vh/esh-pfi-infrastructure
1
0
Fork 0
Code Issues 7 Pull Requests Actions Packages Projects Releases Wiki Activity

memory: re-derive the shipped-corpora split-leak numbers with the committed gate

Browse Source
This commit is contained in:
Vuong Hoang
2026-09-17 11:43:48 -07:00
parent 707fae2b2c
commit 370b16ca15
2 changed files with 18 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
persistent-memory.d/2026-09-17-mccarthy-split-name-leak.md
+13 -4
View File
@@ -55,13 +55,22 @@ control.
## The shipped corpora, checked with the committed instrument
Re-derived with the COMMITTED gate, not a scratch probe:
```
lv-bronte clean of this class (`English women` vs `Englishwomen` is a false positive)
lv-hemingway 3 real: `Primi tivo`, `Pasionar ia`, `Chi cote` — SHIPPED and LIVE
lv-bronte GATE PASSED 0 separator-split survivors (9.7 s)
lv-hemingway GATE FAILED Pasionaria, Primitivo, Chicote (34.6 s)
1 occurrence each per copy, in all 6 copies — SHIPPED and LIVE
```
**Operator call outstanding** on whether 3 occurrences in 958k words warrant re-gating a
live adapter. Not acted on.
⚠ The first version of this scan was **too slow to run** on Hemingway — per-surface scanning
is O(surfaces x copies x corpus) and 881 surfaces x 10 copies was still going at 5 minutes
when it was killed. Rebuilt as one alternation pass, same trick `scan()` already used: 35 s,
identical verdict and identical hit counts on both McCarthy trees. **A gate too slow to run
is not a gate.**
**Operator call outstanding** on whether 3 names in a 958k-word corpus warrant re-gating and
retraining a live adapter. Not acted on.
## The chain was recovered, not remembered — and is now written down
persistent-memory.md
+5 -3
View File
@@ -132,9 +132,11 @@ kept at `corpus-{clean,renamed}.pre-splitfix` on gx10. Recipe: `scripts/mccarthy
### OPERATOR CALL — lv-hemingway carries 3 of the same hidden survivors and is LIVE
`Primi tivo`, `Pasionar ia`, `Chi cote` in a 958k-word corpus, on the adapter serving at
fv-ml1 GPU 0 `:8027`. lv-bronte is clean of the class. Re-gating means a corpus rebuild and
a retrain; three occurrences may not be worth it. **Not acted on.**
`Primi tivo`, `Pasionar ia`, `Chi cote` — one occurrence each per copy, in all 6 copies of a
958k-word corpus, on the adapter serving at fv-ml1 GPU 0 `:8027`. Confirmed by running the
committed gate, which now exits 1 on it (34.6 s). lv-bronte re-run clean (0, 9.7 s).
Re-gating means a corpus rebuild and a retrain; three names may not be worth it.
**Not acted on — this one is yours.**
### SHIPPED — `lv-hemingway` on ckpt850, live on `vllm-voices` (fv-ml1 GPU 0 :8027)