The first vision pass OCR'd every large image. Because the "transcribe the table" prompt makes qwen render any image as a markdown table, it turned 60 product UI screenshots into junk "tables" in the corpus. In reality the only image-only data here is the release-schedule matrix — the upgrade-compatibility matrices are already native HTML tables (no OCR needed).
Fix — allowlist instead of sweep
BundleSpec.ocr (default False): the runner passes the OCR-enabling session only for opted-in bundles. Set ocr=Trueonly on morpheus_release_schedule.
Purged the 60 screenshot cache entries (kept the release-schedule one). The next --force refresh regenerates every other page's .mdwithout the bogus ## Image transcriptions section — cleaning the deployed corpus + index.
Docs (vision.py / refresh.yml) updated to the allowlist model.
Adds scripts/vision_report.py — the self-contained image-vs-transcription inspection report.
Verified
Release schedule still transcribes (cache hit); cache back to 1 entry; other bundles pass no session so their images are left alone. After merge, the refresh will strip the 60 screenshot transcriptions corpus-wide and redeploy.
## Problem
The first vision pass OCR'd every large image. Because the "transcribe the table" prompt makes qwen render **any** image as a markdown table, it turned **60 product UI screenshots into junk "tables"** in the corpus. In reality the only image-only *data* here is the **release-schedule matrix** — the upgrade-compatibility matrices are already native HTML tables (no OCR needed).
## Fix — allowlist instead of sweep
- **`BundleSpec.ocr`** (default `False`): the runner passes the OCR-enabling session only for opted-in bundles. Set `ocr=True` **only** on `morpheus_release_schedule`.
- **Purged the 60 screenshot cache entries** (kept the release-schedule one). The next `--force` refresh regenerates every other page's `.md` **without** the bogus `## Image transcriptions` section — cleaning the deployed corpus + index.
- Docs (vision.py / refresh.yml) updated to the allowlist model.
- Adds **`scripts/vision_report.py`** — the self-contained image-vs-transcription inspection report.
## Verified
Release schedule still transcribes (cache hit); cache back to 1 entry; other bundles pass no session so their images are left alone. After merge, the refresh will strip the 60 screenshot transcriptions corpus-wide and redeploy.
The first vision pass OCR'd every large image and, because the "transcribe
the table" prompt makes qwen render ANY image as a markdown table, it turned
60 product UI screenshots into junk "tables" in the corpus. In practice the
only image-only DATA in this corpus is the release-schedule matrix — the
upgrade-compatibility matrices etc. are already native HTML tables, no OCR
needed.
- Add BundleSpec.ocr (default False); the runner passes the OCR-enabling
session only for bundles that opt in. Set ocr=True only on
morpheus_release_schedule.
- Purge the 60 screenshot cache entries (keep the release-schedule one). The
next --force refresh regenerates every other page's .md WITHOUT the bogus
"## Image transcriptions" section, cleaning the deployed corpus + index.
- Update vision.py / refresh.yml docs to the allowlist model.
Also add scripts/vision_report.py — a self-contained HTML report pairing each
OCR'd source image with its transcription (uncertain-first) for accuracy
inspection. Regenerate anytime from corpus/.vision-cache/.
Verified: release schedule still transcribes (cache hit); cache is back to 1
entry; other bundles now pass no session so their images are left alone.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01LFowQzJu7k97QLCRDSAeh1
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Problem
The first vision pass OCR'd every large image. Because the "transcribe the table" prompt makes qwen render any image as a markdown table, it turned 60 product UI screenshots into junk "tables" in the corpus. In reality the only image-only data here is the release-schedule matrix — the upgrade-compatibility matrices are already native HTML tables (no OCR needed).
Fix — allowlist instead of sweep
BundleSpec.ocr(defaultFalse): the runner passes the OCR-enabling session only for opted-in bundles. Setocr=Trueonly onmorpheus_release_schedule.--forcerefresh regenerates every other page's.mdwithout the bogus## Image transcriptionssection — cleaning the deployed corpus + index.scripts/vision_report.py— the self-contained image-vs-transcription inspection report.Verified
Release schedule still transcribes (cache hit); cache back to 1 entry; other bundles pass no session so their images are left alone. After merge, the refresh will strip the 60 screenshot transcriptions corpus-wide and redeploy.
b31121d90cto1286cb7fdd