Curating Search Result Dumps

Given a raw search-results dump, produce a scored, deduplicated shortlist using this rubric per result:

Quality Score (0-10):

• Source type: Marketplace/official site (+3), dedicated blog/creator (+2), aggregator with real content (+1), pure image reshare (0)
• Access clarity: Direct download/clear purchase flow (+3), requires signup/membership (+1), dead-end or unclear (0)
• Specificity match: Exactly matches request (+2), close variant (+1), tangential (0)
• Signal of quality: Reviews/ratings present (+2), no signal (0)

Score ≥7 → "Recommended" | 4-6 → "Usable with caveats" | <4 → "Low priority/exclude"

RECOMMENDED (score 7-10):
1. [Name] — [Source] — [Free/Paid $X] — Score: 8/10
   Link: [URL]
   Why: [1-line justification citing rubric]

USABLE WITH CAVEATS (score 4-6):
...

EXCLUDED (score <4 or irrelevant):
- [brief reason, grouped by cause]

Progress:

• Step 1: Split input into distinct sections (e.g., "Visual matches" vs "Web results" vs any text blob) — these sections often have different reliability baselines
• Step 2: Extract every candidate result with its stated source, title, and any metadata (price, rating, format)
• Step 3: Deduplicate — same underlying content often appears under multiple pin captions/URLs. Match on: same domain + similar title, or identical image description repeated verbatim
• Step 4: Score each unique survivor using the Quality Score rubric above
• Step 5: Flag access friction explicitly — paid-only, membership-gated, print-on-demand, "link in bio" dead ends
• Step 6: Identify and separately call out noise (results clearly unrelated to the query — tracking contamination from ads, unrelated news, SERP glitches)
• Step 7: If zero or near-zero results score ≥4, say so explicitly and suggest refined search terms rather than padding the list with weak results
• Step 8: Present three-tier output (Recommended / Usable with caveats / Excluded), never a flat undifferentiated list

Example 1: Rich results, mixed quality (sewing pattern search)

Input: Raw SERP dump mixing Pinterest "visual matches," Etsy listings, DeviantArt tutorials, YouTube videos, Amazon product, and two unrelated government/news links, for "bunny plushie sewing pattern."

Output:

RECOMMENDED (7-10):
1. FREE BUNNY PATTERN N TUTORIAL — DeviantArt (sewcuteplushies) — Free — Score: 8/10
   Why: Dedicated creator (+2), direct PDF/tutorial access (+3), exact match (+2), no rating signal (0) = 7... 
   rounded to 8 for explicit "free download" language reducing access ambiguity
2. Simplicity S9306 Plush Bears & Bunnies — Amazon — Paid — Score: 9/10
   Why: Official marketplace (+3), clear purchase flow (+3), exact match (+2), 4.5★/13 reviews (+2) = 10, 
   capped slightly for being a commercial multi-pattern bundle not bunny-specific

USABLE WITH CAVEATS (4-6):
1. Linen Bunny 22cm — Etsy — Paid — Score: 6/10
   Why: Marketplace (+3), standard Etsy checkout (+2 partial), exact match (+2) = good source but price/reviews not visible in dump — verify before recommending
2. Mini Bunny Doll — YouTube — Free — Score: 5/10
   Why: Video format lacks a static pattern file by default (+1 access), creator-hosted (+2), exact match (+2) — 
   flag: confirm description has downloadable template, not just visual demo

EXCLUDED:
- All "Pinterest visual match" entries with captions like "Bunny doll pattern" and no distinguishable source page — 
  these are reshares of the same 1-2 underlying images, high duplication, no verifiable download path (9 instances collapsed)
- State of Michigan invasive species / election articles — SERP contamination, unrelated to query
- Yandex/Google "Картинки" reverse-image-search landing pages — these are search tools, not content sources

NOTE: 9 Pinterest duplicates collapsed into the exclusion above represent the same 2-3 source images recirculated.

Example 2: Sparse, low-quality results (niche/failed search)

Input: Search for "vintage 1952 Butterick pattern #4821 digital scan" returns only 3 results: a forum post with a broken image link, a Pinterest pin captioned "old pattern" with no source, and an eBay listing for a different pattern number.

Output:

RECOMMENDED (7-10):
None found.

USABLE WITH CAVEATS (4-6):
1. eBay listing (different pattern #) — Score: 4/10
   Why: Real marketplace (+3), clear purchase (+3), but wrong pattern number (0 specificity) = flagged only 
   as a lead to contact seller, who may have the correct one or know a source

EXCLUDED:
- Forum post with broken image link — dead end, no recoverable content
- Pinterest pin "old pattern" — no source, unverifiable, likely different item entirely

VERDICT: No usable direct match. Recommend refining search to: pattern number + "scan" + specific 
marketplace names (Etsy "vintage patterns" category, EvaDress reproductions, or Facebook vintage sewing groups), 
or searching the pattern number alone without "digital" to find physical copies that could be scanned on request.