Scale ablation early-stop caps to the compressed window (scale=0.2)

At replay_time_scale=0.2 the 600s arrival window compresses to 120s, so
the inherited 900s wall-clock elapsed cap let overloaded TP1 probes burn
~15min each (the tractability hazard the brief flagged). Scale the caps
proportionately to the time axis: early_stop_max_elapsed_s 900->180,
early_stop_max_lag_s 120->30. Feasible probes (~120s arrival + drain)
finish well inside 180s; overloaded probes die in ~3min. Both configs
still differ only in use_harness + study_id. Adds the ablation doc
skeleton and a read-only trajectory-extraction helper.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
2026-06-16 19:49:57 +08:00
parent a16016a876
commit d975e57bb5
4 changed files with 137 additions and 4 deletions

View File

@@ -131,8 +131,8 @@
"max_input_tokens": 8192
},
"replay_time_scale": 0.2,
"early_stop_max_lag_s": 120.0,
"early_stop_max_elapsed_s": 900.0,
"early_stop_max_lag_s": 30.0,
"early_stop_max_elapsed_s": 180.0,
"adaptive_stop": {
"enabled": true,
"tau": 0.9,

View File

@@ -131,8 +131,8 @@
"max_input_tokens": 8192
},
"replay_time_scale": 0.2,
"early_stop_max_lag_s": 120.0,
"early_stop_max_elapsed_s": 900.0,
"early_stop_max_lag_s": 30.0,
"early_stop_max_elapsed_s": 180.0,
"adaptive_stop": {
"enabled": true,
"tau": 0.9,