Skip to content

demo-script.md

Latest commit

 

History

History
78 lines (59 loc) · 2.69 KB

demo-script.md

File metadata and controls

78 lines (59 loc) · 2.69 KB
Raw

Five-minute demo script

This sequence demonstrates the deterministic path and its safety boundaries. Run it from the repository root after uv sync --locked --dev --group agent --group sim. No API key, network call, GPU, or external service is required.

1. Typed intake and verified report (60 seconds)

uv run --locked qre-workflow --request "qaoa complete size 4 layer 1" \
  --thread-id demo-happy-path

Point out the validated workload, the two synthetic profiles, and the limitations carried in the output. The planner selects work; the deterministic kernel computes the metrics.

2. Tool trace (45 seconds)

uv run --locked qre-orchestration \
  --request "qaoa complete size 4" \
  --evidence-query "topology-aware quantum transpilation"

The JSON trace exposes every step from planning and retrieval through compilation, verification, and report writing.

3. Topology comparison (75 seconds)

uv run --locked python eval/run_eval.py --output /tmp/qre-demo-evaluation.json
uv run --locked python eval/check_regression.py --current /tmp/qre-demo-evaluation.json
uv run --locked qre-report --input /tmp/qre-demo-evaluation.json \
  --output /tmp/qre-demo-report.md \
  --csv-output /tmp/qre-demo-report.csv \
  --html-output /tmp/qre-demo-report.html
sed -n '1,80p' /tmp/qre-demo-report.md

Show the O0 control beside O2 topology-aware placement, then the equal-constants table. The latter isolates topology/compiler effects from illustrative profile parameters.

4. Induced failure (30 seconds)

uv run --locked qre-orchestration \
  --request "teleport size 4" \
  --evidence-query "quantum resource estimation" || true

The unsupported workload fails at the typed planning boundary and produces a structured failed trace; no compilation runs.

5. Checkpoint retry and review resume (60 seconds)

uv run --locked pytest -q -s \
  tests/unit/test_agent.py::test_workflow_can_retry_after_an_invalid_request \
  tests/unit/test_agent.py::test_human_review_interrupt_can_resume_with_checkpoint

The two focused contract tests demonstrate retrying a failed thread with a corrected request and resuming a human-review interrupt through the same checkpointer seam.

6. Consolidated evidence and integrity check (30 seconds)

uv run --locked qre-eval-report --output /tmp/qre-demo-eval-report.md
sed -n '1,220p' /tmp/qre-demo-eval-report.md
uv run --locked qre-calibration-parity --output /tmp/qre-demo-calibration-parity.json
uv run --locked qre-manifest --check

Close on the separate compilation, noise, retrieval, and orchestration results and their shared limitations. The manifest check proves the committed report inputs and outputs have not drifted.