Block a user
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,…
PhraseChat and PhraseQuery hide model failures — a dead server scores as bad phrasing
The reply path can't be tested — llmReplier is stuck in package main
Stale-task sweep 2026-08-04: what is left, and who it is blocked on
llama-server holds 7.9GB RSS for a 1.1GB model, and its startup log goes nowhere
llama-server holds 7.9GB RSS for a 1.1GB model, and its startup log goes nowhere
Bug: "что я говорил про X" escapes the personal boundary and reaches web search
Bug: the memory index stores the raw utterance as a fact's recall text, and nothing ever deletes a fact vector
Bug: a question writes invented knowledge into memory as a self fact, and recall then serves it back for unrelated questions
Name the gap: world questions through the workstation model, and the four remaining callers
Run the big model on the workstation, with admission control and the 1.7B as the floor
Run the big model on the workstation, with admission control and the 1.7B as the floor
Run the big model on the workstation, with admission control and the 1.7B as the floor
Docs: refresh the QA plan against the live task list
Docs: refresh the QA plan against the live task list
Refresh the failure mandate in-loop when a write lands on a file the failure names
Retry budget: track all seen gate fingerprints, charge on repeat (cycle detection)
Docs: refresh the QA plan against the live task list
QA: phraser coverage is 65.3% but the llama-server subprocess lifecycle is 0% — the suspicion in this task was correct
Voice cannot accept a routine (V-367); the last three prompts are Russian (V-404)