Multi-touch Attribution QA Runbook
Release QA, on-call checks and repair steps for multi-touch attribution in Prosper202 1.9.76 and later, with p202 commands.
Run this checklist after you deploy or upgrade Prosper202 ClickServer 1.9.76 or later, and whenever attribution numbers look off. Nearly every check is a command in the p202 CLI; the few that need SQL or a PHP script say so. Reading needs an API key with the attribution:read scope, and changing models also needs attribution:write. The p202 CLI itself requires Prosper202 Pro on the install, trial included; without Pro most commands exit with code 6, while /api/v3/attribution in the REST API still serves the same attribution data.
How the engine works
- Each conversion, whether it arrives by pixel, postback, upload or API, is written to a ledger and queued in an outbox table.
- A worker empties that queue. The every-minute cron gives it 20 seconds each run, and it can also run on its own.
- Per conversion, the worker builds one journey: up to 25 clicks by the same visitor, joined through visitor keys rather than IP address or user agent, inside the longest lookback of your active models (30 days at minimum).
- It then splits credit under every active model. A conversion's credits total exactly 1, and its credited revenue equals its counted amount. Reports and exports read an hourly rollup.
Key checks
- Cron is running:
p202 system cron. The every-minute job should have run within the last two minutes. - Queue is draining:
p202 attribution queue.pendingshould fall back soon after traffic, andfailingshould stay at 0. - Models are usable:
p202 attribution model list. A model markedinvalidfailed validation, andp202 attribution model get <id>shows why. - One conversion end to end:
p202 attribution journey <conv_id>lists the touches, how the journey was built and each model's split. - Totals reconcile:
p202 attribution breakdown --group-by campaign --period last30. All models credit the same total revenue.
Also in the runbook
Model rules and weighting, per-campaign models, exports, safe recomputes, rollback and what to send when you escalate. An upgrade from 1.9.55 or earlier can't be undone, so back up first. Command details are in the p202 CLI Reference.
Full guide on prosper.tracking202.com
The full runbook adds healthy sample outputs, read-only SQL checks and step-by-step incident fixes: the Prosper202 multi-touch attribution QA runbook
Updated about 3 hours ago