Testing loop

First formal introduction 2026-08-08 — cycle CLOSED. Reload the founding conversation. Check the stack. Fix what is fixable. Red-team. Continue only after. Company cadence below.

Ethos fit

Founding demand: frontier-level adversarial pressure on the model and the transparency layer. “Remember you are on drugs.” Soft tissue is green without measurement. The original conversation is the load path — not optional flavour text.

How the first cycle ran (re-run this)

Stop feature thrash
Reload SoT — founding/conversation/ + PROVENANCE
Double-check automated suite (command below)
Triple-check soft tissue (version, seals, nano≠frontier, gates honest)
Fix agent-fixable issues; re-measure
Red-team integrity (RTC) + business (BHA) with real code
File docs/audits/loops/LOOP_YYYY-MM-DD.md · mark CLOSED
Continue free-core only after

Normative: docs/security/TESTING_LOOP.md §8–9 · first record: docs/audits/loops/LOOP_2026-08-08.md

Cadence

Every free-core claim ship — full automated suite
Weekly while building hard — suite + soft-tissue scan
Monthly — suite + Domain 1 audit skim
Quarterly — full loop + campaign plan
Pre-fundraise / first SKU — full loop + BOUNDARY refuse drill
After integrity incident — suite + targeted red-team

Minimum commands (automated half)

python3 scripts/run_testing_loop.py
# expanded:
python3 scripts/oneshot_verify_all.py
python3 -m pytest -q
python3 scripts/redteam_nano_harness.py
python3 scripts/redteam_business_attack.py
python3 scripts/ttllm_status.py --quiet-ok

Hard gates · Placeholders · Status · Demo / inclusion proof