Blame for source/VALIDATION.md

## 2026-11-10 — Benchmark scope, shared parameters, denser residues and Atlas coverage

- Benchmark UI, downloadable metrics and regeneration scripts now retain only
  the 500-pair **Swiss-Prot time-cutoff** dataset. Compared every retained row to
  the before-edit snapshot: all 63 scores/sample counts and all scaling values
  are unchanged. Keyword recovery still scores 476 valid-reference pairs.
- The joint overview animates text and protein parameter sharing in sequence,
  highlighting corresponding encoder/decoder self-attention rows. Both groups
  remain inspectable with direct controls. Rewind reproduces the same SVG state.
- Protein output now shows 72 exact source residues in 26×28 tiles, three numbered
  desktop rows and a compact 12-column phone layout. Complete original outputs
  remain available in the archive.
- Exported the existing full-corpus coverage audit: 151,479 exact Swiss-Prot
  2024_01 matches among 51,321,839 Atlas representatives (0.295155%), with
  51,170,360 beyond-reference sequences. Source audit checksum is preserved.
  The display distinguishes reference overlap from model-generated hypotheses.
- Homepage map starts on reference overlap; the new descriptions layer reveals
  beyond-reference points using the original coordinates and reference flags.
  Community colors, landmarks, map navigation and full ID lookup remain intact.
- Production build passed; **46 Python tests passed**. Focused browser checks
  passed for all nine benchmark metrics, sharing controls, 72 residue values,
  coverage counts, map reveal, pause/resume, selection while paused and reduced
  motion. Actual pointer clicks on the sticky motion control preserve scroll
  position; Playwright's automatic control scrolling is avoided in this check.
- Whole-site Chromium regression passed at 1440, 1280, 1024, 820, 768, 390 and
  320 px, including all original research controls. Targeted 1280×720 checks
  confirmed diagram and navigation containment. No uncaught page errors, HTTP
  failures or horizontal overflow; no inference jobs submitted.
- Current artifacts: `artifacts/refinements-20261110/`, including the sharing /
  residue / Atlas recording and desktop/mobile screenshots. Source backups are
  in `before/`. This work is local; Hugging Face has not been updated.

## 2026-11-10 — Scroll-driven Dual Learning story

This replaces the earlier timed architecture tour described in the historical
refresh notes below. The current review artifacts are in
`artifacts/scroll-story-20261110/`.

- Read CellPack's `ScrollScene.vue`, `engine.js`, `story.js`, `pipeline.js`,
  `scale.js` and `scenes.css` at revision
  `4ef3d7867d4ee1fd09112bee09e9480682b10a58`.
- Implemented one progress-driven React/SVG story with six beats, staggered
  input/state/output objects, focus changes, a final overview of both tasks,
  independent projections and within-modality self-attention sharing.
- Text inputs/output stay blue; protein inputs/output stay orange/red. Hidden
  states are unlabelled vector tiles; residual connections and layer norms are
  omitted from the representative blocks. Complete archive values remain below
  the scene. Added bundled IBM Plex Mono technical labels alongside the existing
  Space Grotesk and DM Sans fonts; retained the smooth infinity identity.
- Desktop offers a sticky stage, chapter buttons, keyboard range input, local
  pause, global motion control and an optional reading view. Widths ≤1,100 px,
  viewports ≤640 px high and reduced-motion preferences use six static reading
  steps with purpose-built, readable diagrams.
- Production TypeScript/Vite build passed. Existing Python suite scoped to
  `tests/`: **45 passed**. An initial unscoped pytest invocation discovered
  duplicate historical tests in artifact backups; the documented scoped command
  excludes those snapshots.
- Chromium regression passed at **1440, 1280, 1024, 820, 768, 390 and 320 px**:
  all six beats, intermediate residue emission, exact SVG reproduction after
  scrolling back, no idle/offscreen progress, local/global pause, chapter and
  keyboard navigation, reading-view switching, source-matched full archive,
  persisted motion preferences and runtime reduced-motion behavior.
- Existing 29 annotation keys, eight Atlas cards, standalone Atlas structure,
  DE2 backbone playback/scrubbing, PETase case/residue controls, original prompt
  reuse and benchmark filters passed. **No uncaught page errors, HTTP failures
  or horizontal overflow; no inference jobs submitted.**
- The final SVG polish gives moving hidden-state packets a clear channel through
  the projection, separate from its label and equation. Targeted checks passed
  at 1280×720 and 1920×1080, including SVG/controls containment, rewind after the
  layout change, short-viewport reading fallback, runtime reduced-motion changes
  and a simulated hidden-tab visibility event. The short-screen review caught
  and corrected a vertical overlap with chapter navigation.
- `scroll-story-preview.webm` records real scrolling through every stage and
  back to an intermediate residue output. Current desktop/mobile screenshots,
  `browser-report.json` and `final-layout-report.json` accompany it.
- Backend/model code, scientific exports and inference APIs were not modified.
  This work is local; no Hugging Face publication was performed.

## 2026-11-10 — Orange/blue presentation and animation refresh

- Reviewed public CellPack source at `4ef3d7867d4ee1fd09112bee09e9480682b10a58`.
  Space Grotesk / DM Sans are bundled locally; language stays blue and proteins
  orange/red. Redesigned the hero, architecture, section hierarchy and evidence
  surfaces, without changing scientific datasets or inference implementations.
- Added a continuous red/blue infinity identity, SVG/PNG mark/avatar and favicon,
  configuration-matched T5 block diagrams, source-matched residue tile playback,
  selectable hero directions and signal animations,
  first-entry charts, and a DE2 sequence-to-backbone interpolation using the actual
  279 Cα coordinates. No folding simulation or measured attention is implied.
- Motion control persists across reload and navigation to the standalone Atlas.
  Reduced motion reveals complete outputs; offscreen and hidden-tab motion stops.
- Production TypeScript/Vite build passed. The Docker frontend stage now copies
  the same model config used by the architecture diagram. Existing Python suite:
  **45 passed**.
- Chromium regression passed: hero direction controls, animation freeze/resume,
  persisted preferences, guided tours, recorded playback, backbone scrub/replay,
  all 29 annotation keys present, all 8 Atlas cards present, standalone Atlas
  structure, PETase case switching and keyboard residue selection, original
  prompt reuse, and benchmark task/metric/set selection.
- Layout checks passed at **1440, 1024, 820, 768, 390 and 320 px**, including all
  homepage sections, mobile navigation and runtime reduced-motion changes.
  **No uncaught page errors, failed HTTP resources or horizontal overflow.**
- Validation ran with inference disabled; **no inference jobs submitted**. API
  behavior is covered by the existing tests, not a new live generation run.
- Original source backups, screenshots and browser report:
  `artifacts/motion-refresh-20261110/`. These remain excluded from Docker.
- This refresh is prepared locally; no Hugging Face publication was performed.

## 2026-11-10 — Nine active PETase designs

This update supersedes the four-design wet-lab snapshot in the historical notes
below. It uses the newly supplied table and FASTAs, with no inferred assay date.

- Displayed cases: DE1–DE9 plus IsPETase. Default: DE2. Screened test outcome:
  9 active / 12 submitted; the three inactive candidates are not individual cases.
- Exact supplied concentration values and CSV precision verified. DE2 total:
  1,514.1531 μM vs IsPETase 1,074.8149 μM (1.41×, 40.9% higher). This is a
  product concentration comparison, not a kinetic or significance claim.
- All nine sequences match original LoRA generation records and prompts.
- ESMFold job `ouroboros-petase-nine-20261110-r1-14531184` succeeded in
  `ailab-dnacoding`, single H200 GPU, idle, 9/9 structures, no restarts.
  PDB sequence, residue numbering, CA confidence and SHA-256 checks passed.
- Production TypeScript/Vite build passed; `.venv/bin/python -m pytest -q tests`:
  **45 passed**. Tests are scoped to `tests/` to exclude archived release snapshots.
- Chromium checked all 10 structures, all assay metric modes, chart-to-case
  navigation, sequence keyboard selection, representation and confidence modes,
  sequence copying, FASTA download and original-prompt reuse in the playground.
- Desktop, tablet and phone widths (1440, 768, 390, 320 px), reduced motion,
  and the unchanged eight-case Atlas preview passed; no page errors or overflow.
- Desktop and mobile screenshots were inspected; molecular framing and compact
  candidate labels were refined during review. No manuscript figures or PDFs.
- Release, browser checks and source snapshots: `artifacts/petase-nine-20261110/`.
  Research inputs, job logs, manifest and PDBs are in the sibling
  `Ouroboros_Showcase_Research/wetlab_20261110/` directory.

## 2026-11-10 — HF publication revision

The manuscript figure gallery, original SVG/PDF downloads and fixed figure numbers
were removed before publication. No manuscript images or PDFs are included in
`public/` or `dist/`. The eight independently rendered protein cases remain.
The homepage typography, molecular stage, map surface and case cards were refined.
The production build and 45 Python tests passed. Browser checks covered desktop,
phone, tablet, all eight structures, history/deep links and real full-corpus lookup.
The release manifest and publication status are under `artifacts/hf-refresh-20261110/`.

## 2026-11-10 — Paper Atlas and homepage refresh

- Production build passed; 45 Python tests passed, including the real Parquet reader.
- Atlas point counts: 12,000 on `/atlas`, 4,000 in the homepage preview. The 51,321,839-record lookup scope is unchanged.
- Precompressed transfer: 201,416 bytes for the map and 70,108 bytes for the preview, versus 1,491,533 bytes for the previous map.
- All eight paper cases retain sequence-verified PDB files, descriptions, Pfam evidence, source hashes and original map coordinates. Both display subsets retain all 34 communities.
- Browser checks passed at desktop, 390 px phone and 768 px tablet widths. All eight structures, separated map landmarks, selection/history/deep links, figure dialogs and reduced motion were exercised without page errors.
- Additional checks used actual point clicks, real full-corpus lookup outside the display sample, two-finger touch input, enlarged paper figures, and a simulated lookup outage with working archived case links.
- Full lookup used the existing published dataset revision `391ef110018106573573bf049aebbc21f28e35b6`. Live inference was disabled for preview; no inference jobs were submitted.
- Browser logs and screenshots: `artifacts/atlas-refresh-20261110/`. Original local files were backed up in its `before/` directory. No Hugging Face deployment was performed.

# Review build — 2026-10-09

The implementation follows the approved scope: original large-v5, dual learning first, the experimentally tested de2 PETase case, and an HF Docker Space with online generation.

## Passed

- Production TypeScript/Vite build, including local fonts and bundled 3Dmol.
- Nine API/asset tests: input validation, queue bounds, serial worker reuse, failed generation handling, exact sequence–PDB correspondence, confidence and source checksums, and both benchmark splits.
- Chromium desktop (1440 × 1000) and mobile (390 × 844) rendering; no horizontal overflow or uncaught page errors.
- Both WebGL protein viewers; de2 and the selected IsPETase reference; cartoon/surface/sticks; confidence colors; residue selection and keyboard navigation; PDB download.
- Dual-learning direction switches, benchmark task/metric/set switches, model-size selection, mobile navigation.
- original de2 LoRA prompt → playground; archived examples in both directions; generated sequence → annotation input.
- Actual original checkpoint loaded from the research workspace. Both text→protein and protein→text jobs completed with a 64-token output cap.
- End-to-end browser submission to the live protein→text API completed and displayed a LIVE result. These short tests validate integration, not output quality or biological function.

## Deployment status

The deployment targets REXWind/ouroboros on its existing CPU Basic hardware, using the Docker SDK. The previous Gradio revision is backed up locally and retained in HF history. A prior local Docker build was blocked by a Docker Hub timeout; the frontend and Python runtime passed local checks. Hosted build and live-generation verification are recorded separately after upload.

Local live preview: http://localhost:7861 (while the review service is running).

## Evidence boundaries

The active case is de2 (279 aa), selected as the highest reported TPA+MHET concentration among four measured designs in the supplied 2026-09-27 workbook. The visible comparison focuses on de2 and IsPETase. The complete seven-sample source table, including LCC and BhrPETase, remains downloadable. Eight other submitted designs have no numerical values and are not treated as zero. Assay conditions and replicate information are awaiting confirmation. The original LoRA generation prompt and exact predicted sequence were recovered from lora_petase_3epoch, prediction index 2890.

ESMFold job ouroboros-wetlab-de2-1009b succeeded on 2026-10-09. All four structures exactly match supplied FASTAs. de2 CA-mean pLDDT: 93.46258; mean PAE: 4.79198 Å. The runner's unmasked atom37 average pLDDT (85.4938) includes absent atom slots; the UI consistently reports the mean over CA residues, verified against PDB B factors. Both calculation definitions remain recorded in the source manifest and exported case metadata.

## Prompt builder follow-up

Production build and browser checks passed for all six original-demo presets; similarity/function editing; adding/removing cofactor; exact original-order prompt assembly; random preset selection; switching to protein→text and loading saved outputs; 390 px mobile layout. No uncaught page errors.

## Wet-lab browser verification

All seven assay rows, the TPA/MHET/combined metric switches, exact de2 readouts and CSV download passed. The selected de2/IsPETase structure tabs, residue selection, confidence coloring, and the recovered LoRA prompt → playground passed. Desktop and 390 px mobile rendering have no horizontal overflow or uncaught page errors. Default structure framing was enlarged for the compact de2 fold.

## Dataset and Core Idea update

Nine tests passed. A dedicated browser run validated all 29 key tiles, top/other filtering, search including the rare interaction key (2 records), empty-search feedback, coverage values, the guided tour with manual pause, changing input/target roles, and the IsPETase-only comparison. Desktop and 390 px mobile layouts have no horizontal overflow or page errors. Dataset membership (full dataset versus training set) remains unconfirmed, so the inferred 545,334-record denominator is documented in metadata and is not promoted as a confirmed training-set size.

## Compact statistics and motion polish

Production build passed. The dataset section now defaults to five rows with an expandable, searchable 29-key list; its desktop section height is approximately 550 px including section spacing. Dedicated Chromium checks passed for both directions' progressive reveal, pause/resume, immediate completion, replay, stable action placement, complete downloads during partial playback, direction-switch cleanup, and reduced-motion behavior. The completed live-job UI was exercised with a mocked API; the unchanged backend's earlier real-model checks remain recorded above. Desktop and 390 px mobile screenshots were inspected, with no horizontal overflow or uncaught page errors. Output playback is explicitly distinguished from live token streaming.

## Final deployment revision

The Core Idea exclusively owns recorded typewriter playback (large-v5 HSP20 example hard1-0151). Playground examples only populate inputs; outputs appear directly after real inference jobs complete. Browser verification covered both Core Idea directions, pause/skip, reduced motion, input-only presets and both completed-job UI paths (mocked API), with no page errors.

The supplied typedata.zip was extracted to the same OSS directory. All 545,334 records were checked against the CSV: all 29 key counts match exactly. Every case is a verbatim annotation from type_with_data_clean.json, with a record index, UniProt ID and source checksum. All 29 key/case selections, search, scrolling and desktop/mobile overflow checks passed. No external fallback examples are used. The full dataset is retained in OSS; only 29 small illustrative cases are shipped.

The remote HF model checkpoint and validated local original large-v5 have identical SHA256 f11f6295794d027e03d6844d5b6d685641ca04d3558536416663ffddaa0b4067. Nine API/asset tests and the production build passed.