration, new navigation destination, UI copy cleanup or dependency change was required. Third-party delta zero: no package/lock/vendor/CDN additions or updates. Temporary acceptance uses existing approved local browser tooling only.

Commits and Study Outcome

Planning f2d718a0; Stage 1 04c18547; Stage 2 bd9e4a5d; Stage 3 7c4f7170; Stage 4 ccb115ce; Stage 5 7c255bfd; Stage 6 5567310c; Stage 7 0eabe1a9; Stage 8 is the closeout commit containing this section. Each stage includes its summary. A separate log-only handoff commit follows Stage 8.

Stable cycle v3/docs_independent_publication, scope v1 clarified / R1 / P2; start ddc9d2c020addbc41b19f543d0247b83259c6f69, predecessor v3/docs_publication_policy. Earlier intake/Step 1/Step 2 0.75 and Step 3/Step 4 entry 0.85 projected total remain unchanged. Retrospective scope assessment: 0.90, Codex single exposed rater; separate from forecast and not an actual-work measure. Compared with R1's provisional 1.0 startup recovery, this cycle reuses established rendering/CLI/recovery infrastructure but required a new ownership boundary, anonymous startup isolation, legacy cleanup retention, version identity and full-library/concurrency acceptance. These details justify the modest retrospective difference without a conversion from line counts or timings. No new request, scope split, omitted approved stage or retrospective forecast replacement. Production corrections remained within Stages 2–5; fixture/assertion corrections are disclosed above.

Exact R1 report/diff/frozen-feedback hashes reverified at closeout (d3a30b64…, 068b3519…, ef335604…; full recorded references remain in the earlier snapshots). R1 is unchanged; no independent rating, pristine intake, measured active work or inferred hours. Local technical outcome complete with focused/normal-path/browser checks above; full suite and production acceptance unclaimed. Maintainer comfortable-fit judgment and acceptance pending; accepted-outcome count unchanged. Duplicate worktree copies are one observation; no upload authorized or performed.