allaun/Research-Stack
1
0
Fork 0
mirror of https://github.com/allaunthefox/Research-Stack.git synced 2026-07-31 03:05:21 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Add autonomous speedrun harness example

Browse source
This commit is contained in:
Allaun Silverfox 2026-05-16 19:12:25 -05:00
parent 200b29ead7
commit c1be5904f3
1 changed files with 54 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

54
shared-data/examples/famm_autonomous_speedrun_harness.example.json Normal file
View file

@ -0,0 +1,54 @@
{
"target_metric": 2990,
"metric_mode": "lower_is_better",
"noise_floor": 10,
"min_seeds": 3,
"source_refresh_hours": 12,
"warden_threshold": 0.45,
"harness": {
"agents": {
"rules_file": "AGENTS.md",
"autonomy": "keep_loop_running_unless_warden_blocks"
},
"goal": {
"file": "goal.md",
"target": "reach benchmark target in fewer steps with valid statistical receipt"
},
"plan": {
"file": "plan.md",
"state": "mutable frontier"
},
"thread": {
"file": "scratchpad/THREAD.md",
"state": "durable mission log"
},
"source_refresh": {
"cadence_hours": 12,
"sources": ["upstream PRs", "records", "papers", "local best"]
}
},
"experiments": [
{
"id": "contra_muon_stack_v3",
"idea": {"name": "Contra-Muon stack", "lineage": ["upstream PR", "previous local frontier"]},
"patch": {"files": ["train_gpt.py"], "summary": "optimizer/schedule/init mutation"},
"config": {"schedule": 3050, "lr_power": 1.2},
"run": {"log": "runs/contra_muon_stack_v3.log"},
"metric": {"value": 2930, "target": 2990},
"judge": {"seeds": 31, "noise_floor": 10, "rule_pass": true, "reproducible": true},
"warden": {"source_stale_hours": 2, "local_loop_score": 0.1, "overfit_risk": 0.15},
"compute": {"gpu_hours": 42, "tokens": 120000000}
},
{
"id": "novelty_only_deadend",
"idea": {"name": "unvalidated novelty optimizer"},
"patch": {"files": ["optimizer.py"], "summary": "novel but weak idea"},
"config": {},
"run": {"log": "runs/novelty_deadend.log"},
"metric": {"value": 3105, "target": 2990},
"judge": {"seeds": 1, "noise_floor": 10, "rule_pass": true, "reproducible": false},
"warden": {"source_stale_hours": 20, "local_loop_score": 0.7, "overfit_risk": 0.4},
"compute": {"gpu_hours": 12, "tokens": 50000000}
}
]
}