From round 2 of the adversarial review of #458. Pre-existing — it was already off by 12 on main, so filing rather than fixing in that PR.
.claude/skills/ground-taxa-gtdb/SKILL.md:163-166 says the grounded blocks are:
715 grounded blocks: 336 s__, 225 g__, 72 d__, 51 p__, 13 c__, 11 o__, 7 f__ … 69.6% grounded, or 62.5% once d__ is excluded
Measured over kb/communities today: 728 blocks — 338 s__, 231 g__, 71 d__, 50 p__, 15 o__, 12 c__, 11 f__; 70.8%, or 63.9% excluding d__.
Every figure is off, including the two percentages, and the 62.5% / 7.0 pp line is quoted as a coverage result.
The underlying problem is that these are hand-maintained snapshots with nothing keeping them honest — the same shape as the status table two sections above, which #458 had to correct twice in one review cycle, and the curated: true enumeration that #450 found naming a block that carried no pin. Worth deciding whether the skill should quote counts at all, or point at a command that computes them:
grep -rho 'gtdb_id: GTDB:[a-z]__' kb/communities | sort | uniq -c
From round 2 of the adversarial review of #458. Pre-existing — it was already off by 12 on
main, so filing rather than fixing in that PR..claude/skills/ground-taxa-gtdb/SKILL.md:163-166says the grounded blocks are:Measured over
kb/communitiestoday: 728 blocks — 338s__, 231g__, 71d__, 50p__, 15o__, 12c__, 11f__; 70.8%, or 63.9% excludingd__.Every figure is off, including the two percentages, and the
62.5% / 7.0 ppline is quoted as a coverage result.The underlying problem is that these are hand-maintained snapshots with nothing keeping them honest — the same shape as the status table two sections above, which #458 had to correct twice in one review cycle, and the
curated: trueenumeration that #450 found naming a block that carried no pin. Worth deciding whether the skill should quote counts at all, or point at a command that computes them: