L4 · AECO Compliance · v3.4
The Kernel
Not a black box. Not a checklist. A deterministic scoring engine anchored to a transparent, versioned calibration framework, producing audit-grade verdicts for AI use in licensed AEC practice.
22 questions · 7 layers · ACS™ · DCS · FitScore · Methodology v3.4
Watch the kernel work
A single scenario walked through every kernel stage. See how inputs become an audit-grade verdict — deterministic, reproducible, stamped with the methodology version.
Illustrative walkthrough — scenario inputs
Before any scoring happens, HF1–HF5 evaluate whether the tool can even enter the pipeline for this project. Licensure jurisdiction, licensure discipline, E&O coverage minimum, GL coverage minimum, and DBE requirement (on public projects). Any single hard-filter failure is a pass/block gate — the tool never proceeds to scoring.
HF1 licensure jurisdiction · HF2 licensure discipline · HF3 E&O · HF4 GL · HF5 DBE (public projects)
The tool passes through the 22-question rubric distributed across 7 liability layers (L1 Responsible Charge → L7). Each answer combines with the layer weights via a deterministic formula — no LLM decides the score. Explicit weights, explicit calibration constants, same inputs always produce the same output.
22 questions · 7 layers · deterministic rubric · ACS 0–10 output
The Data Confidence Score derives from evidence completeness across the layers. Verdict thresholds are explicit, versioned, and derived from a transparent calibration framework — auditable, not arbitrary. Same inputs always produce the same band, and the threshold set itself is stamped as part of the methodology.
DCS 0–100% · GREEN / AMBER / RED bands · versioned calibration framework
ACS + DCS + Stamp-Safe classification roll into an artifact stamped with the methodology version, timestamp, attesting L1 professional, and project ID. That artifact is what a license board, an E&O carrier, or a procurement reviewer receives — evidence-grade, reproducible, timestamped.
ACS + DCS + Stamp-Safe verdict · methodology version · L1 attestation · project ID
Verdict — illustrative output
ACS
8.2
AI Compliance Score
DCS
82% · AMBER
Data Confidence Score
Stamp-Safe
Conditional
Verdict classification
Stamped with methodology v3.4 · timestamp · attesting L1 professional · project ID.
Illustrative — not an actual assessment score.
The DNA
The kernel is the moat. Three properties make it irreplaceable — and define why we built it rather than bought it.
Property 1
“Same input → same output. Always.”
The kernel is not a language model. The 22 questions, 7 layers, weights, and calibration constants are explicit, versioned, and auditable. Same tool + project + person produces the same score, every time. That determinism is what makes the audit artifact defensible to an E&O carrier years after the sign-off.
Property 2
“Explicit, principled thresholds — not arbitrary.”
The GREEN / AMBER / RED verdict thresholds are explicit, versioned, and derived from a transparent calibration framework — auditable in every artifact. Same inputs always produce the same band. The threshold set itself is stamped as part of the methodology, so a decade from now a reviewer can still see exactly which calibration fired the verdict.
Property 3
“Methodology v3.4 is stamped on every artifact.”
When the methodology updates, past artifacts still reference the version they were stamped under — frozen. The kernel evolves; past records stay valid against the version they were signed under. Currently on methodology v3.4; each artifact carries the exact version it was scored against.
Read the full methodology to see every question, layer, and calibration constant — or see how the first app on the kernel ships an audit artifact you can hand to your carrier.