Content-aware classification: analyze document contents, richer prompts, Fable 5 - #9
Merged
Merged
Conversation
A connection failure in propose_groups previously crashed the whole run after all per-file analysis had completed, discarding the work. Fall back to a single 'All Files' group and emit a GroupingFailed event so the warning is visible, mirroring how per-file analysis failures are already tolerated.
The grouping model previously never saw document contents — text files and PDFs were described by filename alone, so distinct legal cases, trips, or subjects all collapsed into one bucket. - Extract text from PDFs (pdf-extract, panic-isolated) and plain-text formats, sending an 8KB excerpt to a new describe_text provider method (default impl keeps existing AiProvider impls compiling) - Describe prompts now demand distinguishing detail: people count, selfie/posed, pets and species for photos; document type, parties, and case identifiers for documents; 5-8 specific-first tags - Grouping prompt prefers specific groups over broad catch-alls, replacing the 'fewer, larger groups' rule that produced one-big-bucket results - Default model: claude-sonnet-4-20250514 -> claude-fable-5 - Analysis cache versioned (.v2.json) to invalidate shallow cached descriptions written under the old prompts
🔒 Upgrade RequiredPrivate repository analysis requires Pro or Enterprise. Upgrade: https://ecc.tools/pricing?plan=pro ECC Tools keeps the core app open, and puts private repos, team features, and enterprise controls behind paid tiers. |
3 tasks
claude-fable-5 has been suspended; fall back to claude-opus-4-8 as the default analysis model.
🔒 Upgrade RequiredPrivate repository analysis requires Pro or Enterprise. Upgrade: https://ecc.tools/pricing?plan=pro ECC Tools keeps the core app open, and puts private repos, team features, and enterprise controls behind paid tiers. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Addresses mediocre classification results: photos of the same person all landed in one bucket regardless of companions/pets/setting, and all legal documents were dumped into a single folder regardless of case.
Root causes fixed
"document file: x.pdf"), so contents could not influence groupingChanges
pdf-extract, panic-isolated on a blocking thread) and plain-text formats (txt/md/csv/json/xml/html/yaml/toml); send an 8KB excerpt via newAiProvider::describe_text(default impl preserves backward compatibility). doc/docx/rtf still fall back to filenameclaude-fable-5.v2.json— existing shallow descriptions are invalidated and will be re-analyzed on next runTest plan