Duplicate checkout trees double-count every estate census — repos/ vs hyper-repos/ overlap on 213 names, _SET containers hold 412 nested repos

[hyperpolymath]

Claim

Every estate census that walks the local disk double-counts: multiple overlapping checkout trees exist, so per-repo measurements (gate scans, shadow-tracking counts, language inventories) inflate or contradict each other depending on which tree a scanner happens to walk.

Evidence (measured 2026-08-31, during #678 re-verification)

• repos/ overlaps hyper-repos/ on 213 repo names.
• 20 _SET container dirs under hyper-repos/ hold 412 nested repos, heavily duplicated against the direct checkouts.
• Known duplicate pairs: gitbot-fleet-fresh-2026-08-28 vs gitbot-fleet-recovery-20260824; metadatastician/* mirrored under bulk_update_oikos/*.
• Concrete effect: Estate: 132 repos track files their own .gitignore hides; template-sync sweep (92 repos) left a SILENT fake gate in 29 #678's original population of 226 repos no longer exists as a set — the same measurement today yields 231 direct + 412 nested, and no two censuses since the reorg have been counting the same universe.

Why it matters

Numbers in issues stop being comparable across time (this is a large part of why re-verified issue numbers keep failing), and any sweep that enumerates the disk risks patching the same repo twice via different paths — or patching a stale duplicate and reporting success.

First step

Declare ONE canonical tree per repo (likely hyper-repos/<name> for org hyperpolymath) in a machine-readable manifest, and make every census/sweep script resolve through it. Then decide the fate of repos/, the _SET containers, and the recovery/fresh duplicates (archive, delete, or mark non-canonical).

Found while re-verifying #678; filed per the discoveries-become-issues rule. Related: #677 (damaged object stores — several of the affected names exist in more than one tree).

[hyperpolymath]

Additional duplicate-tree measurements from the #658/#664 verification pass (2026-08-31), confirming the census-integrity problem from a second instrument: bulk_update_oikos/ holds 134 checkouts and a nested metadatastician/ tree 17, both duplicating canonical checkouts; the rogue repos/ tree has recurred at 383 checkouts (it has been swept before — see the rogue-repos recurrence history). Also one concrete casualty: a stale checkout under the old name bag-of-actions persists on disk although the repo was renamed continuation-mesh — exactly the class of ghost a canonical-tree manifest would catch.