feat: add lda report workflow example

examples/lda_report_workflow/documents/architecture-notes.md

@@ -0,0 +1,21 @@
+# Architecture Notes
+
+Lifecycle:
+
+- Drafts are mutable authoring workspaces with revisions.
+- Artifacts are immutable versioned workflow definitions.
+- Deployments bind logical source ids to configured concrete sources.
+- Runs persist stopped execution records and bounded traces.
+
+Runtime:
+
+- The core executes typed graph nodes and routes by declared outcomes.
+- State writes go through reducer-aware merge semantics.
+- Interrupt nodes pause at explicit human-in-the-loop boundaries.
+- Resume payloads are validated before state mutation.
+
+Source providers:
+
+- Python sources support trusted local demo capabilities.
+- MCP sources preserve upstream session state through a runtime pool.
+- OpenAPI source support exists as an experimental provider.

examples/lda_report_workflow/documents/evaluation-findings.md

@@ -0,0 +1,17 @@
+# Evaluation Findings
+
+Evidence:
+
+- Automated tests cover core runtime, artifacts, deployments, CLI, JSON-RPC,
+  source providers, and examples.
+- A 36-trial audited agent challenge campaign evaluated the product-facing CLI
+  under bounded conditions.
+- Manual audit flags separate product-surface success from source-code or prior
+  answer reads.
+
+Limitations:
+
+- Agent challenge runs are operational evidence, not a controlled model study.
+- The campaign used small sample sizes and changing prototype snapshots.
+- The prototype does not claim production security, scheduling, RBAC, or a
+  general autonomous planning algorithm.

examples/lda_report_workflow/documents/project-brief.md

@@ -0,0 +1,19 @@
+# Project Brief
+
+lda.chat is a workflow substrate for AI-agent-facing workspace automation. The
+prototype separates external planning from deterministic workflow execution.
+
+Key achievements:
+
+- Typed Draft, Artifact, Deployment, Run, and Trace lifecycle records.
+- Source-provider boundary for platform, MCP, Python, and experimental OpenAPI
+  sources.
+- JSON-RPC and CLI surfaces usable by external agents and human operators.
+- Deterministic report and browser-click examples with audited agent challenge
+  runs.
+
+Current positioning:
+
+- The system is not a bundled autonomous planner.
+- External agents or humans operate the workflow lifecycle.
+- The next product-facing step is a local Workflow Console and defense demo.

examples/lda_report_workflow/documents/risk-register.md

@@ -0,0 +1,18 @@
+# Risk Register
+
+Material risks:
+
+- Title and product framing can overstate the implemented autonomous-agent
+  layer if not explained carefully.
+- Evaluation evidence is stronger as systems evidence than as a controlled
+  empirical model comparison.
+- File-backed stores are useful for auditability but not a production
+  transaction boundary.
+- The Workflow Console needs a strict loopback-only first slice to avoid
+  becoming an arbitrary RPC proxy.
+
+Mitigations:
+
+- Keep the agent/substrate boundary explicit in the thesis and defense.
+- Present challenge data as bounded operational evidence.
+- Defer production storage, auth, and remote proxying to future work.

examples/lda_report_workflow/documents/roadmap.md

@@ -0,0 +1,15 @@
+# Roadmap
+
+Near-term:
+
+- Add self-describing interrupt request and resume contracts.
+- Build a deterministic lda.chat report workflow with typed issue approval.
+- Build a local Workflow Console over JSON-RPC.
+- Add live-demo replay support for the defense.
+
+Later:
+
+- Stabilize the experimental OpenAPI provider.
+- Add production secret stores and transactional persistence.
+- Add a surrounding agent interface and planner loop.
+- Explore scheduling, richer debugging, and visual workflow editing.