PILOT WAITLIST · COHORT 1 · Q1 2027

Get your AI coding tool scored on real metrics — before your CFO asks how you justified the spend.

An independent, pre-registered audit of how Claude Code, Cursor Agent, Replit Agent and Google Antigravity perform against your team's actual spec — across five quality dimensions, ten replications, with statistical confidence intervals.

The headline from the public benchmark

The auditor's main study scored four AI coding agents against three task specifications. 600 measurements. Pre-registered protocol. Live dashboard. Some findings:

CLAUDE CODE
0.00
off-spec features per run (30 runs). Tightest discipline.
REPLIT AGENT
1.00
hallucinated CLI commands per run. Ships its own architectural defaults regardless of spec.
CURSOR AGENT
0.17
off-spec endpoints per run. Occasional helpful overreach (/health, /metrics).
ANTIGRAVITY (GEMINI)
0.33
off-spec endpoints per run. Concentrated in web-app specs.

Read these as the protocol recorded them. Claude Code and Cursor Agent contribute distinct captures per replication. Replit Agent and Antigravity are IDE-bound: under Deviation 001 each was captured once per cell and replayed, so their figures describe one observed run per specification, not thirty independent ones. The deviation is in the pre-registered protocol, not discovered after the fact.

→ Browse the live dashboard

What the pilot audit gives you

01
Five-dimension score (security, complexity, duplication, hallucinations, effort)
02
Four-vendor comparison on YOUR spec, not a generic benchmark
03
Statistical tests (ANOVA + Tukey) with 95% confidence intervals
04
Per-spec breakdown so you see condition × task interactions
05
Written PDF report (~10 pages) + private dashboard URL (12 months)
06
Raw CSV + provenance JSON for your own analysis

How a pilot engagement runs

Day 0
You hand me a YAML or Markdown spec of the feature set your team is asking AI tools to build.
Day 0–3
Spec ingestion + a 30-min clarification call.
Day 3–7
Capture runs: 10 replications × 4 vendors × your spec, scored under the pre-registered protocol.
Day 7–9
Statistical analysis + report drafting.
Day 10
Delivery — PDF report, private dashboard, CSV. 30-min walkthrough call included.
Important — pilot timing. I am the sole operator of this instrument, currently completing my MSc dissertation at Aston University under student visa. Commercial pilot engagements begin Q1 2027, once the relevant work-rights transition completes. Joining the waitlist now reserves your slot at the cohort-1 launch price (£2,000 per pilot, no payment until delivery).

Join the Cohort 1 waitlist

No payment. No commitment beyond a 20-min discovery call when slots open. I'll prioritise teams whose spec interests me most for the cohort.

Submissions are held privately by O’Rume Dominic Uririe. No marketing list, no third-party sharing. You can opt out at any time by replying to my outreach email. By submitting you confirm you have authority to engage a vendor on behalf of your organisation for an exploratory call.

Live dashboard Open-source repo PyPI package Pre-registered protocol