Repository navigation
Chunk Quality v3.4: retrieval simulator, coverage matrix, report export - #12
Merged
Merged
Conversation
Closes the loop from diagnosis to action. All client-side, no deps, nothing leaves the browser. - BM25 (k1=1.2, b=0.75) over sections, heading indexed with text and overlap included — the same doc shape the JSONL export feeds a RAG pipeline - Test retrieval: one question -> top-5 ranked sections with scores, bars, and matched-term counts; the winner shows its flag chips and a stands-alone verdict - Coverage check: one query per line -> matrix of query / status / best section / score / terms. covered = winner matches >= 60% of query terms; weak = partial; gap = nothing matches. CSV copy included - Copy optimization report: markdown deliverable with score/grade, top fixes, flags grouped by severity with specific edits, the coverage table, and a content-gaps list Verified with Playwright on the Hayes Barton Place page: staffing query lands on Chunk 2 with a "wins but is flagged" warning, "dog boarding rates" reads as a gap, pricing reads as weak, and both exports copy to the clipboard intact. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This branch was successfully deployed
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Stacked on #11 (v3.3), which stacks on #10 (v3.2). Merge order: #10 → #11 → #12, retargeting each to main as its base lands.
Why
v3.3 diagnoses; this closes the loop to action. Test what searchers actually ask, see which section wins and whether it stands alone, find content gaps, and export the whole thing as a markdown report an SEO can hand to a client. All client-side, no dependencies, nothing leaves the browser — and still zero AI.
What's in the panel
Test retrieval — type a question, get the top 5 sections ranked by BM25 (k1=1.2, b=0.75; headings indexed with section text, overlap included — the same doc shape the JSONL export feeds a real RAG pipeline). The winner is highlighted with its flag chips and a stands-alone verdict: on the Hayes page, "is staff available 24/7" wins with "Chunk 2 · section 1 wins but is flagged (Dangling reference) — it may not stand alone when retrieved."
Coverage check — paste queries one per line (GSC, keyword tools, an Ontologizer fan-out) and get a matrix: query → status → best section → score → terms matched. Status is explainable: covered = winner matches ≥60% of query terms, weak = partial match, gap = nothing matches. Copy CSV button for the matrix.
Copy optimization report — one-click markdown deliverable: URL, Chunkability score/grade, top fixes with point values, every flag grouped Fix first / Then improve / Worth knowing with its specific edit, the coverage table, and a content-gaps list. JSON stays for devs; this is the SEO-facing artifact.
Verified (Playwright, real clipboard)
On the Hayes Barton Place page: staffing query ranks Chunk 2 first with the flagged-winner warning; "dog boarding rates" correctly reads as a gap; "what does assisted living cost per month" correctly reads as weak (2/7 terms — the page genuinely doesn't answer pricing); CSV and report both copy intact (report: score 51/F, 4 top-fix/flag sections, gaps listed).
This completes the three-phase Chunk Quality plan (
tasks/todo.md). Phase 4 items (template query generator,?queries=deep-link from Ontologizer, compare mode, README rubric) remain optional backlog.🤖 Generated with Claude Code