What
The vocabulary census reads local checkouts under MECHS_ROOT. Two are behind their remotes, so the census counts fewer records than the Mech actually publishes:
| Mech |
Local checkout |
Remote main |
Published site |
| CommunityMech |
364 (360 communities + 4 isolates) |
383 (379 + 4) |
379 communities |
| TraitMech |
694 |
696 |
696 trait records |
CommunityMech's checkout sits at 6bd643d (2026-09-14), 21 commits behind origin/main; the corpus gained 19 communities in that window. TraitMech's is exactly 2 commits behind, each adding one genomics trait YAML.
Note CommunityMech's site total and the fleet glob will never be equal even on a fresh checkout: the site hero counts kb/communities only, while the fleet glob also includes the 4 data/isolates records. Expect site N vs fleet N+4.
Why it matters
The card headline numbers on /mechs/ come from the published sites and are correct (379, 696). The heatmap, overlap chords and assets/fleet/{edges,cells} come from the checkouts and are low. The page already dates the census separately from the card totals, so nothing is stated falsely — but the two halves of the page disagree by more than the label explains.
Options
git pull each checkout before a census run, and say so in _fleet/README.md.- Have
roots.py warn when a checkout is behind its remote — it already hard-fails on a missing checkout and an empty glob, so a staleness warning fits its existing job.
Option 2 is the one that survives the next person forgetting.
Found during the refresh in #89.
What
The vocabulary census reads local checkouts under
MECHS_ROOT. Two are behind their remotes, so the census counts fewer records than the Mech actually publishes:mainCommunityMech's checkout sits at
6bd643d(2026-09-14), 21 commits behindorigin/main; the corpus gained 19 communities in that window. TraitMech's is exactly 2 commits behind, each adding one genomics trait YAML.Note CommunityMech's site total and the fleet glob will never be equal even on a fresh checkout: the site hero counts
kb/communitiesonly, while the fleet glob also includes the 4data/isolatesrecords. Expect site N vs fleet N+4.Why it matters
The card headline numbers on
/mechs/come from the published sites and are correct (379, 696). The heatmap, overlap chords andassets/fleet/{edges,cells}come from the checkouts and are low. The page already dates the census separately from the card totals, so nothing is stated falsely — but the two halves of the page disagree by more than the label explains.Options
git pulleach checkout before a census run, and say so in_fleet/README.md.roots.pywarn when a checkout is behind its remote — it already hard-fails on a missing checkout and an empty glob, so a staleness warning fits its existing job.Option 2 is the one that survives the next person forgetting.
Found during the refresh in #89.