Issues

• [RFC] A recipe only defines configuration

  area: corearea: servicearea: trainingenhancementRFC
  Human-Agent-Society/reef#1390 · Filter by author BobbyZhouZijian opened on Nov 12, 2026

• [Bug] AutoEnvScaling numbers its generations from 0 again after #1371, so new tasks take published names

  area: harnessbug
  Human-Agent-Society/reef#1380 · Filter by author Benjamin-eecs opened on Nov 12, 2026
  

• [RFC] Replace a task item's files in place

  area: serviceenhancementRFC
  Human-Agent-Society/reef#1359 · Filter by author BobbyZhouZijian opened on Nov 11, 2026

• [RFC] Run the rollouts of an ended evaluation that have no score again

  area: serviceenhancementRFC
  Human-Agent-Society/reef#1340 · Filter by author BobbyZhouZijian opened on Nov 11, 2026

• [Bug] gepa against AIME - replacing gpt with qwen model shows inverse performance

  area: observabilitybugstatus: needs-triage
  Human-Agent-Society/reef#1339 · Filter by author yanan1116 opened on Nov 11, 2026
  1

• [RFC] Support student and teacher routing with shadow execution and replay

  area: serviceenhancementRFCstatus: needs-triage
  Human-Agent-Society/reef#1308 · Filter by author RickWayne1125 opened on Nov 9, 2026

• [RFC] Give each evaluation trial to a deployment's trial observer

  enhancementRFCstatus: needs-triage
  Human-Agent-Society/reef#1304 · Filter by author BobbyZhouZijian opened on Nov 9, 2026

• [RFC] Link model calls by their messages in ATIF form

  enhancementRFCstatus: ready
  Human-Agent-Society/reef#1300 · Filter by author BobbyZhouZijian opened on Nov 9, 2026
  1

• [RFC] Serving work that runs in parallel, stops when replaced, deploys again on request, and gives its reason

  enhancement
  Human-Agent-Society/reef#1299 · Filter by author BobbyZhouZijian opened on Nov 9, 2026
  1

• [RFC] Keep every reward of an evaluation rollout's verifier

  enhancementRFCstatus: ready
  Human-Agent-Society/reef#1297 · Filter by author BobbyZhouZijian opened on Nov 9, 2026
  1

• [RFC] Give a task's verifier the ATIF trajectory of a Reef adapter's run

  area: harnessenhancementRFCstatus: ready
  Human-Agent-Society/reef#1290 · Filter by author BobbyZhouZijian opened on Nov 9, 2026
  2

• [RFC] Stop an evaluation when too many rollouts fail, and give its progress

  enhancement
  Human-Agent-Society/reef#1287 · Filter by author BobbyZhouZijian opened on Nov 9, 2026

• [RFC] Report an artifact's startup in three fixed phases, with their times

  enhancementRFC
  Human-Agent-Society/reef#1283 · Filter by author BobbyZhouZijian opened on Nov 9, 2026

• [RFC] Limit the rollouts of an evaluation that run at once

  enhancementRFC
  Human-Agent-Society/reef#1276 · Filter by author BobbyZhouZijian opened on Nov 8, 2026
  1

• [RFC] Explain sample processing tasks: triggers, backoff, logs and lineage

  enhancementRFC
  Human-Agent-Society/reef#1267 · Filter by author BobbyZhouZijian opened on Nov 8, 2026
  1

• [RFC] SPADE trains its Environment Designer on its own replies

  enhancementRFCstatus: needs-triage
  Human-Agent-Society/reef#1261 · Filter by author Benjamin-eecs opened on Nov 8, 2026
  

• [RFC] Keep the model calls of each evaluation rollout

  Human-Agent-Society/reef#1258 · Filter by author BobbyZhouZijian opened on Nov 8, 2026

• [RFC] Each model artifact names a serving profile

  enhancementRFC
  Human-Agent-Society/reef#1233 · Filter by author BobbyZhouZijian opened on Nov 7, 2026
  2

• [RFC] Split a Scenario's traffic across its served artifacts by weight

  area: serviceenhancementRFCstatus: needs-triage
  Human-Agent-Society/reef#1225 · Filter by author BobbyZhouZijian opened on Nov 7, 2026

• [RFC] Use an artifact in a scenario through its item, with the same artifact ID and no copy

  area: artifactsenhancementRFCstatus: ready
  Human-Agent-Society/reef#1218 · Filter by author BobbyZhouZijian opened on Nov 7, 2026
  2

• [RFC] Evaluate a model artifact with a harness artifact from another scenario

  area: trainingenhancementRFCstatus: ready
  Human-Agent-Society/reef#1208 · Filter by author BobbyZhouZijian opened on Nov 6, 2026
  1

• [Feature] Check the candidates of one generation at the same time

  area: trainingenhancement
  Human-Agent-Society/reef#1204 · Filter by author Benjamin-eecs opened on Nov 5, 2026
  

• [Feature] Let generated-task rounds run ahead of training

  area: trainingenhancement
  Human-Agent-Society/reef#1202 · Filter by author Benjamin-eecs opened on Nov 5, 2026
  

• [Feature] AutoEnvScaling seeds from TMax v1.1's repaired Harbor release

  area: harnessenhancement
  Human-Agent-Society/reef#1200 · Filter by author Benjamin-eecs opened on Nov 5, 2026
  

• [Bug] Generated-task runs on Tinker have no objective since #1186, so AutoEnvScaling and SPADE do not train

  area: trainingbug
  Human-Agent-Society/reef#1194 · Filter by author Benjamin-eecs opened on Nov 5, 2026