Forge Final

Philosophy

Modules must influence the next iteration, not merely exist. Bridge-only provisional ideas are labeled separately so workflow placeholders cannot masquerade as discoveries.

Architecture & control flow

1. 01

   PhysicsTester and the Aurora bridge share coefficient/role-resolved units.

2. 02

   Evolution consumes findings, chooses routed repairs and persists parent-to-child repair traces.

3. 03

   Each conjecture creates eight workstream records and claim provenance; the message bus broadcasts blockers while viability, repairs and failures are saved.

4. 04

   Bridge-only provisional ideas complete T→I before harvesting, and the CLI applies synthetic-prior caps and synchronizes package/JSONL outputs.

Architecture outline derived from this version’s control flow.

Inputs & outputs

Inputs

Forge equations, conjectures, diagnostic findings, synthetic priors and mode configuration.

Outputs

Viability, repair, failure and provisional-idea JSONL, a claim-linked working_paper_v55.md and a reconciled package.

Implemented components

• Live unit-resolution and routing integration, auditable ledgers, workstream-state creation, bridge-only provisional ideas and verdict caps.

Implementation & evidence scope

Four repair operators still fall through to default geometric-diffusion mutation: traces and routing do not establish their full semantic edits. Workstream goals are created and persisted, but real subagent execution and literature backends are not integrated. Numerical witnesses remain Spark’s small ODE probes, not large-scale or 3D PDE solvers.

Code & bundled material

The introduction draws on bundled notes, changelogs and central code. Software tests, synthetic diagnostics and scientific effectiveness use different evidence standards.