METHODOLOGY · OPEN
P.R.O.D. - four dimensions of production AI.
Most AI projects don't fail at the model. They fail at the four places production reality intersects with a demo. P.R.O.D. is a deliberately small framework - four dimensions, each scored 1-5 - that makes the gap visible.
P
Production Reality
Real process, named owner, measurable outcome.
1
Notebook - Works once, by hand.
2
Internal demo - Repeatable, no real users.
3
Pilot - Real users, narrow scope.
4
Scoped production - Owned, monitored, billable.
5
Cross-org production - Survives a Q-end peak.
R
Reliable Data
Accessible, clean, documented, permissioned, fresh-enough.
1
Ad-hoc CSV - One person knows where it is.
2
Brittle ingestion - Breaks on schema drift.
3
Pipeline + lineage - Documented but unowned.
4
Owned + observed - Drift alerts, freshness SLOs.
5
Audit-grade - Permissioned, retention-policied, replayable.
O
Operational Control
Evals, monitoring, guardrails, cost, incident response.
1
None - We'll add evals later.
2
Manual spot-check - Smoke test before deploy.
3
Eval suite exists - Runs ad-hoc, not in CI.
4
Evals + cost in CI - Gates the release.
5
Drift + incidents - Runbook, on-call, SLOs hit.
D
Delivery Architecture
Documented, scalable, deployable, rollback-able, secure.
1
Single notebook - Lives on one laptop.
2
Ad-hoc service - No rollback, single region.
3
Documented + deployed - Diagrams exist, rollback ≤ 1 hr.
4
Resilient - Multi-region, rollback ≤ 10 min.
5
Hardened - Pen-tested, certified, audited.
SCORING
The gap matters more than the score.
A 4-3-4-5 stack is more dangerous than a 3-3-3-3 one. The first ships before its weakest dimension catches up. The second stays in the lab. Treat the lowest dimension as the ceiling.
RUN IT
Eight minutes for a score. Ten days for the artifacts.
Self-score with the public assessment, or have us run the full audit with you.