Keep persistent Atlas source status readable by screen readers

docs/reports/atlas-persistent-status-20261026/FOLLOWUP.json

@@ -0,0 +1,50 @@
+{
+  "authored_after": "38ba59e77e6cadd78ad39beac8284a655c2a74ff",
+  "original_report_head": "a1bf4b04b5372d5aac1c623d034fe8c44730a2df",
+  "followup": "Include byte-identical six-test module in existing CI discovery and publish exact current metadata/registry after offline rendering",
+  "original_native_failure": {
+    "workflow_run": 40332209208,
+    "job": 121604256581,
+    "attempt": 1,
+    "tested_head": "562252d1e90dcdcac9966e04458803e4836e56ea",
+    "tests": 145,
+    "errors": 2
+  },
+  "local_reproduction": {
+    "tests": 2,
+    "errors": 2
+  },
+  "after_publication_pipeline": {
+    "normal_tests": 151,
+    "optimized_tests": 151,
+    "errors": 0,
+    "normal_log_sha256": "e2d87acda151efd7502dbc4500b658d722d710355bf769a13ee7936d7935d2e9",
+    "optimized_log_sha256": "79ab6c3c7782c27a557b506f7260668f3e9af610b8de444a4525416e89a54f4a"
+  },
+  "generated_metadata_sha256": "8a33e435685b3862f334006d6626bd05279664c25d88bed8b41f0493bc50b049",
+  "registry_sha256": "5f822c4a2b512aa9293e2ebaee58abbf5f4caca4ad87362ef1e114b5d49ad817",
+  "original_archives_and_payload_unchanged": true,
+  "scientific_effect": "NONE",
+  "navigation_followup": {
+    "original_native_run": 40334461953,
+    "original_native_job": 121612197360,
+    "original_native_head": "be15875536bc9724e86250d4bd5675dcd0c0d79e",
+    "native_python": "3.11.17",
+    "local_full_python312_before": {
+      "tests": 1427,
+      "failures": 2,
+      "skipped": 24
+    },
+    "root_cause": "Two historical byte-golden renderer tests used the new current-presentation default; historical output must explicitly request readable_status=False",
+    "repair": "Three renderer call arguments in two tests, preserving original hash fixtures and assertions; no production or policy change",
+    "after_exact_module_discovery": {
+      "python311_tests": 8,
+      "python311_seconds": 24.966,
+      "python312_tests": 8,
+      "python312_seconds": 18.82,
+      "failures": 0
+    },
+    "direct_qualified_module_probe_import_error": "Harness invocation only; full discovery adds tests directory, not a native failure",
+    "full_navigation_gate": "Fresh corrected-head hosted execution remains required"
+  }
+}

docs/reports/atlas-persistent-status-20261026/README.md

@@ -0,0 +1,61 @@
+# Persistent source status: bounded accessibility repair
+
+Source-observation and preserved-viewer status messages now remain ordinary readable paragraphs inside native status regions. Orca46.1/Chromium141 can read their contents with browse, WhereAmI and SayAll. The existing paragraph classes/IDs and freshness/error writers remain intact. Readers keep the compact source panel, native **Sources and scope** disclosure, exact source links and static reading routes.
+
+Implementation: `dfe7051b13023bd627aabc02318aacb4431c4c22`, authored against Main `f8ea0b530fed1942aa23d79a8fce179a679862f8`. [SOURCE.json](SOURCE.json) records production hashes. The follow-up renames the six-test module, byte-identically, to `tests/test_current_site_persistent_status.py` so the existing hosted `test_current_site*.py` controls discover it. The original dfe/a1 execution and source identities remain unchanged. Local tests do not identify a deployed revision; separate current-composition review, integration and actual publication verification remain required.
+
+The matched observation is Main **f8ea0b530fed1942aa23d79a8fce179a679862f8** / Math- **180d6406b0a51b268a152241a70bde6a1b39a142**, observed **2026-10-26T11:41:53Z**: **49 nodes, 55 directed recorded relationships, 108 excerpts**, payload **1,751,539 bytes**, SHA-256 `71eb6b5c26c7660e93c4664d629a4d9b1fd8c37e3df45d88d3adad0c8a4bf96c`. The candidate was generated by the changed source after restoring all eleven original publication archives. Fixture pins and implementation pins have different purposes.
+
+## Same reader journeys
+
+| Journey | Before | After |
+| --- | --- | --- |
+| Actual Orca46.1/Chromium141, two statuses × browse/WhereAmI/SayAll | Contents omitted, 0/6 | Contents recorded, 6/6; repeated on generated implementation |
+| Actual Orca48.1/Chromium154 | Contents already readable | Contents readable; no new-version failure claimed |
+| Actual Orca46.1/Chromium141, JavaScript disabled | No matched baseline speech qualification | All six readings contain the initial source/legacy messages |
+| Desktop/phone × themes, no-JS, static index/node, script-free Formal, refused current payload | Existing text/links/layout retained; new paragraph-structure criteria fail | 10/10 focused cases pass |
+| Source and preserved-status geometry in the same ten cases | Recorded | Identical bounding boxes, typography and margins |
+| Previous full fifty live-page descriptors | Exact old representation | Restored byte-exactly; altered scientific text, source pin or role refused |
+
+Desktop source panel: [before](baseline/1200-light-source.png), [after](candidate/1200-light-source.png). Phone dark: [before](baseline/390-dark-source.png), [after](candidate/390-dark-source.png). Preserved-viewer notice: [before](baseline/390-dark-preserved.png), [after](candidate/390-dark-preserved.png). No-JS desktop: [before](baseline/nojs-1200-source.png), [after](candidate/nojs-1200-source.png). The screenshots were visually inspected. Literal observation text and source access remain legible; the change adds no visible settings or provenance panel.
+
+The first wrapper iteration lost an inherited desktop `78ch` paragraph measure. The matched geometry check exposed that change, and a scoped selector restores the old measure. The first unavailable-state browser probe intercepted metadata that the page never fetches; the corrected probe corrupts both current-payload attempts and verifies visible refusal with disabled Atlas controls. Both initial runs remain in custody. Baseline browser failures concern the new paragraph-structure criteria, not a claim that every existing reading journey was unusable.
+
+## Actual speech and its limits
+
+The operator is the author OpenAI/Codex invocation; this is source-exposed implementation evidence, not blind or organizationally independent review. [Speech summary](speech/summary.json), literal phase records, response identities, runtime versions and the [keyboard scenario](speech/keyboard-scenario.json) accompany this report. Raw Orca/AT-SPI/Speech Dispatcher logs, harness snapshots and null-sink audio are preserved privately. Public runtime-version text copies omit trailing whitespace; summary hashes identify the retained raw originals. Duplicate debug representations of an utterance are not counted as separate spoken messages. **No human audition** was performed.
+
+The corrected protocol uses real Orca heading/Down keys, read-only DOM caret observations to establish each target, keypad Enter for WhereAmI and keypad Add for SayAll. `20261026T123334Z-persistent-legacy-key-base` omits both texts in all six phases. The diagnostic wrapper run `123414Z-persistent-legacy-key-new`, proper generated run `130225Z-persistent-generated-key` and final generated run `131303Z-final46key` each record all six. `130802Z-final46nojs` reads the initial pinned observation and original “Loading the local source snapshot…” text; the latter remains an initial no-JS message, not evidence that the preserved viewer loaded data.
+
+Orca48.1/Chromium154 already reads these texts. The initial 48 extractor accidentally omitted SayAll's SSML records; the corrected [extractor](extract-speech.py) includes both dispatcher strings/SSML and SPEECH OUTPUT. The original partial assessment was corrected and retained. Early 46 pointer-only browse attempts did not establish the intended caret and do not qualify browse/WhereAmI. Later long-labelled harness runs hit Unix socket length/startup limits; owned IPC paths were shortened and actual AT runs serialized. These setup failures are retained, not counted as product failures or passes. The six-second modern baseline SayAll window stopped upstream of one target; its longer thirteen-second replay reads both. See individual run applicability rather than treating `completed` as sufficient qualification.
+
+Test-only official runtime pins: Orca `46.1-1ubuntu1` Ubuntu package (565,724 bytes, SHA-256 `2ea414b8aea922b1ac9a1479c1324cba6b28e48dbe57b99d08f9c76806a51c59`, LGPL-2.1+), and Playwright Chromium revision1194 / `141.0.7390.37` (182,333,649-byte ZIP, SHA-256 `5ad70cf662e7090354edbe84e1a01ac531fd217f37f5001f0d2efbe0d152b396`, Chromium's bundled licensing). Downloads came from the official Ubuntu archive and Playwright CDN over the configured proxy/TLS; no production dependency was introduced. The current stack is Orca48.1 / Chromium154.0.8037.57. Both use Debian13, Python3.13 GI, AT-SPI2.56.2, Speech Dispatcher0.12.0 and eSpeakNG1.52 with isolated Xvfb sessions and Chromium sandbox enabled. Matching the older Orca/Chromium versions does **not** reproduce Claude54's complete original Ubuntu/Python3.12 environment or close that agent's conditional qualification. Its source evidence and unaccepted/conditional roles remain attributed and unchanged.
+
+## Loading, interaction and rendering observations
+
+| Four matched contexts per side | Before | After |
+| --- | --- | --- |
+| Median Atlas/current-source readiness | 440ms (386–502) | 443ms (370–664) |
+| Median 3D activation, all49 nodes/55 edges | 333ms (309–426) | 348ms (288–552) |
+| Decoded resources before requesting3D | 1,996,147B | 1,996,461B (+314B) |
+| Median exact restoration of previous50 live pages | 501ms | 1,004ms |
+
+[Raw observations](performance.json), [summary](performance-summary.json) and the retained [measurement driver](benchmark.py) describe shared Linux x86_64, Playwright1.62, Chromium154.0.8037.57, sandbox enabled, DPR1, reduced motion and software SwiftShader. Contexts alternate side order across desktop/phone and themes; network is local HTTP without throttling. Three.js is absent from pre-activation resource lists. Browser animation-frame intervals during a short pointer orbit had median/p95 about16.7ms on both sides (150/147 samples); these are browser scheduling observations, **not GPU timings or sustained FPS certification**. Existing successful canvas-picking tests separately exercise the real renderer.
+
+These four observations establish no statistically significant speed change, physical mobile/GPU performance or larger-graph capacity. Supporting the additional exact presentation roughly doubles the measured legacy-restoration step, adding about0.5s at this dataset; this is a build/admission cost, not reader latency. The finite source-derived presentation search still refuses unmatched complete descriptors. The initial benchmark waited for a mobile renderer before scrolling its host into view; the corrected driver follows the existing successful-rendering protocol, and the initial diagnostic is retained privately.
+
+## Verification and preserved evidence
+
+The author ran all **14 prescribed local groups** after standalone generation and asset release/check. The first hosted full current-site discovery subsequently caught two publication-pointer errors among145 tests: the standalone source build had omitted the required history-publish step. The follow-up runs the prescribed publish/release/publish/check sequence, restores the current metadata pointer and refreshes all fifty exact live descriptors. It changes no publisher assertion, admission rule, payload or original archive bytes. Complete discovery, including the renamed six tests, then passes **151/151 in normal and optimized Python**. Original native/local failure evidence remains retained. The separate navigation job then exposed two historical-output tests using the new current-view default. Their three renderer calls now explicitly request `readable_status=False`; original golden hashes, fixture bytes and assertions stay unchanged. Full Python3.12 discovery before this correction recorded1427 tests/two failures/24 documented skips; corrected module discovery passes8/8 on both exact CPython3.11.17 and Python3.12. Fresh full navigation execution remains a separate release gate. The combined current data/build/history/focused suite passes **76 tests**; the original current-panel full-byte API hash stays unchanged. The public-shop Node group passes311 with its two documented fixture skips, and museum integration reruns those controls. The museum test correction preserves its two evidence live regions and a separate generated current-source region; original assertion failures are retained. The official current-site driver passes **39/39** through a retained local adapter substituting packaged Chromium154 for unavailable channel Chrome, with sandbox/assertions intact; this is local evidence, not the hosted packaged-Chrome record.
+
+Another **57/57 browser cases** pass: ten persistent-status cases, four live controls, seven guide/provenance, two metadata contrast, four full-label, sixteen spatial, ten source-refusal/fallback and four formal-evidence cases. They include actual canvas rendering/picking, dependency/source-inspector consistency, share/reload/Back/Forward, keyboard/pan/zoom/reset, reduced motion, context loss/initialization failure and 2D/list/static/no-JS fallbacks. [Regression records](regression/) and [prescribed results](prescribed-results.json) retain scope and environment.
+
+Sixteen frozen historical renderer function identities match the original golden record. All100 changed historical HTML files differ only in required asset-release query keys; [canonical-byte checks](historical-preservation.json) preserve their original content. The current mathematical payload is unchanged. The eleven original published archive ZIPs remain byte-exact in the generated acceptance fixture; no historical proof or mathematical record is regenerated by this repair. The generic full `current_section` API retains its legacy default; current page generation explicitly opts into readable paragraphs, and archived sections ignore that option.
+
+Original complete site/generator/test preimages (259 files,10,463,278 bytes, SHA-256 `b41ffe78a9098ab32336d2c42b29005725644876c4ce1c70d78e54e0b91059f0`), actual prior publication (262 files,13,329,629 bytes, SHA-256 `e978a7cb1e6cac44b94e604bf06f133e50d0dce32c6ef998137b4570555f6f7a`) and supplemental museum-test preimage were independently downloaded from restricted private custody and verified member/whole-byte/CRC **before edits**. Neutral receipts accompany this report; private locator mappings remain private. No Drive material is deleted. Source/head/final-custody and publication observations are subsequent separate delivery records, not implied by these preimage receipts.
+
+## Formal connections and next increment
+
+The four formal-source tests preserve the existing limited SIDE24 source-document/arithmetic-companion connection and original execution receipt. The selected declaration includes `UniversalLaw.Side24.pairing_sum_six`; it does not prove the mathematical coefficient claim or parent theorem. The original package remains **REJECTED (AMEND)**, its manifest **PENDING**, and accepted package-level alignment/parent coverage stays zero. No mapping, declaration, hypothesis, status authority, kernel evidence or acceptance record changes. Missing mappings stay explicit; CI success does not promote proof coverage.
+
+Next value: independently qualify these current sources on the original older environment and other AT families, then complete a human-assisted accessibility audit of mathematical navigation and explanations. NVDA/JAWS/VoiceOver, actual mobile touch/GPU, full WCAG conformance, the older View sensitivity finding and scientific acceptance remain outside this bounded evidence. Existing foundational guide/disclosures and 2D/list/static reading are preserved.

docs/reports/atlas-persistent-status-20261026/SOURCE.json

@@ -0,0 +1,39 @@
+{
+  "implementation": "dfe7051b13023bd627aabc02318aacb4431c4c22",
+  "authored_base": "f8ea0b530fed1942aa23d79a8fce179a679862f8",
+  "dataset_repositories": [
+    {
+      "commit": "f8ea0b530fed1942aa23d79a8fce179a679862f8",
+      "repository": "d6g8k5htny-coder/main",
+      "tree": "c3fff8fa74c0afb2af3ab4c5da0df7d4c170f70e"
+    },
+    {
+      "commit": "180d6406b0a51b268a152241a70bde6a1b39a142",
+      "repository": "d6g8k5htny-coder/Math-",
+      "tree": "cf3e699e9deb25a7ce1b12b92b1c2f50dd3ba82f"
+    }
+  ],
+  "observed_at": "2026-10-26T11:41:53Z",
+  "payload_sha256": "71eb6b5c26c7660e93c4664d629a4d9b1fd8c37e3df45d88d3adad0c8a4bf96c",
+  "payload_bytes": 1751539,
+  "graph_nodes": 49,
+  "graph_relationships": 55,
+  "equations": 108,
+  "source_files": {
+    "tools/current_site_build.py": "2a963a61d01db3fbb96979860e9c4af9f4b79140d00b73cc1b0d8a18b4a2d3eb",
+    "tools/current_site_history.py": "589879a59db8531807598100b76cfe02fc75b561cafef4c6c059dcf171bef115",
+    "docs/site/brand.css": "3d7c7b6e0ea5191a8bae038d1a8dc2a5cb2709731ecdcd5957ad5e8599c7a0ab",
+    "docs/site/atlas-layout.css": "e5726dc43b30610e32c4cb3e39efa771bc6569676de0b00c0341d8d3d3feb958",
+    "docs/site/dependencies.html": "df8469d8120cfae021d64d47aefaf38f56a809ddad14cce6c4f2acf575ffb64e",
+    "docs/site/current.mjs": "0cf73485540df3674ff2eb4458614d51dac7fdb7941f64db6215680fbb7a8a76",
+    "docs/site/current-model.mjs": "c9247ea9acec23172893caa03028041be784bf83f08b0209c077167a3de0f4cb",
+    "docs/site/atlas-controls.mjs": "182c5a17be3c67389633f07f50a13abb4cbbb6ebd2824eff8c107f18581ed161",
+    "docs/site/atlas-formal-evidence-data.mjs": "441babc5f605492b361822c26d59ebee4126e45b90695915195bf951c3c943ef",
+    "tools/persistent_status_browser_check.py": "c7916ef456fd5ff37d9e367f4e6b5cccc9987e05be8c1c66e463a103bf4815c7",
+    "tests/test_persistent_status.py": "cd4e9a962daef2761474d57a36e97a8d289323395378234e426ff409f8af0338",
+    "tests/test_museum_frontend.mjs": "1bc61e0202cc6c147bbdd1357d7075f239f8a1bacda405404de09ba138507c72"
+  },
+  "asset_release": "101a762efd0bb2dfd4d9c0ca3c2ec39e5744ae262ed8cb2ac7c6e7b16da2caa0",
+  "scientific_effect": "NONE",
+  "publication": "Pending separate current-composition integration and actual deployed-source verification; authored implementation does not identify the current publication."
+}