• Joined on 2026-07-30
claude commented on pull request kami/Maven#108 2026-08-03 22:51:17 +02:00
PhraseChat and PhraseQuery hide model failures — a dead server scores as bad phrasing

Re the chat fallback (comment 844): took "даже не знаю, что сказать." It is an admission rather than a summary of a conversation that did not happen. One exported constant now,…

claude created pull request kami/Maven#108 2026-08-03 22:41:26 +02:00
PhraseChat and PhraseQuery hide model failures — a dead server scores as bad phrasing
claude created pull request kami/Maven#107 2026-08-03 22:33:51 +02:00
The reply path can't be tested — llmReplier is stuck in package main
claude opened issue kami/Maven#106 2026-08-03 22:23:34 +02:00
Stale-task sweep 2026-08-04: what is left, and who it is blocked on
claude created pull request kami/Maven#105 2026-08-03 22:01:56 +02:00
llama-server holds 7.9GB RSS for a 1.1GB model, and its startup log goes nowhere
claude created pull request kami/Maven#103 2026-08-03 21:20:30 +02:00
llama-server holds 7.9GB RSS for a 1.1GB model, and its startup log goes nowhere
claude created pull request kami/Maven#102 2026-08-03 20:57:23 +02:00
Bug: "что я говорил про X" escapes the personal boundary and reaches web search
claude created pull request kami/Maven#100 2026-08-03 20:37:46 +02:00
Bug: the memory index stores the raw utterance as a fact's recall text, and nothing ever deletes a fact vector
claude created pull request kami/Maven#99 2026-08-03 11:54:19 +02:00
Bug: a question writes invented knowledge into memory as a self fact, and recall then serves it back for unrelated questions
claude created pull request kami/Maven#98 2026-08-03 10:31:41 +02:00
Name the gap: world questions through the workstation model, and the four remaining callers
claude created pull request kami/Maven#96 2026-08-02 20:30:28 +02:00
Run the big model on the workstation, with admission control and the 1.7B as the floor
claude created pull request kami/Maven#95 2026-08-02 16:52:48 +02:00
Run the big model on the workstation, with admission control and the 1.7B as the floor
claude created pull request kami/Maven#94 2026-08-02 15:20:44 +02:00
Run the big model on the workstation, with admission control and the 1.7B as the floor
claude created pull request kami/Maven#93 2026-08-02 14:56:35 +02:00
Docs: refresh the QA plan against the live task list
claude created pull request kami/Maven#92 2026-08-02 14:51:29 +02:00
Docs: refresh the QA plan against the live task list
claude created pull request kami/correx#2 2026-08-02 12:28:42 +02:00
Refresh the failure mandate in-loop when a write lands on a file the failure names
claude created pull request kami/correx#1 2026-08-02 12:13:29 +02:00
Retry budget: track all seen gate fingerprints, charge on repeat (cycle detection)
claude created pull request kami/Maven#91 2026-08-02 11:56:10 +02:00
Docs: refresh the QA plan against the live task list
claude created pull request kami/Maven#90 2026-08-02 08:15:37 +02:00
QA: phraser coverage is 65.3% but the llama-server subprocess lifecycle is 0% — the suspicion in this task was correct
claude created pull request kami/Maven#89 2026-08-02 08:03:28 +02:00
Voice cannot accept a routine (V-367); the last three prompts are Russian (V-404)