fix(website): the silicon table follows the specs that moved repository

[gHashTag]

check:silicon is red on main after #1021, and the same table ships in the bundle: fifteen specs currently read "No hardware run" on t27.ai while their silicon record sits one renamed key away.

{
  "version": 1,
  "head_sha": "8a2629a87fd485925b1a468e71942ecc08537f1d",
  "summary": "The silicon coverage table names spec paths, and the GitHub-sourced catalog refresh moved the GF ladder's second copy from chips/euler to tt-trinity-euler; this follows those fifteen entries and classifies the nine widths the refresh brought in from t27 master.",
  "changes": [
    "src/data/siliconHistory.ts: fifteen SPEC_TO_FORMAT, EXCLUDED and FAMILY_SPECS entries renamed from chips/euler/** to tt-trinity-euler/**, the repository that owns those files; the bytes are the same ones, attributed differently by the refresh.",
    "src/data/siliconHistory.ts: GF6, GF10 and GF14 excluded as widths between the rungs of the silicon-verified GF4-GF24 ladder, and GF48, GF96, GF512 and GF1024 as widths above it, each with the wording the table already uses for its neighbours.",
    "src/data/siliconHistory.ts: the canonical copies of GF64, GF128 and GF256 excluded alongside their tt-trinity-euler twins, so one copy excluded and its twin silent cannot recur.",
    "No spec gains a hardware claim here: every entry added says a cell was not run, which is the understating direction the table's header asks for when a record is absent."
  ],
  "tests": [
    {
      "command": "npm run check:silicon",
      "result": "every spec naming a verified format is now classified",
      "status": "passed",
      "evidence": "1407 specs scanned, 19 mapped, 7 family, 23 excluded; the guard's own self-test still detects an unclassified gf16 path"
    },
    {
      "command": "npm run check:api check:aria check:clients-console check:crons-catalog check:discovery check:explorer-languages check:skills-catalog check:subpath-urls check:tools-site check:tri-identity check:viewport",
      "result": "every other file-driven check in the website-checks list passes",
      "status": "passed",
      "evidence": "run with BOT_ROOT and T27_ROOT pointed at a nonexistent path, as a runner has them"
    },
    {
      "command": "npm run test:sync-crons test:sync-skills test:viewport-spec typecheck:ratchet",
      "result": "the unit tests and the type ratchet hold at baseline",
      "status": "passed",
      "evidence": "179 errors across 26 files, baseline 179 across 26, no file gained type errors"
    },
    {
      "command": "npm run audit:en audit:ru audit:mobile",
      "result": "the three browser audits could not run on this machine",
      "status": "not_run",
      "evidence": "each exits with: Chrome/Chromium not found, set CHROME_PATH; no Chrome or Chromium is installed here, so CI is the first place they run"
    }
  ],
  "limitations": [
    "The three language and mobile audits are unrun here for want of a browser, so this branch relies on CI for them.",
    "The exclusions state that no hardware cell was run for those widths, which is a statement about the absence of a record rather than a measurement of the format.",
    "t27.ai carries the previous bundle until this lands and is published again, so the fifteen understated specs stay understated until then."
  ],
  "tags": ["website", "silicon", "catalog", "specs"],
  "blog": {
    "title": "When a spec changes repository, its record has to follow",
    "summary": "A catalog refresh attributed the GF ladder's second copy to the repository that owns it, and fifteen hardware records quietly stopped matching the specs they belonged to.",
    "outline": [
      "The corpus carried the GF ladder three times over, and the silicon table lists every copy by path precisely so that a spec describing a silicon-verified format does not read as unverified merely because a reader opened a different copy of it.",
      "Vendoring the corpus from GitHub rather than from a branch that inlined the chip repositories changed where the second copy lives, from chips/euler to tt-trinity-euler, and a table keyed by path cannot follow that on its own.",
      "The gate caught it, which is what the gate is for, but the same table ships to the browser, so for the length of one deploy fifteen specs understated what had been measured about them.",
      "The refresh also brought nine widths that declare a format and have no cell behind them, and each is now excluded with a stated reason rather than left to be discovered again by the next person who runs the guard."
    ]
  }
}