EPS

Sessions

Happy sessions driving EPS issues. The summarizer cron writes the cache every 5 minutes; this page auto-refreshes every 30 seconds.

11 sessions · auto-refresh every 30sdata as of
IssueStatusProgressLast activitySummary age
#1112
Matched-install LoRA and full fine-tuning differ in how far, not in what shape, they shift residual-stream activations (MODERATE confidence)
Follow-ups running

Waiting on the analyzer re-fold's completion notification to proceed with interpretation/clean-result critic re-gating and the subsequent 9a-quater extension before re-parking the round.

cmrawddu · claude-haiku-4-5-20251001

#1138
daily-fix: verbatim bare push/merge snippets in SKILL.md
Completed

Recording completion markers and confirming the merged code landed on main.

cmrbwtyw · claude-haiku-4-5-20251001

#1142
workflow-fix: gotchas entry — vLLM H100 prefix-cache IMA differential + mitigation
Completed

Recording completion status and updating session metadata.

cmrby4j5 · claude-haiku-4-5-20251001

#1139
daily-fix: ad-hoc result summaries state data provenance
Completed

Posting final merge markers and updating task completion status after successfully rebasing PR #903 to main and syncing the repository root. The session is now idle, having confirmed the merged code (SHA 0054087524) is live on origin/main with the CLAUDE.md landing paragraph in place.

cmrbxwia · claude-haiku-4-5-20251001

#1137
daily-fix: suppress stalled-alert on fresh poll-tick marker
Completed

Posting the merged marker and updating task status to completed; the session is wrapping up post-merge cleanup and final documentation.

cmrbvrh2 · claude-haiku-4-5-20251001

#1090
The training context, not the contrastive-negative panel, gates where implanted content behaviors express on-policy (MODERATE confidence)
Awaiting promotion

Waiting for the user to promote the completed clean-result to production or select one of three parked follow-up experiments (per-question decomposition, impolite dose-gate, or multi-seed replication).

cmrb1r49 · claude-haiku-4-5-20251001

#742
Linear decoding leaves most of the extractable sycophancy information in base-model activations unread (MODERATE confidence)
Awaiting promotion

Awaiting user promotion decision on the clean result; the pipeline is complete and parked at a user-only gate with no autonomous work remaining.

cmr5fxre · claude-haiku-4-5-20251001

#825
The context→answer-profile map exists in pretrained Qwen2.5-7B at ~87% of instruct strength, and its answer-specificity survives an on-policy separator control only in attenuated form (MODERATE confidence)
Follow-ups running

#825 The context→answer-profile map exists in pret · followups_running / Phas…

cmrbtm01 · self-report

#1092
Realistic sparse-crossed prefix x query corpus: prefix-map vs context-map, averaging-rank collapse, and prefix/query variance decomposition of answer-state transport
Running

#1092 Realistic sparse-crossed prefix x query corpu · lau… · ▓▓▓▓░ 71% overdue

cmraucty · self-report

#1143
workflow-fix: verify_task_body file-count claim-matcher misses footer shape
Reviewing

#1143 workflow-fix: verify_task_body file-count cla · reviewing · ▓▓▓▓▓ 95%

cmrby5a7 · self-report

#1144
Port issue-1090 stranded shared-src fixes to main (sft.py CVD pin, datagen mt500 + posonly panel, bare/ICL contexts)
Running

#1144 Port issue-1090 stranded shared-src fixes to · implementing · ▓░░░░ 27%

cmrbzlgk · self-report