nix.md: archetypes gate peaks at 1.7 GiB per process #65

Merged

vnprc merged 2 commits from agent/archetypes-gate-peak-measured into master

2026-09-23 16:55:57 +01:00

allod-agent commented

2026-09-23 05:42:33 +01:00

Member

Summary

One line in nix.md changes: the note that allod/archetypes "peaks around 5.7 GiB against a 7 GiB box" is replaced with what the gate does now. That figure predates the gate split and was steering sessions to over-serialize and to read any failure under load as an OOM. Nothing else changes. If a later measurement disagrees, edit the same line.

What proves it

Nine runs of ./check.sh in allod/archetypes at master 85d8c54 and on two branches that differ only in check message strings, eval cache disabled, each nix process under GNU time -f %M, free -m sampled every 5 s, with three resident agent processes of about 0.4 GB each on the box: largest single process 1715 to 1716 MiB in every run (runtime-module-selection), wall time 75 to 80 s, system available memory never below 4.8 GiB. Full table on allod/archetypes#31.

R0, docs only.

## Summary One line in `nix.md` changes: the note that `allod/archetypes` "peaks around 5.7 GiB against a 7 GiB box" is replaced with what the gate does now. That figure predates the gate split and was steering sessions to over-serialize and to read any failure under load as an OOM. Nothing else changes. If a later measurement disagrees, edit the same line. ## What proves it Nine runs of `./check.sh` in `allod/archetypes` at master `85d8c54` and on two branches that differ only in check message strings, eval cache disabled, each `nix` process under GNU `time -f %M`, `free -m` sampled every 5 s, with three resident agent processes of about 0.4 GB each on the box: largest single process 1715 to 1716 MiB in every run (`runtime-module-selection`), wall time 75 to 80 s, system available memory never below 4.8 GiB. Full table on allod/archetypes#31. R0, docs only.

allod-agent added 1 commit

2026-09-23 05:42:33 +01:00