[andy-zen-dev]

Product outcome

Reader 0.4.0a1 delivers a complete private manuscript -> actual analyzer -> compare -> HTML/JSON/CSV path and trustworthy frozen-work display. Accumulated unpublished work is now repository code, not a second prototype runtime.

Closes #279. Closes #286. Partially addresses #283 (touch targets only).

Delivered

• Installed scriptorium compare: all 29 canonical metric rows, exact JSON decimal differences (right minus left), percentage-point deltas, explicit units/status and source/profile/dictionary identities.
• Strict bounded input and consistency checks: required counts, mean lengths, punctuation rates, vocabulary partition/windows and same-text/method invariance. Dictionary changes affect only their five dependent metrics. Unknown values remain unavailable.
• Readable source-free offline HTML, deterministic JSON and safe CSV; private operational errors, no uploads or overwrites. Imported reports/hashes are explicitly unauthenticated declarations, not proof.
• Repin-aware source-unit checks for the existing ten-metric frozen-work gallery, without a duplicate metric-ID registry. Historical Grin diagnostics are not assigned invented unit fields.
• Minimum 44-pixel navigation/export targets in the two-work gallery, retained focus/reduced-motion styles; this is not full accessibility acceptance.
• Product-first README, comparison guide, existing compact queue and proportional changelog. Version 0.4.0a1; no registry publication.

Final exact-head verification

Head: 8f4194fdf75be9b1f5de276b6400feff508cd9f9
Base: 5c03ee428df6cc36ebdeff2443d35ac9f7e72dca
Scope: 3 commits, 14 files, +1421/-79 lines. Existing scrip-compare-003-local-cli reused; no new branch.

• 23/23 applicable pull-request workflow runs: SUCCESS on this exact head.
• 818 canonical Python tests: PASS, including 43 new tests. Comparator regressions use real build_local_analysis(), not analyzer stubs.
• Reader package CI: complete wheel build/install and both installed smoke paths outside the checkout PASS. Actual texts -> 29-row comparison -> HTML/JSON/CSV, deterministic exports, create-only destinations and invalid-report rejection PASS.
• Wheel produced by CI: scriptorium_reader-0.4.0a1-py3-none-any.whl, 370400 bytes, SHA-256 35aaa7644bf1a0fb09f51fde441c587eebd6693e5783fca5ede756d63a2c04f5 (build evidence, not a registry release).
• Pages CI: actual canonical site and deterministic rebuild PASS. Deployment job is SKIPPED, not a deployment success.
• Pinned pylem CI and the existing frozen/provenance workflows PASS; no morphology substitution or parity claim.
• Downloaded actual Pages artifact 12407216390, verified SHA-256 ce6c1b062c10599d8a04e0d9e4c69e00764bf3a6fd4c5c3e97164f87c5e54931. Checked all 30 internal links/fragments across seven HTML pages (homepage + six routes), ten shared metrics, twenty CSV measurements, seven charts, exact JSON/CSV values/units/source hashes/deltas and touch-target CSS: PASS.
• Local actual-source subset: 43 tests PASS. Full repository/wheel evidence above is hosted CI, not that limited local closure.

The initial CI run passed 809 tests but exposed a legacy hardcoded 0.2.0a1 smoke expectation. Fixed by requiring project, installed metadata and exact CLI version agreement, with -I isolation; no existing tests or publication protections were weakened.

Reconciled backlog and boundaries

Earlier #278 was freshly reviewed and merged as the base above; native GitHub closed #277. Recovered distinct outbox debts as #287 (POS benchmark join), #289 (EPUB ingestion) and #290 (batch analysis) after open/closed deduplication. The retained private comparator archive is marked RECONCILED_TO_GITHUB; no private storage IDs or manuscript text were published.

Browser #280, readiness #281, AOT #282, remaining accessibility #283, licensing #284 and deployment #285 remain open. EPUB/batch prototype bytes were not recovered here and are not represented as implemented. Current scientific source artifacts and tolerance rules are unchanged; M2 remains 0/5.

Local browser HTTP navigation returned ERR_BLOCKED_BY_ADMINISTRATOR. VISUAL_NOT_VERIFIED; structural checks do not establish visual/live acceptance. The Pages activation interlock is unchanged. No scheduler, access/administration or visibility changes.

PR_CREATED / BUILT / CI_PASS; NOT_MERGED / NOT_DEPLOYED / NOT_LIVE_VERIFIED. Newly authored implementation requires later fresh exact-head acceptance. It is not self-approved or merged in its authoring session.

[andy-zen-dev]

Exact-head technical review: 8f4194fdf75be9b1f5de276b6400feff508cd9f9 on master 5c03ee428df6cc36ebdeff2443d35ac9f7e72dca.

Inspected the 14-file patch, strict JSON/Decimal validator, CLI dispatch, source-unit safeguards, installed-wheel smoke and privacy/error/export boundaries. No new blocking code finding in this review. All 23 applicable GitHub Actions runs are completed/successful on the exact head (Reader installed wheel, Pages build, pinned morphology and frozen/provenance checks). There are no unresolved inline threads. This is not FantLab parity: M2 remains 0/5. Pages deployment is skipped; LIVE_VERIFIED is not claimed.

This is a COMMENT review, not an invented independent approval of my own PR. A later permitted merge still requires native repository gates and fresh SHA verification. The previous safety denial on the identical merge operation has not been bypassed.