STI as governed test intelligence

Core thesis

STI stands out because it treats test intelligence as a governed, contract-bound extension of the Forge control plane — not a chatbot that writes pytest files.

The core message is: test cases need contracts, deterministic oracles, and traceable evidence — especially when agents author them.

Condensed thought

Forge STI layers test strategy through campaigns using lmeta orchestration and LCDL judgment where decisions matter. Humans and agents edit the same JSON/Markdown artifacts. Execution stays on Fleet or local runners; review stays on Lenses.

Why it stands out

Generic AI testing tools optimize for speed of snippet generation. STI optimizes for release confidence: RTM links, bounded campaign flows, verification separate from model fluency, and ForgeRun evidence.

Forge ecosystem hooks

• Schemas — test_case.v1, test_suite.v1, test_run_summary.v1
• lmeta — campaign flows with extends/overrides for smoke vs release-gate variants
• LCDL — case design, triage, coverage tasks (planned)
• Cursor — composer prompts and versona-testing pairing
• Fleet — approved test templates
• Lenses — evidence and approval UI
• Blueprints — ISTQB methodology and versona-testing

Maturity honesty

Defined — architecture, schemas, handbook, Cursor kit stubs. Not yet demonstrated — runnable LCDL task pack, lmeta campaign library, Fleet template catalog, Lenses panels.