Skip to content

PLAN.md

Latest commit

 

History

History
33 lines (23 loc) · 2.56 KB

PLAN.md

File metadata and controls

33 lines (23 loc) · 2.56 KB
Raw

Plan

Status

Built from the first snapshot: 907,093 cards, 308 cards with their own pop report, 186 sets. Not announced yet. Tweet drafts live in TWEETS.md, which is gitignored.

Before posting

  1. Set LAUNCH_DATE in .github/workflows/pages.yml to the day you post, so the refresh schedule counts from launch.
  2. Add the API key secret so the workflow can refresh (run it yourself; the value never prints): 
    ~/.claude/skills/onepassword/scripts/op-sa run --env-file ~/.config/op/opensea.env -- sh -c 'printf %s "$OPENSEA_API_KEY" | gh secret set OPENSEA_API_KEY -R ryanio/card-census'
  3. Run gh workflow run pages.yml -R ryanio/card-census about an hour before posting.
  4. Retake the tweet images and check the numbers: node scripts/shots.mjs (needs Chrome). It writes screenshots and every share card to tweet-images/ (gitignored) and prints the headline numbers on the live page, so the copy in TWEETS.md can be checked against them.

Known limits

  • A facet value list stops at 100 entries, so only cards in the overall top 100 or a franchise's top 30 get their own report. Everything is still in the totals.
  • Grades are compared on each grader's own scale. A 10 means 10 or higher, including 10.5 (CGC Pristine, BGS Black Label). BGS calls 9.5 Gem Mint, so its share of 10s reads low.
  • Values are OpenSea's estimated value, not sales.
  • The cover map shows real photos, but not every card: a small tile stands in for several hundred cards, and a territory cycles through the photos fetched for it (about one per 450 copies for a big card, about one per 750 for a franchise's long tail, up to 800). The most valuable cards in each territory are drawn across 2x2 to 4x4 cells. Cards with no franchise listed aren't on the map; the caption says how many.
  • Platform counts come from the facets endpoint with collection_slugs. The platforms are the collections seen in the crawled photos, so a platform with no photo in any sample would be missed (1,698 cards come from other collections today).
  • The same card can be listed more than once (two issuers, or the same slab in two places). The vault and each card's copies drop repeats by name and value.

Later

  • A set lookup next to the card lookup, using the per-set facets already in the crawl.
  • Pop history: keep a dated snapshot of each card's grade counts to show what got tokenized this week.

Data in the repo

Each fresh crawl writes CSVs to data/ and appends to data/history/. The workflow commits them once per UTC day, from that day's first run. Runs drop to weekly at day 30, so the data commits do too.