Visual tour — see the app working¶

When to read: After (or instead of) a local run — you want screenshots of verify, CLI, Streamlit, and the wiki. Personas: PERSONAS.md.

These images were captured from a real laptop run (DuckDB + Streamlit). The FOIA LLM shot used local Ollama (llama3.2:3b on 127.0.0.1:11434). CI still uses the template insight with no Ollama.

Quarantine payloads may show synthetic emails/phones from samples/public_comments.csv — not real people.

Sam — prove it without an API key¶

./scripts/verify.sh

verify.sh PASS banner

Expected: OPERATOR_ETL_VERIFY=PASS, 41 pytest, FOIA demo silver=10 quarantined=2.

Template etl-graph insight

Priya — FOIA officer view (Streamlit Gov tab)¶

After a FOIA graph run, point Streamlit at that warehouse:

export OPERATOR_ETL_WAREHOUSE=".tmp/mvp-demo/operator.duckdb"
export OPERATOR_ETL_ORDERS_WAREHOUSE=".tmp/orders-demo/operator.duckdb"
uv run streamlit run dashboard/app.py

Gov / FOIA dashboard — 12 bronze, 10 silver, 2 quarantined, gate PASS

What “good” looks like: Bronze 12, Silver 10, Quarantined 2, Gate PASS, PII flagged 4. Latest insight numbers must exist in gold (critic).

Gov tab with quarantine expander open

The two quarantined rows are intentional (empty body, bad timestamp). Details: DASHBOARD.md.

Riley — orders pipeline and local Ollama¶

Orders demo (fresh warehouse): 21 in → 17 silver, 4 quarantined, quality PASS.

Orders CLI

Orders demo tab

Optional: same FOIA graph with local Ollama. Critic still has to pass.

etl-graph against local Ollama llama3.2:3b

Recipe: LLM.md. The model may misread a rate (e.g. treat 0.4 as comments-per-agency); the critic only checks that digits exist in gold, not that the sentence is semantically perfect. Do not claim production LLM insights.

Jordan — wiki as the public proof surface¶

Operator ETL wiki home

Live site: https://khaosans.github.io/operator-etl/. Audit: FINAL-REVIEW.md.

Recapture screenshots¶

Seed warehouses, then:

uv pip install playwright
uv run python -m playwright install chromium
./scripts/demo_mvp.sh   # or etl-graph into .tmp/mvp-demo-ollama for the LLM insight
OPERATOR_ETL_WAREHOUSE=.tmp/orders-demo/operator.duckdb uv run etl run --source demo
uv run python scripts/capture_screenshots.py

CI does not run Playwright. PNGs under docs/assets/screenshots/ are committed artifacts.