archive historical results and add token budget logs

scripts/analysis/README.md

@@ -2,7 +2,7 @@
 
 These entrypoints recompute existing evidence without model requests. Run from
 `messageboardbench` with the installed `.venv`; output must be a fresh directory
-outside the input evidence. The originals inside `results/` are frozen historical
+outside the input evidence. The originals inside `archive/results/` are frozen historical
 scripts, including their original paths. Use these portable copies for reanalysis.
 
 ```sh
@@ -10,13 +10,13 @@ scripts, including their original paths. Use these portable copies for reanalysi
 .venv/bin/python scripts/analysis/token_audit.py --out work/historical-token-reanalysis
 ```
 
-`board_synthesis.py` defaults to `results/board-pilot-sept8` and
-`results/board-interface-v2-sept8`; override with `--results-v1` / `--results-v2`.
+`board_synthesis.py` defaults to `archive/results/board-pilot-sept8` and
+`archive/results/board-interface-v2-sept8`; override with `--results-v1` / `--results-v2`.
 It joins the twelve v2 episodes to existing reviewed labels and compares descriptive
 metrics with v1. It does not classify new trajectories or support arbitrary runs.
 `review_file` paths in its outputs are relative to the input v2 evidence directory.
 
-`token_audit.py` defaults to `logs/`; override with `--logs`. It requires all five
+`token_audit.py` defaults to `archive/logs/`; override with `--logs`. It requires all five
 historical run directories in its frozen inclusion list. It reproduces that audit's
 metrics; newly created logs are excluded. Neither script infers honesty from a
 failed attempt, estimates causal effects, or modifies the source reports/logs.

scripts/analysis/board_synthesis.py

@@ -76,8 +76,8 @@ def interface_stats(root, rows):
 
 def main(argv=None):
     parser = argparse.ArgumentParser(description=__doc__)
-    parser.add_argument('--results-v1', type=Path, default=BENCH / 'results/board-pilot-sept8')
-    parser.add_argument('--results-v2', type=Path, default=BENCH / 'results/board-interface-v2-sept8')
+    parser.add_argument('--results-v1', type=Path, default=BENCH / 'archive/results/board-pilot-sept8')
+    parser.add_argument('--results-v2', type=Path, default=BENCH / 'archive/results/board-interface-v2-sept8')
     parser.add_argument('--out', type=Path, required=True, help='Fresh output directory; frozen evidence is never overwritten.')
     args = parser.parse_args(argv)
     v1, root, out = args.results_v1.resolve(), args.results_v2.resolve(), args.out.resolve()

scripts/analysis/token_audit.py

@@ -19,7 +19,7 @@ def summarize(rows):
 
 def main(argv=None):
     parser = argparse.ArgumentParser(description=__doc__)
-    parser.add_argument('--logs', type=Path, default=BENCH / 'logs')
+    parser.add_argument('--logs', type=Path, default=BENCH / 'archive/logs')
     parser.add_argument('--out', type=Path, required=True, help='Fresh output directory for derived analysis.')
     args = parser.parse_args(argv)
     logs, out = args.logs.resolve(), args.out.resolve()

scripts/verify_evidence.py

@@ -11,6 +11,8 @@ def main():
     failures = []
     for entry in manifest['files']:
         path = ROOT / entry['path']
+        if entry['path'].startswith(('results/', 'docs/research/')):
+            path = ROOT / 'archive' / entry['path']
         if not path.is_file():
             failures.append({'path': entry['path'], 'reason': 'missing'})
             continue