Historical record: Gray–Scott counterexample

Note

This is a preserved version 0.1 classic-workflow record. For the current minimal workflow, start with the orbital investigation and research-loop guide. The original numerical artifacts and decisions below have not been rewritten.

This completed 23.8-minute campaign is the primary version 0.1 autonomous demonstration. The agent received one natural-language hypothesis and no campaign instruction. It selected its own mathematics, solver, parameters, diagnostics, daughter hypothesis, prospective evidence contracts, and stopping condition.

At fixed reaction parameters, whether a two-dimensional Gray-Scott
reaction-diffusion system develops a persistent spatial pattern is determined
solely by the diffusion ratio D_u/D_v, independent of the absolute diffusion
scale.

The agent rejected the installed WarpX capability as inappropriate, authored a NumPy pseudospectral reaction-diffusion solver, and found a finite-domain counterexample. With F=0.072, k=0.062, L=40, and D_u/D_v=3 fixed, the s=1 case retained a pattern while the s=10 case decayed to homogeneity.

The contract evidence changes only the common diffusion scale. Both timestep resolutions lead to the same disposition.

Hypothesis graph and guarded closure

The campaign created one scientific daughter. The child was supported and the root was falsified. When the agent first tried to use child-contract evidence to close the root, the harness refused because the root lacked its own prospective evidence contract. The agent then registered the missing contract, ran a fresh four-case experiment with different perturbation phases, and closed the root with qualifying evidence.

× claim_root                         falsified
└─ ✓ claim_finite_gs (refines)       supported

The present interface attached read-only to the preserved run. Durable JSON and JSONL files, rather than the interface, remain authoritative.

Audit or replay

The repository includes the complete transcript, portable claim ledger, final report, provenance map, all 22 generated workspace artifacts, final fields, and the exact decisive program under demos/gray_scott_counterexample/.

uv run simjecture status demos/gray_scott_counterexample/record
uv run python demos/gray_scott_counterexample/verify_record.py
uv run simjecture web demos/gray_scott_counterexample/record --read-only
uv sync --extra tui
uv run simjecture tui demos/gray_scott_counterexample/record

This is evidence for autonomous, evidence-governed computation within a specific numerical model. It is not evidence that arbitrary hypotheses can be solved reliably, and it is not plasma validation.