How lyslama/k5-day8-vincon-fromworkflowstoagents works
LogiX-VN is a teaching harness for processing simulated shipping-refund claims through four competing architectures, returning a common run result and comparing measured outcomes. The workflow is: a user runs a shape for a claim; it reads scoped claim/order and policy data, computes a quote, and asks the host to authorize and commit any refund; the run emits a trace and result for evaluation. README documentation establishes the intended roles of the unsampled orchestrator and benchmark subsystems; their internal calls are not inferred here.
Main entry points: app.py Shape runner · workflow.py Fixed workflow · single.py Single agent · swarm_reference.py Reference swarm
Components 22
Execution and shapes
- Run CLI
src/lab8/run.py - Shape runner
src/lab8/app.py - Fixed workflow
src/lab8/shapes/workflow.py - Single agent
src/lab8/shapes/single.py - Orchestrator shape
src/lab8/shapes/orchestrator.py - Reference swarm
src/lab8/shapes/swarm_reference.py - ReAct runtime
src/lab8/runtime.py - Safe handoff
src/lab8/handoff.py - Seeded MockLLM
src/lab8/mock_llm.py
Claim and policy domain
- Read tools
src/lab8/tools.py - Compensation quote
src/lab8/compensation.py - Pinned configuration
src/lab8/config_loader.py - Public claim data
src/data/public_world_state.json - Public claims
src/data/public_claims.jsonl
Host safety and state
- Refund host
src/lab8/durability.py - Authorization policy
src/lab8/policy.py - SQLite world and ledger
src/lab8/db.py - Run result contract
src/lab8/schemas.py
Trace and evaluation
- Run trace
src/lab8/trace.py - Benchmark metrics
src/lab8/metrics.py - Shape benchmark
src/lab8/benchmark.py
Other components
- Lab user
Connections 30
- Lab user → Run CLI: runs claims
src/lab8/run.py - Run CLI → Shape runner: invokes runner
src/lab8/run.py - Shape runner → SQLite world and ledger: prepares store
src/lab8/app.py - Shape runner → Fixed workflow: dispatches shape
src/lab8/app.py - Shape runner → Single agent: dispatches shape
src/lab8/app.py - Shape runner → Orchestrator shape: dispatches shape
src/lab8/app.py - Shape runner → Reference swarm: dispatches shape
src/lab8/app.py - Shape runner → Public claims: loads claims
src/lab8/app.py - Shape runner → Public claim data: loads world
src/lab8/app.py - Fixed workflow → Read tools: reads and quotes
src/lab8/shapes/workflow.py - Fixed workflow → Refund host: requests refund
src/lab8/shapes/workflow.py - Single agent → ReAct runtime: runs ReAct
src/lab8/shapes/single.py - Single agent → Refund host: requests refund
src/lab8/shapes/single.py - Reference swarm → Safe handoff: creates handoff
src/lab8/shapes/swarm_reference.py - Reference swarm → Refund host: requests refund
src/lab8/shapes/swarm_reference.py - ReAct runtime → Read tools: dispatches tools
src/lab8/runtime.py - ReAct runtime → Safe handoff: uses handoff
src/lab8/runtime.py - Read tools → Compensation quote: calculates quote
src/lab8/tools.py - Read tools → Pinned configuration: reads policy date
src/lab8/tools.py - Seeded MockLLM → Pinned configuration: reads pricing
src/lab8/mock_llm.py - Refund host → Authorization policy: checks authorization
src/lab8/durability.py - Refund host → SQLite world and ledger: commits ledger
src/lab8/durability.py - Fixed workflow → Run trace: records run
src/lab8/shapes/workflow.py - Single agent → Run trace: records run
src/lab8/shapes/single.py - Reference swarm → Run trace: records run
src/lab8/shapes/swarm_reference.py - Shape benchmark → Benchmark metrics: summarizes trials
- Benchmark metrics → SQLite world and ledger: checks ledger
src/lab8/metrics.py - Seeded MockLLM → Single agent: provides completions
src/lab8/shapes/single.py - Seeded MockLLM → Reference swarm: provides completions
src/lab8/shapes/swarm_reference.py - Fixed workflow → Run result contract: returns result
src/lab8/shapes/workflow.py