# Review and operating runbook

## Refresh

1. Install the pinned capstone dependencies in Python 3.12.
2. Run the project's `analysis.py` from the full collection; review source hashes if publisher files change.
3. Require all `outputs/quality.json` checks to pass, then review excluded observations and evaluation dates.
4. Compare metrics with the prior receipt. Do not silently replace a holdout with a newly tuned result.
5. Rebuild the portfolio only after notebooks and case-study text match the recomputed metrics.

## Before an operational pilot

Use the dashboard to compare risk and review workload by hotel and channel. Before operational use, obtain versioned booking events and run a randomized follow-up trial to estimate actual cancellation prevention and contribution margin.

Assign an owner, define acceptable data latency, obtain permissions for operational inputs, and record baseline service performance. Stop the pilot when required predictors are missing or the input population changes outside the evaluated scope. Track outcome delay, denominator coverage and error by important operating segments. Use newly collected data for monitoring and future evaluation; this portfolio is not a live service.

## Native Power BI

The NYC companion contains three report pages and 12 DAX measures. Extract its ZIP, run `powerbi/Set-DataFolder.ps1`, open `NYCMobility.pbip` and Refresh in Power BI Desktop. Verify dates, borough filters, weighted averages and total trip counts. Native engine refresh and rendering remain unverified in this environment. Hotel and electricity dashboards run directly in the portfolio browser.