The causal auditor for acted-on recommendations.
OpenLift does not create Levered's forward-looking probability. It grades decisions after they are acted on by estimating incremental impact against an appropriate counterfactual and writing the causal result back into the feedback loop.
OpenLift was created by Daramola Ben and is maintained as a public GitHub project. Inside Levered it acts as the causal auditor: recommendation, action taken, treatment/control design, incremental effect, credible interval, validity checks, audit grade and training eligibility.
| Audit ID | Decision | Design | Lift | p-value | Grade | TS |
|---|---|---|---|---|---|---|
| LX-8821 | GTM-31 · scale segment | Region holdout · 14d | +18.2% | p=0.003 | SUPPORTED | 12:04:11 |
| LX-8820 | PM-09 · prioritise fix | Switchback · 10d | +11.9% | p=0.012 | SUPPORTED | 12:03:57 |
| LX-8819 | REV-12 · route account | Observed outcome | +6.4% | p=0.041 | INCONCLUSIVE | 12:02:44 |
| LX-8818 | INV-04 · update plan | Cohort holdout | +4.3% | p=0.110 | UNSUPPORTED | 12:01:32 |
| LX-8817 | MKT-07 · retest creative | Power check | +2.1% | p=0.280 | INVALID | 12:00:08 |
Did the acted-on recommendation create incremental revenue or conversions?
Which geos should be test markets and which should be controls?
Was the experiment strong enough to trust?
What should the next experiment be?
Can this outcome safely inform future model learning?
Experiment Runner, Geo Matcher, Power Analysis, Multi-Cell testing, Creative Lift, Next Experiment, and Scorecard workflows.
openlift init openlift run experiment.yaml --out results.json openlift report results.json
What OpenLift does
The product is built around the full audit workflow: design or evaluate the test, measure lift, check validity, attach an audit grade, and feed eligible outcomes back into the Levered learning loop.
Estimate incremental revenue or conversions for acted-on recommendations with Bayesian synthetic control, posterior lift estimates, and credible intervals.
Match test and control markets, plan holdouts, run power analysis, calculate MDE, and choose experiment duration.
Attach supported, unsupported, inconclusive, or invalid grades to the original recommendation.
Translate lift into incremental CAC, ROAS, profit, budget scenarios, and payback curves.
Analyse creative-level lift from campaign or creative performance data, not just channel-level reporting.
Decide whether the observed result is safe to write back into the model learning and evaluation store.
Store experiment history locally and summarise cumulative evidence over time.
Track pre-trend, contamination, power and sensitivity checks before trusting the causal interpretation.
From geo time-series data to a causal audit grade.
OpenLift expects long-format geo time-series data and guides the team from upload through market selection, lift estimation, economics, validity checks, audit grade and training eligibility.
Upload or connect data
CSV, Google Sheets, Google Ads, and Meta Ads connector interfaces are supported.
Map the experiment fields
At minimum: date, geo, and outcome. Optional fields include spend, treatment, period, channel, and creative_id.
Choose markets
Use geo matching to select treatment and control markets before running the test.
Estimate incremental effect
Run Bayesian counterfactual measurement and inspect lift, uncertainty, diagnostics, validity and economics.
Write the audit back
Attach audit grade, credible interval, validity checks and training eligibility to the original recommendation.