Four tiers. Sequenced honestly.

Tier 1 · sold today

Python AI code teams under external challenge

Mid-market and enterprise engineering teams running Python AI coding agents in production whose patches will be challenged by someone outside the engineering org — a SOC 2 auditor, a customer's procurement officer reviewing an RFP response, a cyber-insurance underwriter, a release manager defending a regression, or in regulated industries, counsel.

Trigger events

• SOC 2 Type II or ISO 27001 audit kickoff cites AI development practices
• Customer trust questionnaire asks "how do you verify AI-written code"
• Cyber-insurance rider underwriting asks for AI E&O posture
• Production regression traces back to an AI-written patch with no audit trail
• RFP for a regulated buyer requires evidence of AI code controls

Verification scenario

Engineering wires Mejepa as a Model Context Protocol server into Claude Code, Cursor, Windsurf, Cline, Continue, or Replit. Each AI-written PR carries a Mejepa verdict packet: Pass/Fail/Abstain, named failure mode on Fail, closest historical exemplar, Docker oracle report.json hash, ed25519 signature. The buyer's auditor or counterparty replays the chain offline with the published public key.

Buy mode: founder-fielded pilot, one Python repo, one agent loop, per-engagement priced .

Tier 2 · sold today (co-sell)

AI coding agent platforms

Claude Code, Cursor, Windsurf, Cline, Continue, Replit. Each platform has an enterprise-credibility ceiling that "AI is shipping our code" creates. Adding "verified by Mejepa" on the verdict path raises the ceiling without the platform having to build the verifier itself.

Trigger events

• Platform's enterprise buyers ask for an independent verifier on the agent's output
• Compliance / governance feature gap blocks an enterprise deal
• Platform wants a defensible answer to "how do we know your agent didn't ship a bug"

Verification scenario

Platform embeds Mejepa's 57 mejepa_* MCP tools into its agent runtime. Verdicts surface inline in the platform's UI; the signed witness chain becomes a platform feature.

Buy mode: co-sell pilot, OEM substrate negotiation deferred until ship-gate green .

Tier 3 · roadmap rung, not sold today

Engineering-risk extensions

Areas where Mejepa's verified architecture extends but calibration and productization are still required. We do not sell these as products today; we accept named-design-partner research engagements with explicit fences .

• Multi-language code verification — Rust, JS, TS, Go, Java, C, C++, C#, Ruby, PHP. Corpus already chunked; per-language parser-backed operators and Docker-oracle equivalents in scope. Deferred until Python ship-gate green.
• AI coding agent observability — pre-merge gate + Stop hook. Capture infrastructure shipped; productization deferred until M4.
• SOC 2 / ISO / FedRAMP evidence packs — templating layer over the witness chain. Witness chain is domain-agnostic, verdict producer is Python today.
• Hosted SaaS for Python AI code verification — multi-tenant calibration is engineering-only risk, but deferred until M4 productization.

Tier 4 · roadmap rung, not sold today

Science-risk extensions

Areas where new frozen instruments and new ground-truth oracles must be designed and proven before Mejepa can sell verification. We do not market against these surfaces today and we decline pilots that conflate them with code verification .

• AI document output verification (citations exist, quotes match) — instrument design for document corpora not started.
• AI underwriting output verification (claim summary vs underlying records) — same instrument-design gap.
• AI clinical output verification — instrument design + regulatory-deployment bar not met.
• AI training-data provenance attestation (paired with Leapable.ai OCR Provenance) — joint witness chain pattern viable; two independent science projects.