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

Debug mode: enabled

Effort study: railspike-effort / v3/reply_composer_focus_layout · Host ledger

Current outcome

Implementation is ready for user testing on feat/reply-composer-focus-layout. All three planned stages and focused browser/application checks are complete. Physical iPhone Safari/Firefox keyboard, chrome, caret and recovery acceptance remains pending; use the checklist below. No full-suite pass, merge, push or deployment is claimed.

Execution state

• Branch: feat/reply-composer-focus-layout; approved plans/debug log organized and committed as e690c0d2 before this summary or application changes.

• Step 3 approved. At Step 4 entry no effort study was selected; subsequent P2 standing enrollment is recorded below. Actual iPhone Safari/Firefox acceptance remains pending user testing, with no upfront device gate. No merge, push or deployment authorized.

• Request state: Preserve the board correction's genuine collapse/expansion and stable first paint; implement approved Option B for standard reply forms. No additional or deferred implementation requests.

Stage 1 - One editor and preserved inline interaction

• Changes: Standard inline replies now render the canonical editable body directly in a server-collapsed wrapper, without a readonly prompt or disclosure arrow. Board/reply composers share the existing state lifecycle and collapsed CSS. Reply links use direct native focus once; no deferred focus handoff or scripted scrolling. Signing recognizes the new inline wrapper; original form controls and APIs remain intact. Updated consumers/tests that targeted the removed prompt.

• Verification passed: node tests/browser/reply_compose_focus.mjs covers 390px/1280px across Light/Word97/Chicago/Midnight, scripts-held first paint and initialized geometry, same editor/one focus, actual height expansion/empty collapse, drafts/selection, error/retry and one real reply, no-script controls. inline_reply_link.mjs covers native reveal, modified/middle-click and no-script fallback; board_compose_focus.mjs passes unchanged board behavior. php tests/run.php LocalAppSmokeTest::testApplicationRendersCoreRoutes BrowserSigningNormalizationTest LazyComposeSigningTest passes 86/86 after replacing obsolete prompt/scroll source assertions. Changed PHP/JS syntax and diff checks pass.

• Test corrections: Thread pages make background requests, so first-paint coverage waits for explicit script initialization rather than network idle. Reply success waits for the resulting post anchor before counting records, rather than matching the thread URL that was already open.

• Notes at the Stage 1 boundary: Stage 2 had not started. Physical iPhone behavior remains pending user acceptance; existing phone toolbar behavior is replaced in Stage 2.

Stage 2 - Separate phone writing, feedback and actions

• Changes: Standard inline and dedicated replies now share a bounded phone composing panel. The original form, textarea, feedback and action row stay together; the editor flexes into the remaining space and Post stays in normal layout within the panel. Open menus reserve their footprint instead of covering the editor. A noninteractive spacer preserves page flow during editing. Focus exit, zoom, tiny viewports or oversized feedback/attachments restore ordinary flow without moving/replacing controls or clearing the draft.

• Verification passed: Chromium reply_viewport.mjs reproduces the previous overlap at multiple page offsets with a simulated 420px viewport, asserts editor/action/error/menu separation, preserves long draft/selection through viewport changes/scroll/rotation, tests readable menus/native primary and anonymous submitters, oversized-feedback/tiny-viewport/zoom fallback, focus restoration, desktop and Forte exclusion across Light/Word97/Chicago/Midnight. Firefox and WebKit pass the same viewport matrix, including fallback checks, on final runs. The first WebKit run timed out before finding the synthetic form; rerunning the same check passed all assertions.

• Verification passed: reply_viewport_flow.mjs covers both isolated real reply routes, recoverable failure, retained draft, retry, duplicate pending activation and exactly one persisted reply. compose_layout_stability.mjs passes 24 route/theme/viewport/permission cases and both Outbox saves. The Stage 1 reply focus/paint/collapse matrix also passes with the new panel. JS/PHP syntax and diff checks pass. Stage 1 commit: 30846b00.

• Notes: Original reply tests asserted only action visibility; the matrix now explicitly rejects intersections and verifies fallback. Physical iPhone acceptance remains pending. Concurrent changes to project workflow/study files are outside this feature's code changes and will not be included in its stage commit.

Effort study enrollment and Step 4 discovery snapshot

• Recorded UTC: 2026-10-21T18:01:30.581350+00:00.

• Study/cycle: railspike-effort / v3/reply_composer_focus_layout; parent: merged iphone_reply_submit cycle (not previously assigned a study ID). Protocol P2 standing enrollment appeared in the project's AGENTS/ledger during Stage 2; this existing cycle joins at this checkpoint. Intake and Steps 1–3/Step 4 entry projections are missing, not reconstructed.

• Scope v1: approved Step 2/3 contract, starting revision bc42277b; single-editor inline entry, non-overlapping phone composing space in both reply routes, real submission/recovery and shared-consumer checks, with actual iPhone user acceptance pending. No scope expansion or split.

• Reference R1: calibration report SHA-256 d3a30b6400b83bb25283b4234b41ba07515df9a696ea44de59c10186103771b0 and diff review 068b3519326e08a939df40b5abe0106ddcbab80bd796616c76b5084a21462b84 verified in /home/wsl/railspike-dev/docs/research/effort_estimation/. Frozen feedback SHA-256 ef335604ecd5b51e5c04067d88e4d6f01cb4fdb21b8f4d7dba7b5b90e4012bfc restored and verified from Git revision 4f6cfe2b436fae8d939b1349fe724a3879ee0a0e rather than silently using updated feedback.

• Rater/exposure: Codex, single rater; approved plans, original reply implementation/feedback and current Stage 1/2 work were visible. This is an exposed estimate collected after implementation began, not an independent or prospective intake rating.

• Total projected scope score: 0.65 from the cycle's starting point, including planning, implementation, required verification/documentation and ordinary corrections. Similar to R1 composer decluttering (0.60), with additional viewport/focus recovery checks; smaller than the package/recovery startup references. This is not remaining work, elapsed effort or user acceptance.

• Approval context: Steps 1–3 approved; study collection does not widen implementation scope. Ledger index update is included as a feature-only closeout patch; unrelated concurrent ledger changes remain outside stage commits. Maintainer comfortable-fit judgment and accepted study outcome remain pending.

Stage 3 - Recovery, regressions and device-test handoff

• Changes: Extended isolated real-route tests to use a simulated keyboard viewport for both entries, verify long-draft selection through resize, correct a real unsupported-character validation warning, and require editor/error/action separation across Light/Word97/Chicago/Midnight. Explicit Reply-link intent activates the phone layout before typing while retaining native focus once; modified clicks and no-script destinations remain unchanged. Corrected theme specificity so Midnight card padding/stack margins cannot shift the panel outside the visible area; added a whole-panel viewport assertion. Phone signed-image publication now checks anonymous refusal retains the reply draft/image and actions remain clear of the editor. Updated Feature 16 and UI audit notes without claiming iPhone resolution.

• Verification passed: Chromium real-route flow covers both entries, validation correction, long draft/selection, full panel bounds/non-overlap, recoverable prepare failure, preserved draft, retry, pending duplicate suppression and exactly one persisted reply after reload. Native keyboard activation tests the Reply link with the mocked viewport; inline_reply_link.mjs separately covers real pointer/modified/middle clicks with the native viewport. Mocking VisualViewport does not resize browser hit testing, so attempting a real pointer click under the mock was corrected to this explicit split rather than forcing a covered click.

• Verification passed: Final shared viewport matrix covers errors/open menus clear of the editor, viewport offsets/scroll/rotation, selection, native primary/anonymous submitter identity, large-feedback/tiny-viewport/zoom fallback, focus restoration and desktop/Forte exclusion. Stage 1 reply and board focus/paint/collapse matrices pass after the layout changes. First-paint action stability remains covered by the 24-case compose layout check and Outbox saves.

• Verification passed: php tests/run.php LocalAppSmokeTest::testApplicationRendersCoreRoutes BrowserSigningNormalizationTest LazyComposeSigningTest OfflineOutboxComposeTest ImageContentTest (93/93); reply_actions_layout.mjs (20 combinations); compose_disclosure.mjs; image_compose_transport.cjs; isolated image_content_browser.mjs (signed thread/reply, lost-response retry, phone reply anonymous-image refusal, protected image delivery, Forte, backup/restore, rebuild, flag/member handling). No full-suite pass is claimed. Final Firefox/WebKit real-flow runs pass the same two-entry, four-theme assertions, including keyboard Reply-link entry before typing. WebKit navigation-cancelled background requests are collected separately using the existing harness convention; unexpected page errors still fail.

• Visual review passed: Inspected actual inline Light/Word97 and dedicated Chicago/Midnight phone panel screenshots. Found and corrected Midnight's spacing override, then reinspected the corrected dedicated Midnight panel. Local screenshots: /tmp/reply-composer-review/; not committed as product assets. Simulated keyboard screenshots are supplementary evidence, not physical iPhone proof.

• Notes: Planned 3 stages implemented. Planning commit e690c0d2; Stage 1 30846b00; Stage 2 cc7eb3bc; this stage's code, tests, closeout and feature ledger index are committed together. The final handoff log records the final stage revision. Corrections above are within approved scope; no new product requests, splits, schema migration, deployment, merge or push. Other workflow/study edits in the shared workspace are preserved and excluded from this feature's commits.

User iPhone acceptance checklist (pending)

1. In Safari and Firefox, open an ordinary thread and tap the compact reply editor once, then repeat via the post's Reply link. Confirm one editor expands, Post stays keyboard-reachable and neither the action row nor open menu covers the draft/caret.

2. Try a long draft, selection, page/editor scrolling, portrait/landscape, closing/reopening the keyboard and pinch zoom. When space is inadequate, ordinary flow must expose the full draft, feedback and retry controls. Leaving an empty, attachment-free, warning-free inline composer should restore the compact prompt.

3. Repeat writing/submission on the dedicated reply page. Correct a validation warning; after a recoverable connection failure, verify the draft survives and retry creates exactly one reply. Test an attachment and anonymous option as applicable.

4. Record iPhone model, iOS/browser versions, orientation and failures. Actual keyboard/browser-chrome/caret behavior and device acceptance remain pending user testing; automated desktop-engine evidence does not complete these checks.

Study closeout

• Recorded UTC: 2026-10-21T18:11:07.230919+00:00.

• Stable ID v3/reply_composer_focus_layout, study railspike-effort, scope v1 and frozen R1/P2 unchanged. Total projected score 0.65 reaffirmed from the starting point; no newly expanded scope. Earlier checkpoints remain missing because enrollment occurred during implementation.

• Separate retrospective scope assessment: 0.65, Codex single exposed rater (precise model identity not independently verified). Delivered: one-editor inline interaction/first-paint behavior and reserved phone writing/feedback/actions on both reply routes, with normal-flow recovery and shared-consumer regression evidence. Existing canonical submission/uploader/signing/test machinery reduced new work relative to the larger R1 references.

• Baseline bc42277b; delivered stages 3/3. Final production revision is this stage's commit, recorded in the handoff debug log after creation; preserve all stage commits/diffs. Ordinary corrections: obsolete test fixtures, thread background-request waiting, synthetic-viewport hit-test limitations, Reply-link intent and theme specificity. No separate new requests or split outcomes.

• Verification/omissions: Focused and browser evidence above; full suite not run; physical iPhone acceptance and maintainer comfortable-fit/acceptance date pending. Unknown active work remains unknown; no labor or elapsed-effort claim. Technical implementation handoff does not increment accepted study outcomes.