Campaign & Predictor Performance
Report type key: recommender_general_markdown
Configuration type: Interactions
Group: Interactions
Campaign-first performance for selected predictors: executive funnel, campaign scorecards, time/velocity/latency, convergence evidence, economics, data quality, and methodology guards.
UI families channels, convergence, economics, and fairness also land in this report (they do not get separate markdown keys).
How to get it
- Create and save an Interactions configuration.
- Enable: Interactions family general (on by default).
- Run the job and wait until completed.
- Open Reports (
/reports), select that job, and choose this type.
Markdown viewer: /markdown/JOB_ID/recommender_general_markdown
API: GET /reports/JOB_ID/recommender_general_markdown
The general family was off, predictor selection was missing, or the step failed. Check job logs and Analysis & LLM.
What it contains
- Validation findings and data fidelity (exact funnel vs sampled customer profiles).
- Catalogue scorecard for the general family.
- Declared experiment comparison when you configured control + treatment campaigns (uptake comparison only; randomization is not independently verified).
- Executive interaction funnel — recommendation calls → attributed responses → calls with an accepted offer → accepted offer impressions.
- Campaign-first scorecards — calls, responses, accepted-call rate per campaign.
- Time, velocity, latency — daily volume, ratios, attributed response latency.
- Convergence and stability — logged error observations if present. A zero-valued runtime error field is not proof that ranking has converged.
- Economics — estimated net value only when cost/margin exist on the logs; otherwise skipped (this is not Financial ARPU).
- Data quality, attribution coverage, diagnostics, not measurable catalogue metrics, methodology/prerequisite guards.
If attributed responses equal accepted calls, the response collection likely records uptake events only — do not read that as 100% conversion.
How to read
Start with the funnel, then campaign response rates, then daily ratios. Treat “What to change next” as advisory experiments. Role-specific lenses (business analyst / data scientist / ML engineer) may appear when LLM interpretation is on.
Companion artifacts
| Key | Artifact |
|---|---|
recommender_general_markdown | Markdown report |
recommender_general_notebook | Jupyter notebook |
recommender_general_results | JSON results |
recommender_general_graphs | Trend graphs |
Markdown, notebook, and graphs for the same family usually land together. JSON feeds Analysis Insights.