Calibrate lossless Strategy Prefill budget

autoresearch/prefill/program.md

@@ -58,7 +58,7 @@ must be deployed outside this supervisor. Cold benchmarks are explicit,
 separate invocations of `scripts/benchmark_prefill_architecture.py`.
 
 Prefill budgets are hard admission limits, never truncation instructions.
-Strategy input must fit 4096 tokens by carrying the complete active leaf
+Strategy input must fit 8192 tokens by carrying the complete active leaf
 ancestry and its exact experiment records. Generator and Critic inputs must fit
 6144 tokens; the Critic always receives the complete current Generator output.
 If any complete semantic unit exceeds its budget, reject it before remote

autoresearch/prefill/supervisor.py

@@ -596,7 +596,7 @@ def propose_candidate(
     current: dict,
     results_text: str,
     ledger: dict,
-    max_prefill_tokens: int = 4096,
+    max_prefill_tokens: int = 8192,
 ) -> dict:
     from kakeya import Client
     from transformers import AutoTokenizer
@@ -1087,7 +1087,7 @@ def main() -> int:
     parser.add_argument(
         "--strategy-max-prefill-tokens",
         type=int,
-        default=4096,
+        default=8192,
     )
     parser.add_argument(
         "--tokenizer-id",