Causal · Counterfactual · RLHF Research Console
Tenantconsolidated · 14 tenants
Window
live · refreshed 14s ago
Causal ROICounterfactualsSimulation LabLearning TimelineRec QualityPolicy EvolutionTreatment vs Control
Counterfactual Analysis Workspace
“What would have happened if…” — DR-learner with stratified replay
live · 28d window · n=18,420
Active counterfactual · CF-001
What if we accept all blocked recs?Factual (what happened)
$4.82M
30d horizon · DR-learner
OTIF %
94.2
Cost/load
$1,842
Recs accepted
62%
Risk (VaR95)
$310k
Counterfactual (would-have)
$5.96M
30d horizon · DR-learner
OTIF %
93.8
Cost/load
$1,791
Recs accepted
84%
Risk (VaR95)
$420k
Scenario Catalog
5 active · ranked by |Δ|ID
Scenario
Baseline
Scenario
Δ ($M)
Conf
Risk
CF-001
What if we accept all blocked recs?
$4.82M
$5.96M
+1.14
84%
med
CF-002
What if guardrail confidence ≥0.7 (vs 0.85)?
$4.82M
$5.42M
+0.60
78%
high
CF-003
What if fuel was flat YoY?
$4.82M
$4.31M
-0.51
92%
low
CF-004
What if no carrier C-7 capacity?
$4.82M
$3.94M
-0.88
81%
high
CF-005
What if policy v4.2 was kept?
$4.82M
$4.10M
-0.72
86%
low
Replay overlay
factual vs counterfactual trajectoryΔ realized
+$1.14M
Risk uplift
+0.110
Confidence
0.84
External factor overlays
contributors held constantFuel price
−$0.84M
contribution removed
Seasonality
−$0.31M
contribution removed
Volume mix
−$0.45M
contribution removed
Carrier capacity
+$0.12M
contribution removed