Per-game coverage: where coverage stands, the expansion runway, and the one accuracy gap.

1. Coverage is uneven across the games already ingested. Of the four games with active eBay listings, the flagship game leads and pokemon is the worst live game: 36% of its active listings match, 50% of the listings classified as single cards. That gap is per-game matcher coverage, the rule-pack work the client sketched.
2. The catalog is far ahead of ingest, which is the expansion runway. MTG is fully catalogued at 112,693 cards with zero eBay ingest yet. Yu-Gi-Oh holds 46,188 catalogued cards and currently has zero active listings. The cards are already in the database; what is missing is ingest plus a per-game parser, not the catalog.
3. One accuracy gap is durable and reproducible: the foreign-title wrong-card risk is zero on the flagship game and non-zero on every other game (one-piece 212, dragon-ball-super 105, pokemon 7). The flagship game's foreign and sealed guards work; they were never ported to the expansion games. The counts move with listing volume (pokemon's active feed has shrunk sharply since late May), but the structural fact does not: the flagship game returns zero by construction, the others do not. This is the wrong-card risk the client named as the thing that loses paying users, and it maps straight to the rule written into the platform's own matcher code.

• One live read of production, 2026-06-09 12:27 UTC, via a dedicated read-only audit role against the platform's managed Postgres 17.6 instance.
• Per-game coverage is computed two ways and both are shown: card-only (matched over listings the matcher classified as single cards) and all-active (matched over all active listings, which includes lots, sealed, and accessories that should never match one card). The all-active number is always lower; the honest read is to show both.
• Catalogued counts are distinct catalog cards per game. A game with a large catalogue and zero active listings is expansion-ready in the data sense: the cards exist, the eBay feed does not yet.
• The foreign-title risk count per game is the number of active listings carrying a foreign-language print marker that matched (or would queue) against an English card. The flagship game's guard returns zero by construction; the other games have no equivalent guard.

These are the games with active eBay listings today. Card-only is the rate over listings the matcher itself called single cards. All-active is the rate over the full active feed.

The spread from 94% to 36% is per-game coverage, not a single global number. A rule pack per game is what closes it, and pokemon is where the signal is largest.

The catalogue holds far more than the four ingested games. Several games are fully catalogued with no eBay ingest yet, so the cards are sitting in the database waiting for a feed and a parser.

MTG is the headline. The client named it as a target, and it is already fully catalogued at 112,693 cards with zero eBay ingest. The runway is ingest plus a per-game parser, not catalog work. Yu-Gi-Oh is the same shape one tier down: 46,188 cards catalogued, zero active listings, expansion-ready the moment a feed and a set-code rule pack land.

This is the structurally durable reading: between the late-May proof and this live read the raw counts moved with listing volume, but the shape held exactly: the flagship game zero, every expansion game non-zero. It is the clean, reproducible one to lead the accuracy argument with.

The flagship game returns zero because its foreign-print and sealed guards exist. The expansion games return non-zero because those guards were never ported. The raw count moves with each game's active listing volume (pokemon's feed has shrunk from roughly 260k active listings in late May to about 25k at this read, so its foreign-title count fell with it), but the structural gap does not move: the flagship game is zero by construction, the others are open. Every one of those listings is a place where a foreign-language print can resolve to (or queue against) the wrong English card. That is the wrong-card risk in the client's own words, and it is exactly what the strict matcher's header comment guards against:

"Wrong matches = wrong arbitrage = [the client] buys the wrong card. Return null rather than guess." (verbatim from the platform's strict-matcher header comment; the client's name is replaced)

The same per-game work-stream that lifts coverage also ports these guards, so coverage and accuracy move together.

For context on the read-path side, the live arbitrage surface is about $56,598 of addressable spread across 252 cards that have a buyable raw print under the sold-comp, roughly 10% of matched cards carrying any arbitrage row. (This is card-market spread, domain data, and it moves as eBay listings age in and out; it was higher in late May before listings aged off.) The cross-link that puts the opportunity side and the raw-buy side on one screen is still unbuilt; it is sized separately at /m3-arbitrage.

• The coverage and foreign-risk numbers are a single live read at one timestamp. The eBay feed moves daily; the card-only and all-active rates will drift a few points as listings age in and out. The catalogue counts and the per-game gap structure are stable; the exact percentages are a snapshot.
• "Worst live game" means worst of the games that have active listings. MTG and Yu-Gi-Oh are not in that ranking because they have no active listings yet; they are expansion-ready, not low-coverage.
• The foreign-title risk count is the number of at-risk listings, not a count of confirmed wrong matches. The point is that the guard surface is zero on the flagship game and open on the others; a funded slice would port the guard and measure the wrong-match rate it removes per game.

1. Per-game set-code and name rule packs, starting with the worst live game (pokemon) and the named expansion target (MTG, on ingest). Parse each game's code format and resolve to the exact printing, the way the flagship game's tier already does.
2. Port the foreign-print and sealed guards per game. They exist for the flagship game and return zero risk; porting them closes the foreign wrong-card surface (212 / 105 / 7 across one-piece / dragon-ball-super / pokemon at this read, and it scales back up with each game's listing volume) and the lots-as-cards surface in one pass.
3. Ingest plus parser for the catalogued-but-not-ingested games (MTG first), so the 112,693 MTG cards already in the catalogue become matchable against a live feed.

One work-stream, two payoffs per game: coverage up, wrong-card risk down. The worst live game and the named expansion target are where it pays first.

• Per-game coverage, catalogue counts, and foreign-risk counts: the recon22 / recon22b live queries against production (dedicated read-only role, 2026-06-09 12:27 UTC). Arbitrage surface: the same read, cross-referenced with the sold-comp join used on the /m3-arbitrage sizing.