Verification Matrix: Claims vs. Automated Evidence

September 2, 2026 · View on GitHub

This document provides a comprehensive, rigorous traceability matrix between every architectural claim made by dsh-continuum and its automated verification test.

1. System Invariants & Verification Traceability Matrix

Architectural Invariant / Claim	Implementation Source	Automated Test Suite	Test Location	Result
Strict Sequence Monotonicity Events are assigned strictly incrementing integer sequences ( $seq_{n+1} = seq_n + 1$ ). Non-positive/non-integer sequences rejected.	`src/core/events.ts` `src/domain/engine.ts`	`Core Events and Monotonicity`	`tests/unit/events.test.ts`	PASSED
Sub-Millisecond Event Determinism Events occurring in the exact same millisecond are deterministically ordered by unique monotonically incrementing sequence numbers.	`src/core/events.ts` `src/domain/state.ts`	`Core Events and Monotonicity`	`tests/unit/events.test.ts`	PASSED
Optimistic Concurrency Control (OCC) Entity mutations require `expectedVersion`. Concurrent writes with stale version throw typed `VersionConflictError`.	`src/domain/engine.ts`	`Optimistic Concurrency Control`	`tests/unit/concurrency.test.ts`	PASSED
Strict Idempotency with Mismatch Detection Replaying identical mutation returns cached result (`isCachedResult: true`). Reusing key with different entity/action throws `IdempotencyMismatchError`.	`src/domain/engine.ts` `src/storage/filesystem.ts`	`Strict Idempotency with Mismatch Detection`	`tests/unit/idempotency.test.ts`	PASSED
Task State Machine Invariants Tasks strictly follow permitted state transitions (`queued` $\to$ `ready` $\to$ `claimed` $\to$ `running` $\to$ `submitted` $\to$ `verification` $\to$ `completed`/`rejected`). Illegal jumps throw `InvalidStateTransitionError`.	`src/domain/tasks.ts`	`Task State Machine Invariants`	`tests/unit/state-machine.test.ts`	PASSED
Renewable Execution Leases Claiming a task issues a renewable time-bound lease. Expired leases without renewal are automatically swept and returned to `ready` state with attempt counter incremented.	`src/domain/engine.ts`	`Task Execution Leases`	`tests/unit/leases.test.ts`	PASSED
Separation of Duty (4-Eyes Principle) When `verificationPolicy` is `independent` or `dual`, the implementer/claimer principal is forbidden from self-verifying. Violations throw `SeparationOfDutyError`.	`src/domain/tasks.ts` `src/domain/engine.ts`	`Separation of Duty Verification Enforcement`	`tests/unit/separation-of-duty.test.ts`	PASSED
Deterministic Context Budgeting Context Pack adheres to strict token limit ( $N$ tokens) using deterministic 10-tier priority ranking. Active task, North Star, and unresolved conflicts receive top priority.	`src/context/budget.ts` `src/context/ranking.ts` `src/context/builder.ts`	`Context Budgeting & Deterministic Ranking`	`tests/unit/context-budget.test.ts`	PASSED
Untrusted Content Isolation Third-party outputs, findings, and evidence are encapsulated in strict boundaries (`<continuum:untrusted-finding>`). Breakout tags are neutralized.	`src/context/security.ts`	`Untrusted Content Isolation & Data Delimiting`	`tests/unit/security.test.ts`	PASSED
Property-Based Invariants (fast-check) 100 randomized executions verify: (1) Sequence numbers are monotonically increasing; (2) Event replay is completely deterministic; (3) Snapshot + tail fold is strictly equivalent to full fold.	`src/domain/state.ts` `src/core/events.ts`	`Property-Based Invariant Verification (fast-check)`	`tests/unit/property.test.ts`	PASSED
Crash Recovery & Partial Line Truncation Storage detects incomplete JSON fragment caused by power loss or SIGKILL at the tail of `events.jsonl`, truncates the partial fragment cleanly, and permits subsequent appends.	`src/storage/filesystem.ts`	`Storage Integration & Crash Recovery`	`tests/integration/storage.test.ts`	PASSED
Cooperative File Locking with PID Liveness Acquires cooperative write lock. If existing lock is held by a dead process, tests `process.kill(pid, 0)`, detects process death (`ESRCH`), breaks stale lock safely, and acquires cleanly.	`src/storage/locking.ts`	`Cooperative File Locking & Liveness Detection`	`tests/integration/locking.test.ts`	PASSED
Agent-Scoped Capability Enforcement & Zero Leakage Tool restrictions applied to Agent A via `ctx.tools.restrict` do NOT leak to concurrent Agent B. Accurately distinguishes policy expressed vs. technically enforced when shell tools (`bash`) exist.	`src/capabilities/pack.ts` `src/capabilities/enforcer.ts`	`Agent-Scoped Capability Enforcement & Isolation`	`tests/integration/capabilities.test.ts`	PASSED
Cryptographic Backup & Verified Restore Exports complete project snapshot and event log into `.continuum-backup` bundle with SHA-256 digest. Rejects tampered files with `checksum mismatch` error. Faithful state restoration.	`src/storage/migration.ts`	`Complete Backup & Verified Restore`	`tests/integration/backup.test.ts`	PASSED
Large Artifact Spilling (10KB Threshold) Evidence content exceeding 10,000 bytes is automatically spilled to disk artifacts (`artifacts/ART-EVD-*.bin`), replacing the event payload with a 1,000-character snippet and SHA-256 hash.	`src/domain/engine.ts` `src/storage/filesystem.ts`	`End-to-End Multi-Agent Project Lifecycle & Continuity`	`tests/e2e/continuity.test.ts`	PASSED
Complete End-to-End Multi-Agent Lifecycle Simulates full collaboration across 4 agent roles (Orchestrator, Researcher, Implementer, Reviewer), red-team rejection, implementer remediation, approval, crash recovery, and context regeneration.	Entire subsystem stack	`End-to-End Multi-Agent Project Lifecycle & Continuity`	`tests/e2e/continuity.test.ts`	PASSED
DSH Web UI Plugin Settings Integration Mounts namespace `dsh-continuum` with Schemastery schema into DSH `settings` service. Dynamically syncs edits from Web UI and `~/.dsh/settings.yaml` without requiring a process restart.	`src/dsh/settings.ts` `src/dsh/service.ts`	`DSH Web UI Plugin Settings Integration`	`tests/unit/settings.test.ts`	PASSED
Marketplace Distribution Package `npm pack` produces clean tarball containing compiled JS, types, license, and notice, without test or development bloat. Installs and imports cleanly in isolated consumer projects.	Build & packaging pipeline	`Package Smoke Test`	`npm pack` + clean isolated runner	PASSED

2. Test Execution Summary

> tsx --test 'tests/**/*.test.ts'

ℹ tests 35
ℹ suites 15
ℹ pass 35
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 809.771417

All 35 test cases pass deterministically on Node.js 24 (Darwin ARM64 / macOS) and are configured to execute in GitHub Actions across Ubuntu and macOS on Node 20, 22, and 24.