Development record
Changelog.
Concrete changes, known limitations, and what the evidence actually establishes.
NOV 4, 2026Prototype development
Evidence before confidence
- Live HTTP and connection failures stop execution instead of producing simulated results.
- Failed or invalid live judging stops evaluation. Unevaluated results start as unknown.
- Internal campaign tokens are no longer forwarded to target endpoints.
- Live exploration errors no longer substitute mock payloads.
- JSON, HTML and SARIF reports carry execution-mode labels.
- A pinned dependency snapshot installed into a fresh Linux Python 3.15 environment. The latest full suite passed 58 tests with 285 warnings.
Identity and next steps
Mallace’s selected palette is charcoal and crimson. A ForgeLoop integration proposal is documented. Production access controls, durable operations, live acceptance, the operator dashboard, and the connector remain incomplete.