Causal Attribution — 7 Method Convergence · All methods consuming OpenG2P data
Diff-in-Differences
−18.3%
95% CI: −16.1% to −20.5%
Treatment: PBMS enrollment · Control: SR non-enrolled
Propensity Score Matching
−17.8%
95% CI: −15.9% to −19.7%
Matched on SR demographics + G2P Bridge disbursement dosage
ML Causal Forests
−18.7%
Heterogeneous treatment effects
SR + PBMS + Bridge reconciliation logs as feature set
Regression Discontinuity
−16.9%
Bandwidth: ±2 PMT points
PMT threshold from OpenG2P PBMS eligibility rules
Synthetic Control
−17.4%
Donor pool: 12 comparable districts
Baseline: SR demographic data pre-OAL deployment
Instrumental Variables
−19.1%
Instrument: SPAR financial address availability
IV sourced from OpenG2P SPAR rollout timeline
All 6 methods converge on −17% to −19% poverty reduction. Treatment assignment from OpenG2P PBMS enrollment records. Treatment dosage from G2P Bridge disbursement logs. Outcome measurement from SR dynamic registry (ODK field sync). No separate survey apparatus required — OpenG2P was already collecting the data.
Evidence Chain — OpenG2P Data → RBF Verification Package
OpenG2P PBMS
→
Enrollment records · Treatment assignment · Disbursement cycles · Eligibility decisions
OpenG2P G2P Bridge
→
Disbursement logs · Payment timestamps · Reconciliation · Treatment dosage data
OpenG2P SR (ODK sync)
→
Outcome measurements · MPI updates · Field enumerator data · Post-intervention resurvey
SkyEdgeAI OAL
→
Causal inference engine (7 methods) · Governance metadata · Decision rationale · Counterfactual analysis
GuardianLedger™
→
Immutable audit chain · Every decision timestamped from Day 1 · Cannot be retrofitted
RBF Verification Package
→
Causal attribution + audit trail + governance sign-off → $40M tranche unlock · Donor-grade evidence
Critical Architectural Note — The Data Trail Starts at Enrolment
OpenG2P timestamps every SR record, every PBMS disbursement cycle, every G2P Bridge transaction from Day 1.
SkyEdgeAI's causal attribution is retroactively valid from the moment OpenG2P is deployed — but only if SkyEdgeAI OAL was also running from Day 1 to capture governance metadata alongside operational data.
If SkyEdgeAI is added later, the causal baseline is permanently compromised.
In 18 months when your donor asks for RBF evidence, you will not be able to produce causal attribution for the first 18 months.
The architecture decision is now. It cannot be undone.