NRP Test Documentation

The NRP tests are grouped under tests/NRP. They protect the mesh resolver, scoring engine, observability loop, and the parser compatibility fixes that make the bridge work with cleaker v3.

This document mirrors the exported NRP_TEST_CATALOG, so TypeDoc also includes the test taxonomy.

---

Test Groups

File	Category	Purpose
tests/NRP/bridge.parse.test.ts	parsing	Verifies me:// parsing through cleaker v3 __ptr.target, including dot paths like .mesh/monads
tests/NRP/monadIndex.test.ts	index	Verifies local monad index read/write/list, claimed namespace lookup, name lookup, and ordering
tests/NRP/meshSelect.test.ts	selection	Verifies selector matching, stale filtering, self-exclusion, runner-up tracking, and exploration
tests/NRP/scoring.test.ts	scoring	Verifies score invariants, learning loop, introspection breakdowns, and scoring integration
tests/NRP/decisionLog.test.ts	observability	Verifies decisionId correlation, reward calculation, JSONL output, and failure isolation
tests/NRP/adaptiveWeights.test.ts	learning	Verifies global/namespace weight updates, maturity blending, and learning-loop integration

---

Hard Invariants

The scoring suite explicitly protects these invariants:

• normalized scores stay in [0, 1]
• identical inputs produce identical scores
• NaN and Infinity do not propagate
• scaling all weights by the same constant does not change the score
• computeScore and computeScoreDetailed.total stay identical
• runner-up is present only when more than one claimant exists
• decision correlation uses decisionId, not monadId:namespace
• reward stays in [-0.7, 1]
• namespace weights are not created on read
• namespace maturity reaches 70% blend at 140 samples
• global prior still receives 5% background learning at full maturity

---

Why Tests Are Also in TypeDoc

Runtime test files are not exported as public API. Instead, the production module exports NRP_TEST_CATALOG, a typed catalog of the NRP test groups. This keeps the package runtime clean while letting generated API docs show the coverage map.

See:

import { NRP_TEST_CATALOG } from "monad.ai";

---

Commands

# Full suite
npm test

# Focus a test group
npx vitest run tests/NRP/scoring.test.ts
npx vitest run tests/NRP/decisionLog.test.ts
npx vitest run tests/NRP/adaptiveWeights.test.ts

# Build docs API
npm run docs:api

---

Legacy/Non-NRP Tests

The older tests are still relevant. They protect the base semantic runtime:

File	Area
appFactory.test.ts	app bootstrap and env isolation
claimSemanticSeeds.test.ts	claim semantic seeding
claimsOpenProfile.test.ts	opened claim profile hydration
commitsync.test.ts	memory ledger and replay smoke tests
hostProjection.test.ts	host memory projection
hostResolver.test.ts	canonical host-to-namespace projection
observerRelation.test.ts	observer relation and disclosure envelopes
persistentClaim.test.ts	claim persistence and verification
replayCanonicalization.test.ts	replay payload normalization
rootUsersProjection.test.ts	root namespace user projection
selfMapping.test.ts	self identity, selector parsing, and surface entries
semanticBootstrap.test.ts	semantic bootstrap idempotency
semanticBranchReader.test.ts	semantic branch prefix isolation

tests/claim_test_verification.ts remains a manual script invoked by npm run test:claims.