Design Fidelity Metrics
29 metrics. Five layers. One score for the board.
Each metric has a formula, thresholds, the people who read it, and the ways it misleads. Start with DHS or pick your layer.
L4 · The headline numberDHSDesign Health Score. One composite for the board, built from four conformance metrics.How DHS is built →
L1 · 4 metricsIntent HealthIs our intent written down, deep and current?
L2 · 5 metricsPillar HealthIs each domain holding, improving or slipping?
L3 · 11 metricsConformance HealthDo PRs follow the intent and the process?
L4 · 1 metricsDesign Health ScoreOne number for leadership.
L5 · 8 metricsAgentic Workflow HealthAre agents governed as well as people?
| Abbr | Metric | What it tells you | |
|---|---|---|---|
| L1Intent Health4 metrics | |||
| SCS | Steering Completeness Score | Are all 7 governance artifact categories present? | → |
| GD | Governance Depth | How many architectural rules are machine-readable and enforced on every PR? | → |
| SF | Steering Freshness | Is governance current or quietly abandoned? | → |
| VPA | Vision-Pillar Alignment | Are pillars connected to a stated purpose? | → |
| L2Pillar Health5 metrics | |||
| PCR | Pillar Conformance Rate | Pass rate of Lens evaluations per architectural domain. | → |
| PDT | Pillar Drift Trend | Is a domain getting better or worse sprint-over-sprint? | → |
| PDRP | PR Degradation Rate per Pillar | Is drift concentrated in one team or systemic? | → |
| LPP | Lenses per Pillar | How deeply is each domain covered? | → |
| PSI | Pillar Staleness Index | Days since a pillar was last updated. | → |
| L3Conformance Health11 metrics | |||
| DCS | Design Conformance Score | % of PRs that passed all Lens checks. The headline metric. | → |
| DPAR | Design Process Adherence Rate | % of audited PRs with an approved Change Intention linked to the work. | → |
| IFR | Intent-First Rate | % of PRs with design approved before the first commit. | → |
| UDR | Ungoverned Drift Rate | % of structural changes with no approved Change Intention. | → |
| CLPR | CHI-less PR Rate | % of audited PRs with no Change Intention, pass or fail. | → |
| ABR | Audit Block Rate | % of audits that actively blocked a PR from merging. | → |
| MTTCD | Mean Time to Change Decision | Average time from blocked PR to resolution. | → |
| ADI | Architecture Drift Index | Normalised 0.0–1.0 measure of drift severity. | → |
| PRE | PRs Evaluated | Count of design audits triggered by GitHub PR events. | → |
| LPPR | Lenses Per PR | Average number of Lenses evaluated per design audit. | → |
| TLA | Total Lenses Applied | Cumulative Lens evaluations across all design audits. | → |
| L4Design Health Score1 metric | |||
| DHS | Design Health Score | Weighted composite of DCS, ADI, IFR, UDR. The board-reportable metric. | → |
| L5Agentic Workflow Health8 metrics | |||
| AHCR | Agent vs Human Conformance Rate | Differential: do agents conform better or worse than humans? | → |
| ADRIFT | Agent-Introduced Drift Rate | % of agent PRs that produced ungoverned drift. | → |
| AIFI | Agent Intent-First Index | % of agent PRs with CHI approved before first commit. | → |
| MCPCAR | MCP Context Access Rate | % of agent PRs where rkito MCP was queried before coding. | → |
| LSCR | Lens Satisfaction Before Commit Rate | % of agents that loaded Lenses and produced zero violations. | → |
| AFPR | Agent False Positive Rate | % of Lens findings on agent PRs dismissed by an architect, vs. human PRs. | → |
| MACCR | Multi-Agent CHI Collision Rate | % of concurrent agent sessions with overlapping CHI scope. | → |
| AGVR | Agent Governance Bypass Rate | % of agent PRs merged with no CDA evaluation. | → |
Put metrics to work
Guide · PlaybookCTO Architecture Health PlaybookA CTO's weekly check on architecture health, board-ready DHS reporting, AI adoption governance via ADRIFT, and knowing when to escalate.Guide · PlaybookVP Engineering PlaybookHow VPs of Engineering track sprint conformance, team-level drift, and quarterly process health using DCS, IFR, PDT, and PDRP signals.ConceptArchitectural DriftThe silent, incremental divergence between agreed architectural intent and what actually ships — now accelerating as AI coding tools multiply PR volume beyond review capacity.