uditgoenka/autoresearch v1.9.11 on GitHub

What's New

Mandatory numeric validation — extracted values must match ^-?[0-9]+\.?[0-9]*$ before any decision logic runs
metric-error status — new iteration status for non-numeric extraction failures
Two-consecutive-error halt — stops the loop (even unbounded) when the verify pipeline is confirmed broken
Diagnostic output — shows raw verify output on failure so the problem is visible
Whitespace trim — strips leading/trailing whitespace before validation
macOS grep compatibility — uses grep -oE instead of grep -oP throughout

Unbounded mode safety net — tracks best_metric and iterations_since_best to detect when the loop burns tokens without real progress
Configurable patience — Plateau-Patience: N (default 15), or Plateau-Patience: off for overnight runs
Three-option pause UI — stop / continue with reset patience / change strategy
Measured-only counting — skips no-op, crash, metric-error, hook-blocked iterations
Bounded mode unaffected — iteration limit already bounds the run

Three-state detection — dirty tree (uncommitted edits), unverified experiment (committed but not verified), clean state
Automatic recovery on resume — discards uncommitted changes, reverts unverified experiments, resumes normally from clean state
Phase 4 safety — cleans up staged changes if git commit itself fails (disk full, hook timeout, permissions)
Simplicity override removed — subjective "complexity" and "simpler" judgments eliminated from Phase 6; the metric alone decides

Threshold-based guards — Guard-Direction + Guard-Threshold params for numeric tolerance instead of binary pass/fail
Backward compatible — without the new params, guards operate in pass/fail mode as before
guard-metric column — new TSV column tracks guard-metric values for drift visibility over time
Plan wizard integration — line-count guard and metric-valued guard options folded into Phase 4.5

Improved dry-run validation — shows raw output, extracted value, numeric check, and troubleshooting table
Common failure fixes table — documents fixes for 85.2%, 342ms, empty output, wrong awk field, multi-line output
Guard options expanded — line-count bloat guard + metric-valued threshold guard added to Phase 4.5

None. All changes are backward compatible. Existing configs work without modification.

140eca2 fix: add metric extraction validation to prevent garbage-data decisions
be54359 fix: add whitespace trim and macOS grep compatibility
ff3eea7 feat: add plateau detection to unbounded mode
0c67f58 fix: add best_iteration tracking and broaden halt condition references
1b17ad0 feat: add session crash recovery, metric-valued guards, remove simplicity override