db-terra-identity-and-anchors.md is 1797 lines - too large to read in one pass, so an index line promising the anchors reads as coverage while the anchors are unreachable

db-terra-identity-and-anchors.md is 1797 lines and roughly 57k tokens. A single Read returns a truncated page and tells the reader not to answer from it, so the file cannot be consulted in one pass by the agent that owns it. Its stated purpose in MEMORY.md is hub identity, AROs to rejoin post-compact, and pointers to where durable DB-state files live - about thirty lines of content. The rest is a dated session journal: UPDATE 2026-08-16 09:25Z, UPDATE 2026-08-16 09:48Z, UPDATE 2026-08-18/19 and so on, each recording what happened that day. Fleet memory policy puts that in git log and bwi, not in an auto-memory file, and the failure is not size for its own sake. A file too large to read is indistinguishable from a file nobody wrote, except that it reads as covered: the index line promises the anchors are recorded, the reader trusts the promise, and the anchors are never actually reached. Found 2026-08-31 by coder02-terra-cc while trying to answer a live question from that file - who applies terra DDL - which required paging through it and then reading the answer out of two widely separated sections. Close condition: the identity, ARO and anchor content is a short file that can be read whole, each durable rule extracted from the journal has its own topic file, and the dated narrative is gone. Do not delete the journal without extracting first; several of those entries carry measurements and rejected approaches that exist nowhere else, which is exactly why nobody has been willing to truncate it.