kami/Maven
1
0
Fork 0
Code Issues 1 Pull Requests 1 Actions Packages Projects Releases Wiki Activity

Stale-task sweep 2026-08-04: what is left, and who it is blocked on #106

New Issue
Open
opened 2026-08-03 22:23:33 +02:00 by claude · 2 comments
claude commented 2026-08-03 22:23:33 +02:00
Contributor
Copy Link
Copy Source

Stale-task sweep of Vikunja project Maven, 2026-08-04. Ten tasks closed. The rest
are sorted into what is blocked on the owner and what is not. Vikunja stays the
durable store. This issue is only the reading list, and every number below is a
Vikunja id, not a Gitea issue.

Decisions only the owner can make

  • V-357. Hexis: is entity-existence validation the permanent target guard, or does blessing land in Nexus?
  • V-464. The router buys 4 accuracy points, not a doubling. Keep it, demote it to a classifier-unconfident call, or push more into stage 0.

Needs a person at a machine

  • V-15. Deploy the desk_active presence script on the desk PC.
  • V-16. Create the Kuma API key for service_down polling.
  • V-122. Finish CPT training of Qwen3-1.7B on workpc.
  • V-355. Hexis: deploy the auth change.
  • V-463. mavwaked and mavenclient run nowhere, so the wake path is untestable.
  • V-492. Workstation offload end to end: card free, card held, machine asleep.
  • V-320, item 4. Kill llama-server and confirm the classifier answers. Needs a permission to grant.
  • V-472, items 4 and 5. Nexus holds no entities. The ecosystem trace fix has never been validated on this box. Items 1 to 3 are now cleared: morning_routines, feeds and crawl are in deploy/mavend.json.

Blocked on something outside the repo

  • V-14. Cold-start unlock needs a PRF authenticator.
  • V-125. Expense and income tracking via zenmoney.
  • V-256. SmartHome through Home Assistant.
  • V-477. The model swap and the self-update are both unreachable on this deploy.
  • V-478. The Vikunja MCP server binds loopback, so no container reaches it.

Live defects, verified still real, nothing blocking them

  • V-396. llmReplier is stuck in package main, so the reply path cannot be scored.
  • V-397. PhraseChat and PhraseQuery return canned strings with a nil error, so a dead server scores as bad phrasing.
  • V-402. Sampling temperature is a single global 0.7 across world.go and llmphraser.go, never swept.
  • V-323. The 60s startup-timeout arm in startLlamaProc is a literal. It needs a StartupTimeout config field to be testable.
  • V-479. An unconfigured capability declines its own turn instead of naming the gap, and the question escapes to web search.
  • V-491. No LimitCORE in deploy/mavgpud.service, and a yield logs identically to a crash.
  • V-498. Point 1 of 470: что дальше? and расскажи про X still reach the model with no stage 0 rule.

Closed in this sweep

V-362, V-365, V-375, V-387, V-398, V-400, V-401, V-403, V-406, V-407, V-446,
V-447, V-459, V-485, V-490.

V-485 and V-490 shipped their code. The residual verification and the card-free
fraction both live in V-492.

Not covered

Roughly 130 tasks were not read one by one. Most are QA sittings that need a
person at the box. The sweep targeted three classes. A premise naming
Qwen3.5-0.8B or llama.cpp b9351. Acceptance criteria the tree now satisfies.
Work superseded by an owner's call.

Stale-task sweep of Vikunja project Maven, 2026-08-04. Ten tasks closed. The rest are sorted into what is blocked on the owner and what is not. Vikunja stays the durable store. This issue is only the reading list, and every number below is a Vikunja id, not a Gitea issue. ## Decisions only the owner can make - [ ] V-357. Hexis: is entity-existence validation the permanent target guard, or does blessing land in Nexus? - [ ] V-464. The router buys 4 accuracy points, not a doubling. Keep it, demote it to a classifier-unconfident call, or push more into stage 0. ## Needs a person at a machine - [ ] V-15. Deploy the desk_active presence script on the desk PC. - [ ] V-16. Create the Kuma API key for service_down polling. - [ ] V-122. Finish CPT training of Qwen3-1.7B on workpc. - [ ] V-355. Hexis: deploy the auth change. - [ ] V-463. mavwaked and mavenclient run nowhere, so the wake path is untestable. - [ ] V-492. Workstation offload end to end: card free, card held, machine asleep. - [ ] V-320, item 4. Kill llama-server and confirm the classifier answers. Needs a permission to grant. - [ ] V-472, items 4 and 5. Nexus holds no entities. The ecosystem trace fix has never been validated on this box. Items 1 to 3 are now cleared: `morning_routines`, `feeds` and `crawl` are in `deploy/mavend.json`. ## Blocked on something outside the repo - [ ] V-14. Cold-start unlock needs a PRF authenticator. - [ ] V-125. Expense and income tracking via zenmoney. - [ ] V-256. SmartHome through Home Assistant. - [ ] V-477. The model swap and the self-update are both unreachable on this deploy. - [ ] V-478. The Vikunja MCP server binds loopback, so no container reaches it. ## Live defects, verified still real, nothing blocking them - [x] V-396. `llmReplier` is stuck in `package main`, so the reply path cannot be scored. - [x] V-397. `PhraseChat` and `PhraseQuery` return canned strings with a nil error, so a dead server scores as bad phrasing. - [ ] V-402. Sampling temperature is a single global 0.7 across `world.go` and `llmphraser.go`, never swept. - [x] V-323. The 60s startup-timeout arm in `startLlamaProc` is a literal. It needs a `StartupTimeout` config field to be testable. - [x] V-479. An unconfigured capability declines its own turn instead of naming the gap, and the question escapes to web search. - [x] V-491. No `LimitCORE` in `deploy/mavgpud.service`, and a yield logs identically to a crash. - [x] V-498. Point 1 of 470: `что дальше?` and `расскажи про X` still reach the model with no stage 0 rule. ## Closed in this sweep V-362, V-365, V-375, V-387, V-398, V-400, V-401, V-403, V-406, V-407, V-446, V-447, V-459, V-485, V-490. V-485 and V-490 shipped their code. The residual verification and the card-free fraction both live in V-492. ## Not covered Roughly 130 tasks were not read one by one. Most are QA sittings that need a person at the box. The sweep targeted three classes. A premise naming Qwen3.5-0.8B or llama.cpp b9351. Acceptance criteria the tree now satisfies. Work superseded by an owner's call.
kami commented 2026-08-04 00:02:00 +02:00
Owner
Copy Link
Copy Source

Overnight run 2026-08-04. The four live defects with nothing blocking them are done, each on its own stacked branch and PR:

  • V-479 → PR #114. Netscan and the crawler name the gap when unconfigured instead of letting the question escape to search.
  • V-498 → PR #115. Stage 0 claims "что дальше?" and "расскажи про X". Fixture gained ru-query-020 and ru-query-021; classifier+onnx 53/77 → 55/79, no new false clarify.
  • V-323 → PR #90 (the branch was already open). StartupTimeout is a Config field and phraser.startup_timeout, and the timeout arm is tested including the reap.
  • V-491 → PR #116. LimitCORE=0 in the unit, and a deliberate stop logs as a yield. No matching upstream llama.cpp issue found, so filing it is still open and needs an account on that tracker.

V-402 was left alone deliberately (owner's call before the run). The string-families umbrella V-500 also advanced: PRs #109, #110, #111, #112, #113 cover families 1 to 4 and the summaries.

Overnight run 2026-08-04. The four live defects with nothing blocking them are done, each on its own stacked branch and PR: - V-479 → PR #114. Netscan and the crawler name the gap when unconfigured instead of letting the question escape to search. - V-498 → PR #115. Stage 0 claims "что дальше?" and "расскажи про X". Fixture gained ru-query-020 and ru-query-021; classifier+onnx 53/77 → 55/79, no new false clarify. - V-323 → PR #90 (the branch was already open). StartupTimeout is a Config field and phraser.startup_timeout, and the timeout arm is tested including the reap. - V-491 → PR #116. LimitCORE=0 in the unit, and a deliberate stop logs as a yield. No matching upstream llama.cpp issue found, so filing it is still open and needs an account on that tracker. V-402 was left alone deliberately (owner's call before the run). The string-families umbrella V-500 also advanced: PRs #109, #110, #111, #112, #113 cover families 1 to 4 and the summaries.
kami commented 2026-08-04 08:44:26 +02:00
Owner
Copy Link
Copy Source

V-479 is fixed and closed. PR #114: an unconfigured netscan or crawler now claims its own turn and names the gap (net_off, page_off in the query family), instead of falling through to the search leg. The two strings are the ones QA 257 step 1 and QA 259 step 2 expect.

queryHome keeps the old fall-through deliberately, for a reason the ticket records.

V-479 is fixed and closed. PR #114: an unconfigured netscan or crawler now claims its own turn and names the gap (`net_off`, `page_off` in the query family), instead of falling through to the search leg. The two strings are the ones QA 257 step 1 and QA 259 step 2 expect. `queryHome` keeps the old fall-through deliberately, for a reason the ticket records.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
No items
No Label
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
kami
No Assignees
2 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: kami/Maven#106
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?