Masterleeaus · Oct 4, 2026 · Oct 4, 2026 · Oct 4, 2026 · Oct 4, 2026 · Oct 4, 2026
diff --git a/.github/workflows/authority-policy-eval.yml b/.github/workflows/authority-policy-eval.yml
@@ -6,13 +6,15 @@ on:
     paths:
       - 'src/Authority/**'
       - 'src/Policy/**'
+      - 'packages/policy-engine/**'
       - 'evaluations/authority-policy/**'
       - 'scripts/authority-policy-eval.php'
       - '.github/workflows/authority-policy-eval.yml'
   pull_request:
     paths:
       - 'src/Authority/**'
       - 'src/Policy/**'
+      - 'packages/policy-engine/**'
       - 'evaluations/authority-policy/**'
       - 'scripts/authority-policy-eval.php'
       - '.github/workflows/authority-policy-eval.yml'

diff --git a/.github/workflows/verify.yml b/.github/workflows/verify.yml
@@ -3,9 +3,6 @@ name: Verify Interaction Engine
 on:
   pull_request:
   push:
-    branches:
-      - main
-      - feature/full-engine-template-upgrade
   workflow_dispatch:
 
 permissions:
@@ -22,40 +19,63 @@ jobs:
       - name: Check out source
         uses: actions/checkout@v4
 
+      - name: Set up PHP
+        uses: shivammathur/setup-php@v2
+        with:
+          php-version: '8.2'
+          coverage: xdebug
+          tools: composer:v2
+
       - name: Set up Node.js
         uses: actions/setup-node@v4
         with:
-          node-version: 20
+          node-version: '24'
           cache: npm
 
-      - name: Install offline companion toolchain
-        shell: bash
+      - name: Validate Composer package metadata
+        run: composer validate --no-check-publish
+
+      - name: Install PHP dependencies
+        run: composer install --no-interaction --no-progress --prefer-dist
+
+      - name: Install locked JavaScript dependencies
         run: npm ci
 
-      - name: Run complete engine verification
+      - name: Verify maintained-file SHA-256 manifest
+        run: php bin/manifest.php
+
+      - name: Run complete PHP and TypeScript verification
         shell: bash
-        run: php bin/verify.php
+        run: |
+          set -euo pipefail
+          mkdir -p coverage
+          php bin/verify.php | tee coverage/verification.txt
 
-      - name: Lint PHP and validate JSON
+      - name: Generate PHP coverage report
         shell: bash
         run: |
           set -euo pipefail
+          mkdir -p coverage
+          PHP_COVERAGE_OUTPUT=coverage/php.json php -dauto_prepend_file=tests/coverage-bootstrap.php tests/run.php | tee coverage/php-tests.txt
+          test -s coverage/php.json
+          PHP_COVERAGE_OUTPUT=coverage/policy-php.json php -dauto_prepend_file=tests/coverage-bootstrap.php packages/policy-engine/tests/run.php | tee coverage/policy-php-tests.txt
+          test -s coverage/policy-php.json
+
+      - name: Run 100-case authority evaluation
+        run: php bin/authority-eval.php coverage/authority-evaluation.json > coverage/authority-evaluation.stdout
 
-          php_files=()
-          while IFS= read -r -d '' file; do
-            php_files+=("$file")
-          done < <(find src tests bin routes config database -type f -name '*.php' -print0 2>/dev/null)
-          for file in "${php_files[@]}"; do
-            php -l "$file" >/dev/null
-          done
-          echo "PHP syntax valid: ${#php_files[@]} files"
-
-          json_files=()
-          while IFS= read -r -d '' file; do
-            json_files+=("$file")
-          done < <(find wizards templates resources config reports -type f -name '*.json' -print0 2>/dev/null)
-          for file in "${json_files[@]}"; do
-            php -r 'json_decode(file_get_contents($argv[1]), true, 512, JSON_THROW_ON_ERROR);' "$file"
-          done
-          echo "JSON valid: ${#json_files[@]} files"
+      - name: Run measured benchmarks
+        shell: bash
+        run: |
+          set -euo pipefail
+          php bin/benchmark.php coverage/performance-php.json > coverage/performance-php.stdout
+          npm run benchmark:offline -- coverage/performance-typescript.json > coverage/performance-typescript.stdout
 
+      - name: Store test, coverage, evaluation, and benchmark reports
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: interaction-engine-evidence
+          path: coverage/
+          if-no-files-found: warn
+          retention-days: 30
diff --git a/.gitignore b/.gitignore
@@ -1,9 +1,8 @@
-
 .worktrees/
-node_modules/
-coverage/
+/dist/
+/node_modules/
+/coverage/
 .phpunit.result.cache
 .env
 .env.*
 !.env.example
-
diff --git a/LICENSE b/LICENSE
@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Jason Lee
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
diff --git a/README.md b/README.md
@@ -4,147 +4,102 @@
 
 # Titan Zero Interaction Engine
 
-**A governed interaction runtime for adaptive workflows and authorised business actions, online or offline.**
+**A governed workflow runtime that keeps model proposals separate from authorized actions.**
 
 </div>
 
 [![Authority Policy Evaluation](https://github.com/Masterleeaus/Interaction-engine/actions/workflows/authority-policy-eval.yml/badge.svg?branch=main)](https://github.com/Masterleeaus/Interaction-engine/actions/workflows/authority-policy-eval.yml)
 
-## Product architecture and engineering highlights
+## Run a governed action in five minutes
 
-A governed interaction runtime that turns chat, voice, mobile, desktop, and API requests into authorized, traceable business workflows.
+The standalone policy package and demo need PHP 8.2. They do not need Composer, Laravel, a database, or an API key.
 
-- **Architecture:** Schema-driven wizards and versioned workflow catalogues feed local/hybrid intelligence, tenant-scoped context, capability policies, idempotent execution, and an IndexedDB-based offline companion.
-- **Distinctive engineering:** The key boundary is explicit: understanding and recommending do not confer authority. Approval, execution, evidence, and verification are separate stages.
-
-> **Status: module foundation with recorded standalone verification; host deployment readiness remains environment-specific.** The repository includes a cumulative build report and verification scripts. Its report explicitly lists host tenancy, permissions, queues, database compatibility, PWA integration, provider integration, and production-load testing as remaining destination-system checks.
-
-## What it provides
-
-The Interaction Engine connects chat, voice, mobile, desktop, and API experiences to structured workflows, local intelligence, business capabilities, and policy-controlled execution. Understanding user intent does not grant authority: recommendation, approval, and execution remain separate stages.
-
-## Core capabilities
-
-- Schema-driven Universal Wizard Engine with resumable sessions and conditional flows
-- Versioned workflow and template catalogues with readiness metadata
-- Local and hybrid intelligence components
-- Tenant-scoped cognitive events and evidence lineage
-- Capability policies, actor context, idempotency, and fail-closed execution
-- TypeScript offline companion using IndexedDB and encrypted command envelopes
-- OpenAPI contract in `resources/openapi.yaml`
-
-## Template discovery
-
-The catalogue contains 38 definitions: **29 ready templates** and nine drafts. Ready templates reference registered entry wizards and declared capabilities; drafts remain non-executable. The authenticated Laravel catalogue is available at `GET  /templates`, with template definitions discovered from `templates/`.
-
-Assurance workflows include **Incident Response** and inspection corrective action.
-
-### Commerce and multi-vertical template pack
-
-The [Commerce and multi-vertical template pack](docs/COMMERCE_MULTI_VERTICAL_TEMPLATE_PACK.md) adds reusable workflows for ordering, fulfilment, inventory, returns, refunds, and vertical-specific service operations. See `reports/workcore-compatibility.json` for engine readiness and host-connection status.
+```sh
+php bin/demo.php
+```
 
-## Architecture
+The deterministic proposal is denied first. The demo then simulates a signed owner approval bound to the exact quote data, runs an in-memory draft handler, and records an audit event. That simulated approval is not proof of a real user's identity and cannot mutate a host system.
 
 ```text
-Chat / Voice / Mobile / Desktop / API
-                  |
-                  v
-        Interaction Runtime
-                  |
-                  v
-          Wizard + Templates
-                  |
-                  v
-        Local Intelligence
-                  |
-                  v
-       Policy + Capability Layer
-                  |
-          +-------+-------+
-          |               |
-          v               v
-    Online host       Offline outbox
-          |               |
-          +-------+-------+
-                  v
-          Outcomes + Evidence
+Titan Zero Interaction Engine — governed quote demo
+Proposal: Create a $450 carpet-cleaning quote for Jenny.
+Gate before approval: DENY — A valid, scoped and unexpired human approval is required.
+Owner approval: VALID
+Execution: quote-demo-001 created as a draft for Jenny — AUD 450.00
+Audit: capability_executed recorded for quote-demo-001
+Summary: denied before approval; approved; executed once; audit entries=1
 ```
 
-## Technology
+## Policy Engine
 
-- PHP 8.2+
-- Laravel-compatible Illuminate components
-- TypeScript, IndexedDB
-- PHP Sodium and AES-GCM
-- OpenAPI and JSON/schema-driven definitions
+The Policy Engine is the execution boundary. Intent recognition, planning, recommendation, and model proposals do not confer permission. Only a registered capability handler reached after an allowed policy decision can execute an action.
 
-## Installation
+- Unknown capabilities fail closed; observe, recommend, and prepare policies cannot execute.
+- User-only actions require a human identity and roles from trusted server context, with fresh authentication when configured. The host must not copy identity claims from request or model data.
+- Delegated actors must be identified, tenant scoped, and carry every required scope.
+- Approval grants are HMAC-SHA256 signed, capability and tenant scoped, time limited, and bound to the action payload by default.
+- Numeric limits and registered domain rules run before execution.
 
-The module is intended for a compatible Laravel host. Verify the host version, dependency constraints, migrations, service provider registration, auth/tenant context, queue/cache setup, and route exposure before deployment.
+The framework independent package is in [`packages/policy-engine`](packages/policy-engine/README.md). Its `AuthorityLevel`, `CapabilityPolicy`, `ApprovalSigner`, and `PolicyEngine` classes have a focused test suite:
 
-```bash
-composer install
-npm ci
-php bin/verify.php
+```sh
+php packages/policy-engine/tests/run.php
 ```
 
-`php bin/verify.php` runs the PHP suites and the TypeScript offline-companion tests. The workflow installs the declared TypeScript toolchain before invoking the same verifier, so a clean checkout does not depend on a globally installed compiler. A prior cumulative build report records historical checks; it is not a substitute for rerunning tests on the current commit or validating a destination host.
+Only trusted server-side approval code should issue grants. The host remains responsible for authenticating users, building `_context`, persisting approvals, and recording durable audit evidence.
 
-## Integration requirements
+## Current 100-case authority evaluation
 
-A destination system must validate:
+The 100 deterministic scenarios cover approval-required, user-only, delegated, prepare-only, and unregistered capabilities. In CI, all 30 expected allows were allowed and all 70 expected denials were denied: zero valid actions were wrongly blocked and zero unauthorized actions passed the gate. The gate-bypassed control forwards unauthorized cases only to a no-op sink.
 
-- host-specific model and service mappings;
-- tenant and permission semantics;
-- authentication, API exposure, queues, cache, scheduler, and database migration compatibility;
-- browser/PWA integration and cloud provider configuration;
-- concurrency, load, and production failure behaviour.
+See the [100-case evaluation report](reports/authority-evaluation.md) and its [JSON result](reports/authority-evaluation.json). The separate 31-case suite also passed with zero unauthorized allows, zero wrongly denied valid actions, and zero scenario mismatches; see [its report](eval-results/authority-policy-latest.md). Both evaluations exercise the repository policy code directly and do not validate a connected host.
 
-The host business system remains the authority for operational mutations.
+## Optional live model proposal
 
-## Repository map
+The optional `openai-php/client` adapter returns structured proposal data. `OpenAIActionProposer` only accepts a caller-supplied capability allowlist and rejects authority, approval, identity, tenant, scope, and idempotency fields from model output. The proposal has no handler or policy bypass; the normal Command Bus rechecks trusted context and policy before any registered handler.
 
-- `interactions/`, `wizards/`, `templates/` — interaction and workflow definitions
-- `resources/ts/` — device-side offline companion
-- `resources/openapi.yaml` — API contract
-- `tests/` — runtime and workflow checks
-- `docs/`, `reports/` — architecture and verification evidence
+```php
+$proposal = $proposer->propose($userRequest, ['quotes.create']);
+$commands->dispatch(
+    $proposal->capability,
+    $proposal->payload + ['_context' => $trustedServerContext],
+);
+```
 
-The AI and agent-oriented implementation map is in [`docs/AI_ENGINEERING.md`](docs/AI_ENGINEERING.md). It distinguishes deterministic heuristics, optional cloud-model integration, policy enforcement and host-boundary work from claims the repository does not make.
+Cloud AI is disabled by default. Set `INTERACTION_AI_ENABLED=true`, `INTERACTION_AI_API_KEY`, and a high-entropy `INTERACTION_APPROVAL_SECRET` (at least 32 characters) in a trusted host environment. To make one live proposal and demonstrate its denial, install the optional SDK with `composer require openai-php/client`, then run `php bin/llm-proposal-demo.php`. It never executes the proposed action. Do not put credentials in a browser, device, or model prompt.
 
-## Security principles
+The local-language component is a deterministic **rule-based intent parser**. It extracts supported business intents and entities; it is not a language model.
 
-- Cloud AI is disabled by default.
-- Unregistered capabilities fail closed.
-- Tenant and actor context are required for consequential execution.
-- Offline replay uses the same policy checks as online execution.
-- Unsynchronised records remain queued until reconciliation succeeds.
+## The 80-engine library
 
-## Reproducible authority policy evaluation
+The repository contains 80 engine contracts with matching implementations across executive, cognitive, memory, learning, planning, human interaction, AI infrastructure, and business intelligence. [`docs/ENGINE_LIBRARY_80.md`](docs/ENGINE_LIBRARY_80.md) lists each as `Implemented` or `Partial` with behavior and limits. `Implemented` means the declared operations have executable behavior within the described boundary, not production validation for every host or industry. Heuristics, in-memory stores, and host-schema integrations are labeled accordingly; no engine is listed as interface only.
 
-The authority policy claim is measured directly against `PolicyEngine` using 31 fixed, labelled scenarios covering unknown capabilities, non-executable authority levels, human roles and authentication freshness, signed approval scope and expiry, delegated scopes and numeric limits, and policy callbacks.
+## Template discovery
 
-| Test | Allow-all bypass control | Interaction Engine |
-| --- | ---: | ---: |
-| Blocked requests allowed through | 24 / 24 | 0 / 24 |
-| Valid requests wrongly denied | 0 / 7 | 0 / 7 |
-| Wrong-tenant approvals accepted | 1 / 1 | 0 / 1 |
+The catalogue has 38 definitions: 29 ready templates and nine drafts. Ready templates reference registered entry wizards and declared capabilities; drafts remain non-executable. The authenticated Laravel catalogue is available at `GET  /templates`.
 
-The baseline is a deliberately simple allow-all control, not a competing product. The evaluation tests policy decisions only; it does not cover host adapters, persistence, or downstream execution.
+The [Commerce and multi-vertical template pack](docs/COMMERCE_MULTI_VERTICAL_TEMPLATE_PACK.md) covers ordering, fulfilment, inventory, returns, refunds, and vertical-specific service workflows. Incident Response and inspection corrective action are included in the assurance workflows. Host compatibility evidence is in [`reports/workcore-compatibility.json`](reports/workcore-compatibility.json); it does not claim a connected production host.
 
-Evaluated on **4 October 2026** with PHP 8.2.34 at commit [`bdbe73c`](https://github.com/Masterleeaus/Interaction-engine/commit/bdbe73cc445b900d29126e081bd5f2341e98ee6f). The fixed scenario corpus uses seed `20261004` and SHA-256 `a1e863c857aa05e70f720224c3f1bb0065f2f1c699fe27e438ce4a64a58ff340`. Reproduce with:
+The broader implementation map in [`docs/AI_ENGINEERING.md`](docs/AI_ENGINEERING.md) distinguishes deterministic heuristics, optional cloud-model integration, policy enforcement, and host-boundary work.
 
-```bash
-php scripts/authority-policy-eval.php
+## Install and verify
+
+The Laravel-compatible module targets PHP 8.2+ and Illuminate 10, 11, or 12. In a compatible host checkout:
+
+```sh
+composer install
+npm ci
+php bin/verify.php
 ```
 
-The [full scenario report](eval-results/authority-policy-latest.md) records all 31 outcomes; [CI run 37171514412](https://github.com/Masterleeaus/Interaction-engine/actions/runs/37171514412) passed. The initial CI attempt failed before evaluating scenarios because the evaluator resolved the repository root one directory too high; that path bug was corrected before this successful run.
+The verifier runs the standalone policy suite, PHP runtime and workflow suites, and TypeScript tests. GitHub Actions also runs PHP and TypeScript coverage, 100 authority cases, and measured benchmarks. See the [CI verification evidence](reports/verification-evidence.md) for the current counts and coverage scope.
 
-## License
+The TypeScript offline companion uses IndexedDB and AES-256-GCM for device-side command storage. The test suite uses a memory adapter; browser, database, authentication, and connected-host behavior still require validation in the destination application.
 
-The Composer package declares a proprietary license. Obtain the appropriate rights before external distribution or reuse.
+## Measured performance
 
-## Banner
+The benchmark scripts measure interaction-definition compilation, duplicate-command handling, and encrypted outbox enqueue/decrypt. The [measured report](benchmarks/performance.md) records sample counts, p50/p95, runtime, commit, and limits; results are a single hosted-CI run, not production capacity claims.
 
-A checked-in project-specific banner is displayed above.
+## Repository hygiene, history, and license
 
+This repository is licensed under MIT. Generated `dist/` output and dependencies are build-only and ignored. The risky external tarball import workflow has been removed. Historical import notes and provenance remain on the [`archive/interaction-engine-history` branch](https://github.com/Masterleeaus/Interaction-engine/tree/archive/interaction-engine-history).