Follow-up to #91, found by running the gate on the host it guards:
A by-digest pull leaves the image carrying every digest it was pulled under (4beb81ee… and the pin 69add58f…), so reading RepoDigests[0] reported a healthy pinned image as wrong and refused a window that was ready to run. The fact is membership, not position: podman image inspect -f '{{range .RepoDigests}}{{println .}}{{end}}' and match the pin as a line.
Also records today's measured step curve on the archon page: one 2-way 7,168-token prefill at a 32,768 calibration window runs 106 tok/s at an 8,192-token step and 76 tok/s at 16,384 (activation arena 3.39 → 4.50 GiB) — the wide step has an optimum, and the QSA score plane refusal that hid the whole option is an artifact of the 200,000-token production window.
Follow-up to #91, found by running the gate on the host it guards:
A by-digest pull leaves the image carrying **every** digest it was pulled under (`4beb81ee…` and the pin `69add58f…`), so reading `RepoDigests[0]` reported a healthy pinned image as wrong and refused a window that was ready to run. The fact is membership, not position: `podman image inspect -f '{{range .RepoDigests}}{{println .}}{{end}}'` and match the pin as a line.
Also records today's measured step curve on the archon page: one 2-way 7,168-token prefill at a 32,768 calibration window runs **106 tok/s at an 8,192-token step** and **76 tok/s at 16,384** (activation arena 3.39 → 4.50 GiB) — the wide step has an optimum, and the `QSA score plane` refusal that hid the whole option is an artifact of the 200,000-token production window.
---
cwd: /home/blank/.omp/wt/wt-20261027-142335-6d1fc1f · session: 01a112cd-f9a8-724e-8ec6-285213a85a68
A by-digest pull leaves the image carrying every digest it was pulled under, so
reading RepoDigests[0] reported a healthy pinned image as wrong and refused a
window that was ready to run. Membership, not position. Also records the measured
step optimum: 106 tok/s at an 8,192-token step against 76 at 16,384 on a 32,768
calibration window, so the wide step has a ceiling, not a plateau.
dank
merged commit 4db985b90d into main2026-10-30 16:30:07 -07:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Follow-up to #91, found by running the gate on the host it guards:
A by-digest pull leaves the image carrying every digest it was pulled under (
4beb81ee…and the pin69add58f…), so readingRepoDigests[0]reported a healthy pinned image as wrong and refused a window that was ready to run. The fact is membership, not position:podman image inspect -f '{{range .RepoDigests}}{{println .}}{{end}}'and match the pin as a line.Also records today's measured step curve on the archon page: one 2-way 7,168-token prefill at a 32,768 calibration window runs 106 tok/s at an 8,192-token step and 76 tok/s at 16,384 (activation arena 3.39 → 4.50 GiB) — the wide step has an optimum, and the
QSA score planerefusal that hid the whole option is an artifact of the 200,000-token production window.cwd: /home/blank/.omp/wt/wt-20261027-142335-6d1fc1f · session: 01a112cd-f9a8-724e-8ec6-285213a85a68