docs: clarify Layer 2 after/ directory validation behavior

Address review comment: distinguish file-based tests (which diff
before/ to after/) from monitoring-based tests (which generally
do not diff after/ contents).

Author: data-douser

docs/public.md

@@ -255,10 +255,11 @@ dependencies:
 
 ## Further Reading
 
-- [Tools Reference](./ql-mcp/tools.md) — Complete list of MCP tools
-- [Prompts Reference](./ql-mcp/prompts.md) — MCP prompts for CodeQL development workflows
-- [Resources Reference](./ql-mcp/resources.md) — MCP resources for CodeQL learning and reference
 - [Getting Started (developer)](./getting-started.md) — Building from source and advanced configuration
+- [Testing Reference](./testing.md) — Overview of this repo's release testing strategy
+- [MCP Tools Reference](./ql-mcp/tools.md) — Complete list of MCP tools
+- [MCP Prompts Reference](./ql-mcp/prompts.md) — MCP prompts for CodeQL development workflows
+- [MCP Resources Reference](./ql-mcp/resources.md) — MCP resources for CodeQL learning and reference
 - [VS Code MCP Server Documentation](https://code.visualstudio.com/docs/copilot/customization/mcp-servers) — Configuring MCP servers in VS Code
 - [Publishing and Using CodeQL Packs](https://docs.github.com/en/code-security/tutorials/customize-code-scanning/publishing-and-using-codeql-packs) — Managing CodeQL packs with the CodeQL CLI
 - [Model Context Protocol](https://modelcontextprotocol.io/) — The MCP specification

docs/testing.md

@@ -26,12 +26,12 @@ Unit tests verify the TypeScript implementation of the MCP server.
 Integration tests exercise individual MCP tools against a live server instance using the custom MCP client.
 
 - **Client**: `client/src/ql-mcp-client.js` — starts the MCP server, invokes tools, and validates results.
-- **Test data**: `client/integration-tests/primitives/tools/` — each test has `before/` and `after/` directories capturing expected state.
+- **Test data**: `client/integration-tests/primitives/tools/` — each test has `before/` and `after/` directories that define the initial fixture state and, for file-based tests, the expected final state.
 - **Run command**: `npm run test:integration:default -w client` (or `npm run test:client` from the repo root).
 - **Key properties**:
   - Tests are deterministic and repeatable.
   - No mocks — tests use real CodeQL databases and queries bundled under `server/ql/`.
-  - The `before/monitoring-state.json` file supplies tool arguments; the `after/` directory captures expected output files.
+  - The `before/monitoring-state.json` file supplies tool arguments. For file-based tests, the integration-test runner diffs filesystem state from `before/` to `after/`; for monitoring-based tests, `after/` artifacts are generally not diffed and are only interpreted for specific validations (for example, `codeql_query_run` interpreted output).
 
 ## Layer 3 — Agentic Validation