AI RadarEvidence Portfolio

AI Radar Evidence Portfolio

Designing and building a governed AI system that makes evidence, constraints, and human judgment explicit.

AI Radar is a live system I independently designed and built to turn fragmented AI signals into reviewable intelligence. This portfolio shows where governance judgments changed the implementation—and which claims the available evidence can and cannot support.

Live product and verified implementation trace

The product is deployed. This boundary is inspectable.

The live link proves that AI Radar is deployed. The trace below follows a real Project Takeaway request through the public route, source-aware policy, and tests that enforce the boundary.

Ordinary Project Takeaway candidate creationCode-traced · Test-supported
Real enforced case

An unverified manual entry cannot enter the ordinary creation path.

Scope: ordinary Project Takeaway candidate creation only.

  1. Route receives candidate requestprojects.py
  2. Source-aware policy classifies eligibilityproject_takeaway_candidate_policy.py
  3. Policy failure returns HTTP 400Persistence is not called
  4. Tests preserve the ordinary-path boundaryExplicit override remains separate and audited

What each artifact proves

Live product
Deployment and accessible machinery
Implementation trace
A narrow rule is enforced before persistence
Tests
The ordinary path and explicit exception remain distinct
Product outcome
Not claimed here; a real output still requires human approval for provenance, redaction, and publication scope
Open the deployed AI Radar

Governance narrative

Three places where a governance system can fail

  1. IngestionWhat may enter?Notebook #5
  2. EnforcementWhat actually binds?Notebook #6
  3. AdmissionWhat does approval prove?Notebook #9

Selected evidence

Three judgments, traced to what can actually support them

Each case starts in plain language; the deeper reasoning and evidence boundary remain available for inspection.

Notebook #5Ingestion

Three Borrowed Shells

I stopped three plausible conclusions from carrying more weight than their evidence allowed.

Evidence-aware AIProduct judgmentQuality diagnosis

An external insight, a source match, and a clean metric each tried to carry a stronger conclusion than the evidence allowed. I rejected additive scope, replaced token overlap with traceable source-span grounding, and audited 175 claims. The audit returned zero quoted claims, but mostly summary-level evidence packs meant the number could not support a verdict about the whole product. Relevance is not provenance, and a real symptom is not automatically a diagnosis.

Relevance ≠ admissionOverlap ≠ provenanceSymptom ≠ diagnosis
Evidence tierPublic reasoning + repository ADRs + bounded audit claim

Boundary: The bounded source-excerpt policy was a downstream consequence later admitted through ADR-0011, not part of the original three-shell judgment.

Notebook #6Enforcement

A Constraint Is Only as Strong as Its Layer

I turned a written rule into an enforced boundary that prevents an unverified manual entry from becoming an ordinary Project Takeaway.

Agent governanceEnforcement architectureSystem design

The repository path is code-traced and test-supported: an ordinary Project Takeaway request is classified and evaluated before persistence, and policy failure returns HTTP 400. The boundary is deliberately source-aware rather than one-size-fits-all. Knowledge convergence remains review context; insufficiently verified signal completion is marked unverified; and manual override requires a separate, explicit, audited exception.

1Documented
2Reviewed
3Constructed
4Sandboxed
Evidence tierCode-traced + test-supported

Boundary: This claim is limited to ordinary Project Takeaway candidate creation. It is not a claim that every invalid write or downstream commitment is blocked.

Notebook #9Admission

A Gate You Can Pass Without Understanding Isn't One

I found a comprehension gate that could be passed without comprehension, located the same assumption in my system, and did not rush to build a feature.

Human-in-the-loop evaluationGate diagnosisEpistemic restraint

Reported answer-pattern leakage made an explain-diff quiz passable without demonstrating comprehension. That exposed a distinction in AI Radar: artifact gates can inspect claims, evidence, and inference while still assuming the admitter is competent to judge. Comprehension may resist reliable proxy measurement, so the result remains tracking-only until multiple admitters or operator load makes the failure mode real.

Green quizReported comprehensionGreen quizDemonstrated comprehension
Evidence tierExternal specimen + public reasoning + tracking-only validation record

Boundary: This is a tracked framing gap, not a shipped comprehension feature or a new ADR requirement.

Evidence boundary

What this portfolio proves—and what it does not

AI Radar is a live, independently built system, not a multi-tenant enterprise deployment. It demonstrates end-to-end product, architecture, governance, and evaluation patterns under realistic operating constraints. Evidence is linked at the level available: public reasoning, curated code and ADRs, public implementation traces, and explicit validation records.

Supported here

Not claimed

A working cloud-hosted system exists.

Enterprise production ownership or customer outcomes.

Selected governance claims trace to reviewable artifacts.

External sources endorse AI Radar.

Advisory and unresolved evidence is labelled.

Human approval automatically proves comprehension.

Build and operating context

Built from judgment to operation

ProductSignal, review, project-learning, and manual-intelligence workflows
ApplicationNext.js frontend and FastAPI backend
OperationsAWS-hosted services with documented deployment boundaries
GovernanceADRs, explicit approval gates, candidate policies, and blocked actions
EvaluationClaim verification, composition checks, review records, and calibration

Continue the inspection

Inspect the work, not just the description.