Throughput
How many issues were completed per week.
12
completed in range
9
by AI agents
Contributing tickets (12)
š
12
| Keyā | Summaryā | Assigneeā | Completedā¼ |
|---|---|---|---|
| TAA-18 | Model registration silently didn't happen on completed training jobs with register_model=true | Semenyuk Dmitriy | 7/7/2026 |
| TAA-20 | Fail fast when a training config needs the solver but the job image doesn't have it | OliverAI | 7/7/2026 |
| TAA-19 | Training runs don't log the demonstration-imitation signal (bc_loss / bc_weight) ā add the instrumentation | OliverAI | 7/7/2026 |
| TAA-16 | Autoplay orchestrator: serve the new 3900-input observation schema (blind_v8) alongside the current 1011-input one | OliverAI | 7/7/2026 |
| TAA-13 | Wire parallel eval to MLflow and verify it matches serial on the real RL champion | Semenyuk Dmitriy | 7/4/2026 |
| TAA-7 | Make the canonical lever-screen statistically valid (train-to-budget + Nā„2000), forbid underpowered verdicts | OliverAI | 7/4/2026 |
| TAA-12 | Deterministic (argmax) eval arm logs NaN/empty win_rate instead of a number | OliverAI | 7/3/2026 |
| TAA-11 | Rule-server loss-heuristic false-positives on draw-heavy-but-winnable positions | OliverAI | 7/3/2026 |
| TAA-6 | Background solvable-seed collector (Azure CPU job) | OliverAI | 7/3/2026 |
| TAA-5 | TAA-2 follow-up: run the argmax-collapse E1-E7 investigation and deliver the findings doc (PR #34887 added only the tooling) | Semenyuk Dmitriy | 7/1/2026 |
| TAA-2 | Root-cause the argmax-collapse: why does greedy (deterministic) eval score ~0% when stochastic scores ~43%? | OliverAI | 7/1/2026 |
| TAA-1 | Orchestrator autoplay UI: always show the real/actual seed for RL agents | OliverAI | 7/1/2026 |
Showing 1ā12 of 12
Page 1 of 1