docs: publish real SFT log_history and add provenance note

docs: publish real SFT log_history and add provenance note

docs: add final blog post for submission

docs: add notebook and concise proof screenshots

docs: add training reproduction notebook link

docs: add terminal proof screenshots

docs: add final training evidence plots

docs: finalize phase-aware training evidence

fix(eval): add phase-aware hard and cascade action selection

fix(training): allow grpo script to import app config

fix(grpo): keep reward rollouts aligned with prompt seeds

fix(grpo): align hard and cascade training with final eval

fix(grpo): gate hard-task rollouts on correct first action

fix(grpo): salvage malformed action prefixes for reward scoring

fix(grpo): prefix-anchored generation + continuous reward ladder

fix(grpo): pre-apply chat template and tighten generation params

fix(docker): use python:3.13-slim to fix build (audioop-lts requires >=3.13)

fix(rl): cycle patch files in hard/cascade rollouts and trim default RL pool

docs(evidence): add multi-agent eval, long-horizon trace, and constrained eval logs

feat(rl): GRPO continuation from SFT adapter with multi-step task rollouts

feat(curriculum): adaptive difficulty for telemetry, masking, and secondary failures

test(reward): make reward audit run in-process when server is offline

fix(reward): preserve task pass thresholds and switch MER to efficiency multiplier

feat(env): add fleet_coordination multi-agent task

Harden submission evidence and reward integrity

Document training evidence and improve model evaluation

Add training evidence summary

Document training results and artifacts

Use constrained action scoring for model evaluation

Use multi-step SFT pipeline for reliable training evidence

Add SFT warmup and model evaluation pipeline

Improve GRPO rewards and add model evaluation

Handle chat completions in GRPO reward parser

Use vanilla TRL GRPO training stack

Patch Unsloth GRPO text-only trainer attributes

Fix Hugging Face Space and GRPO training config

Fix Hugging Face Space Docker config

Add Hugging Face Spaces config

chore(grading): raise medium pass threshold to 0.7 and document audit notes

feat(eval): add seed variance reporting script

feat(eval): add benchmark harness and reward integrity artifacts

test: add coalition and anti-gaming grader coverage

feat(env): add curriculum, challenge generation, coalition, and black-swan mechanics

chore(submission): add baseline snapshot and validation script

chore(deps): pin full runtime requirements

feat(env): add robustness, uncertainty consensus, and partial observability

Ship Round 2 manifest/docs, dashboard, and GRPO training pipeline

Add Fleet supervisor-worker delegation layer with /delegate API and tests

Add cascade grader and regression tests for cascade task

Track LLM token usage and send token_count in actions