Executive Decision Memorandum
Decision, uncertainties, evidence gates and management actions.
ENGAGEMENT 001
Regional Jobs & Resilience Evidence Review
A commissioned evidence engagement involving source reconstruction, programme-data QA, causal inference, uncertainty analysis and decision support. Client identity, restricted data, client-specific results and sensitive recommendations remain confidential.
Decision question
How should a regional programme portfolio distinguish apparent performance from defensible impact when implementation quality, participant attrition, economic conditions and evidence quality differ materially across markets?
Scope
What the Lab was asked to own
- 01Reconstruct the evidentiary perimeter.
- 02Reconcile inconsistent measures and denominators.
- 03Establish provenance for material statistics.
- 04Evaluate selection and attrition risk.
- 05Separate descriptive performance from causal evidence.
- 06Stress-test programme comparisons.
- 07Integrate relevant country-level economic conditions.
- 08Identify claims that should be withheld.
- 09Design the final decision and allocation framework.
Deliverables manifest
The engagement is an evidence system, not one report.
Technical Evidence Report
Identification strategy, diagnostics, sensitivity analysis, estimators and robustness.
Evidence Room
Sources, transformations, programme evidence, metadata and structured analytical outputs.
Claim Ledger
Every material claim mapped to its evidence, status, limitation and release decision.
Data Contract
Required fields, definitions, joins, provenance and acceptance criteria for secure client data.
QA Register
Critical and non-critical release gates.
Red-Team Review
Reasons findings may be wrong, overstated or non-identifiable.
Reproducibility Package
Executable analytical pipeline and specification record.
Public method note
Why the analytical architecture matters
RANDOMIZED / SYNTHETIC METHOD REHEARSAL — NOT CLIENT PERFORMANCE DATA
Before applying the analytical pipeline to restricted client records, the Lab stress-tested the estimator on a controlled 180,000-record environment with confounding and informative attrition. The exercise tests whether the proposed method can recover a known effect under conditions resembling the identification problem. It is not evidence about the client's programme.
Evidence withheld
Some questions should end without a number.
The engagement architecture permits a ranking, impact estimate or allocation recommendation to be withheld where comparability, identification, denominator integrity or structural breaks make the result indefensible.
| Question | Status |
|---|---|
| Can the public macro layer support contextual triage? | YES |
| Can public data alone rank programme impact? | NO |
| Can completed participants be treated as the programme denominator? | NO |
| Can observational treatment differences automatically be called causal? | NO |
| Can conflict-affected series be inserted mechanically into a regional ranking? | WITHHELD |
Public Evidence Perimeter
Public context remains source-linked.
institution → release → reference period → retrieval date → transformation → comparability note
This public record identifies the source institutions and required provenance schema. It does not disclose restricted client records or claim that every listed source was material to every client-specific conclusion.
Evidence status
Evidence types are never merged.
OFFICIAL SOURCEIssued by the originating public institution.
REAL PUBLIC DATARetrieved and transformed from public records.
CLIENT DATARestricted data supplied under the engagement; not published here.
SYNTHETICGenerated for method or system testing.
FOUNDER PRODUCEDAnalysis produced internally by the Lab.
EXTERNAL REVIEWReviewed outside the producing team when a review record exists.
PENDING VALIDATIONNot yet externally checked.
Publication boundary. This page is an Engagement Record, not a public copy of the confidential report. Client-specific data, conclusions, restricted appendices and sensitive recommendations are withheld unless publication rights explicitly permit disclosure. The public object contains the decision question, evidence architecture, deliverables manifest, public-source perimeter, method rehearsal and limitations.