PiPER incremental hold v0.3 — bounded comparison

The bounded incremental hold reaches its correction limit after 320 ms in every admitted scene. The fresh 32-attempt comparison retains all failures: none of the four controllers achieves a five-second stationary grip while remaining controller-qualified. The candidate’s one native stationary-grip outcome includes terminal fallback; no controller is promoted.

This source preview adds the candidate, fixed CLI benchmark, independent saved-record audit, reproducible archive and report. EGM-inspired fresh-context review caught a joint-margin clipping counterexample before execution; Harness² informed the propose/test/retain-lessons workflow. Neither upstream framework was installed or run.

• Source freeze: 3f30847. Publication tools: 569e8d3.
• 32/32 valid; 640,000 native rows, 32,000 issued commands; all 8 shared acquisition prefixes independently verified.
• Validation: 102 controller/runner regression tests, 13 audit tests, 8 package tests, 6 publisher tests, and actual browser report inspection.
• Native stationary-grip outcomes: measured 0/8, issued 0/8, Cartesian 0/8, incremental 1/8. All active stationary-grip outcomes 0/8.
• No GPU or API key needed for local CPU reproduction. This tests authored simulator-state controls, not learning, videos, transport, full placement or hardware.

Reproduction guide · Research note · Hugging Face evidence

Archive SHA-256: 5129187d14f8d4ad691a69e3554cff2b5ba65489a19c1583609c767b36f42c44. Original DVIDIA code and synthetic records MIT; upstream notices included.