Universal Quality Development Harness

The platform's core engine. Nine phases. Self-mode and target-mode. Any codebase.

UQDH is the quality engine at the heart of CODITECT per ADR-320. It runs in two modes: self-mode (CODITECT applies the harness to itself for continuous self-improvement) and target-mode (CODITECT applies the harness to any customer codebase). The harness is parameterised - it discovers the target system's characteristics and generates a tailored quality plan. No hardcoding for any specific stack.

The problem - in Mark Walker's words
"We and every other software company in the world are outstripping our ability to test what we're building."

Why now: the velocity of agentic coding has decoupled from the velocity of testing, auditing, and validation - the knowledge and proof that AI agents did what they were tasked to perform, i.e. testing, in this case. An AI agent can produce more code in a day than a team used to write in a sprint. The test, audit, and compliance layers did not get faster at the same rate. The gap is structural and widens with every model release.

Three deficiencies - in every company today - that no software addresses:

  • determining which tests need to run for a particular release
  • checking whether they ran
  • recording the outcome

Mark Walker, nue.io - meeting transcript [00:46:36]

UQDH is the engine that does all three across any target. Every phase carries inputs, outputs, and standards mappings; every run produces evidence in the audit trail.

The 9-phase harness loop (ADR-320)

flowchart LR classDef phase fill:#e0f2fe,stroke:#0284c7,color:#0c4a6e,font-weight:bold classDef obs fill:#dcfce7,stroke:#16a34a,color:#14532d,font-weight:bold P1[1 DISCOVER] P2[2 ANALYZE] P3[3 PLAN] P4[4 SPECIFY] P5[5 RESEARCH] P6[6 IMPLEMENT] P7[7 REMEDIATE] P8[8 DEPLOY] P9[9 OBSERVE] P1 --> P2 --> P3 --> P4 --> P5 --> P6 --> P7 --> P8 --> P9 P9 -. feedback to next iteration .-> P2 class P1,P2,P3,P4,P5,P6,P7,P8 phase class P9 obs

The two modes

SELF CODITECT applies the harness to itself. The platform runs continuously against its own codebase, surfacing quality gaps, regressions, drift, and standards violations in its own delivery pipeline. The platform's audit trail is itself a UQDH OBSERVE output.

TARGET CODITECT applies the harness to any customer codebase. The DISCOVER phase reads the target's characteristics (language, framework, architecture, infrastructure) and generates a stack-tailored plan. No assumption about the customer's tech stack; no hard-coded paths for any specific framework.

Phases 1-5: understand the target, decide what to do

PhaseInputOutputKey actionsStandards mapped
1. DISCOVERTarget system path, repo URL, or cloud endpointsystem-profile.jsonTreesitter AST language detection, framework identification, architecture pattern detection, test/CI/database discovery, dependency graph, security scan, regulatory surface detection, documentation inventoryISO/IEC 25010 §4.2, IEC 62304 §5.1.1
2. ANALYZEsystem-profile.jsonquality-assessment.jsonMap system to applicable regulatory frameworks, score current state per framework, identify quality gaps, calculate technical debt, blast-radius risk, benchmark against DORA metrics, generate compliance gap matrixISO/IEC 42001 §6.1, NIST AI RMF (GOVERN, MAP)
3. PLANquality-assessment.jsonremediation-plan.json12-dimension critical-path scoring, ≤20-task sprint groups, effort estimates, sprint dependency graph, autonomy level per sprint (L1-L4), rollback plan per sprintISO/IEC 5338 §6.2, CMMI GP 2.2
4. SPECIFYremediation-plan.jsonPer-sprint specifications with acceptance criteriaAgent-executable system prompts, test criteria, deploy gates, traceability matrix (task → requirement → standard), Given/When/Then acceptance criteria, rollback proceduresIEC 62304 §5.2, ISO 13485 §7.3.3
5. RESEARCHSprint specifications + tech contextImplementation guidance with evidence-based recommendationsSearch best practices for detected stack, dependency CVE search, framework-specific patterns, relevant ADRs, domain regulatory guidanceNIST AI RMF (MAP), ISO/IEC 42001 §7.2

Phases 6-9: execute, fix, ship, learn

PhaseInputOutputKey actionsStandards mapped
6. IMPLEMENTSprint spec + research guidanceCode changes, config updates, documentationExecute sprint system prompt (agent or human), apply STD-IMPLEMENTATION-LIFECYCLE, syntax/compile checks per change, mark tasks complete, generate diffs and commit hashes as evidenceIEC 62304 §5.5, ISO 13485 §7.3.4
7. REMEDIATETest or quality-gate failuresFixed code, updated tests, root-cause analysisParse error output, trace root cause, minimal fix only, re-run failing + regression suite. After 3 iterations: escalate with blocking note. Root-cause analysis per failure for post-sprint reviewISO 13485 §8.5.2, ITIL Incident Management
8. DEPLOYTested, reviewed code changesProduction deployment with validationPre-deploy gate checklist, platform-appropriate deploy, post-deploy smoke tests, 15-minute SLO-watch with auto-rollback, ITIL change record, CMDB updateITIL Change Enablement, ISO 20000 §8.5.1, SOC 2 CC8.1
9. OBSERVEProduction metrics, quality data, sprint outcomesQuality report + feedback into next iterationDORA metrics (deploy frequency, lead time, MTTR, change failure rate), sprint velocity and accuracy, quality trends, process-improvement candidates, feedback to Phase 2, compliance evidence artifacts, dashboard updateISO/IEC 42001 §9.1, NIST AI RMF (MEASURE), CMMI GG 3

Universal applicability - how DISCOVER adapts the harness

Phase 1 reads the target codebase and generates a stack-specific quality plan in subsequent phases. The same harness produces different gates depending on what it finds:

Detected stackHarness adaptation
Python / Djangopytest gates, ast.parse checks, Django migration safety
TypeScript / Reacttsc --noEmit, npm run build, bundle-size gates
Rustcargo test, cargo clippy, MSRV enforcement
Gogo test, go vet, golangci-lint gates
Java / SpringJUnit, Checkstyle, SpotBugs gates
NestJS / Nodejest, eslint, Prisma migration gates
Mobile (Swift / Kotlin)XCTest / JUnit, screenshot testing gates
Infrastructure (Terraform)tflint, terraform validate, plan review gates
Documentation (Markdown)markdownlint, link verification, frontmatter validation
MonorepoPer-package gates, affected-package detection
MicroservicesContract testing, service-mesh health, distributed tracing
DatabaseMigration safety, schema-drift detection, data-integrity checks

Why UQDH closes Mark's three deficiencies

Absence 1 - which tests must run. Phases 1-4 (DISCOVER, ANALYZE, PLAN, SPECIFY) generate per-change test selection from the live system profile, the gap analysis, and the SPECIFY traceability matrix that links every task to a requirement and a standard.

Absence 2 - that they ran. Phases 6-7 (IMPLEMENT, REMEDIATE) execute the planned tests, capture pass/fail and root-cause for each. Phase 8 (DEPLOY) gates production on pre-deploy and post-deploy verification.

Absence 3 - recording the outcome. Phase 9 (OBSERVE) collects DORA metrics, sprint accuracy, quality trends; the audit trail of every prior phase is the recorded outcome the regulator queries.

Standards covered in one engine

UQDH integrates ITIL, ISO, IEEE, and NIST controls into a single quality engine. Every phase is mapped to specific clauses (see the phase tables above). The standards are not bolted on; they are the gates and outputs of the loop.

  • ISO/IEC 25010 (software quality model) - DISCOVER, ANALYZE
  • ISO/IEC 5338 (software lifecycle processes) - PLAN
  • ISO/IEC 42001 (AI management system) - ANALYZE, RESEARCH, OBSERVE
  • NIST AI RMF (GOVERN, MAP, MEASURE, MANAGE) - ANALYZE, RESEARCH, OBSERVE
  • IEC 62304 (medical software lifecycle) - DISCOVER, SPECIFY, IMPLEMENT
  • ISO 13485 (medical device QMS) - SPECIFY, IMPLEMENT, REMEDIATE
  • ITIL Change Enablement and Incident Management - DEPLOY, REMEDIATE
  • ISO 20000 (IT service management) - DEPLOY
  • SOC 2 CC8.1 (change management) - DEPLOY
  • CMMI GP 2.2, GG 3 (process discipline) - PLAN, OBSERVE

The mapping is recorded in CODITECT's Research Foundations bibliography for every clause cited.

UQDH and the rest of the platform

UQDH is the central engine; every other surface on this site is one expression of it.