Skip to main content

Continuation Work

The active FCV program is now operating as a scientific-instrument project rather than an archive-recovery exercise.

The current stack is:

empirical-data-contracts
+ spatial-data-foundation
        ↓
fcv-empirical-data
  source-native facts + reusable measurements
        ↓
fcv-experiment-harness
  scientific projection + gates + estimators
        ↓
Africa Observability Lab
  known-behavior calibration + commissioning
        ↓
human status / interpretation here
  1. Current Research Status
  2. Research System Architecture
  3. Africa Observability Lab
  4. Calibration Benchmark Catalog
  5. Empirical Product Catalog
  6. DHS Empirical Stack for survey work
  7. Experiment Surface Catalog
  8. Research Workflow and Validation
  9. Validation Status
  10. Experimental Design Status
  11. Source Data Implementation Status
  12. Project Classification Status

Current authority/status pages​

  • Current Research Status — shortest full-system snapshot and next evidence transitions.
  • Research System Architecture — ownership/boundary map.
  • Africa Observability Lab — calibration kernel, observability engine, recovery semantics, and commissioning boundary.
  • Calibration Benchmark Catalog — authoritative/published targets and prerequisites.
  • Empirical Product Catalog — current empirical products and grains.
  • DHS Empirical Stack — HR/GC/GPS/integration/semantic survey state.
  • Experiment Surface Catalog — substantive FCV experiment surfaces only.
  • Research Workflow and Validation — empirical → experiment → gates → observability/commissioning → estimation.
  • Validation Status — three evidence lanes: current contracted architecture, observability/commissioning, recovered calibration.
  • Experimental Design Status — current downstream scientific-use choices.
  • Source Data Implementation Status — implemented versus candidate empirical source capabilities.
  • Project Classification Status — annotation versus treatment boundary.

What materially changed in the latest harness wave​

The experiment harness now has three additional capabilities that should change how we plan research:

Reusable observability​

The old one-point E2 injection check is now an effect-size grid characterizing sign recovery, detection, interval coverage, and recovery error under the existing design.

Calibration Lab kernel​

Known-behavior checks are first-class purpose = calibration runs with commissioning, positive/negative controls, synthetic injection, measurement agreement, and independent Level 1/2/3 recovery states.

External benchmark queue​

The research queue is now concrete:

issue #16 — generic auxiliary empirical input
→ Nigeria DHS 2018 electricity commissioning
→ Briggs 2017 positive control
→ reassess instrument bottleneck
→ Breckner–Sunde only after truthful grid/month support

This means the next question is increasingly what the instrument fails to recover, not simply what additional pipeline component can be built.

Current DHS continuation frontier​

DHS now has:

HR + GC + GPS
→ integration QA
→ codebook-backed household measurements

The preferred continuation path is:

close calibration auxiliary-input seam
→ protected real DHS materialization
→ official DHS report commissioning
→ explain discrepancies
→ choose substantive outcome/exposure
→ household ↔ cluster scientific projection
→ displacement-aware exposure
→ survey-design gates / estimator

This is more diagnostic than jumping directly from source ingestion to a treatment-effect regression.

Current panel continuation frontier​

GeoGCDF + ACLED is still the strongest current contract-backed panel reference surface.

The next valuable real run should ideally produce both:

experiment gate/estimator evidence
+
observability curve evidence

so we know not only what the estimate is but what signal magnitudes the same apparatus could recover.

Evidence tracks​

Current contract-backed architecture​

Core source/measurement/projection/treatment machinery and the new observability/calibration software pass synthetic acceptance.

A canonical real current-artifact fully contracted panel run is still pending.

Observability / commissioning​

Synthetic E2 observability and the generic calibration kernel are implemented.

Official DHS and published-study external benchmarks are designed but not yet executed.

Recovered real-data calibration​

The historical WBad/WBkg → ACLED E2 checkpoint remains real recovered-lane evidence and should not be relabeled as current contracted or new commissioning evidence.

Reference / design-history pages​

Use these after the current overlays when their detail is needed:

Before interpreting any estimate​

Check:

  • exact empirical authority and hashes;
  • experiment projection/timing/eligibility;
  • gate evidence;
  • observability evidence when available;
  • external commissioning status when the component has a suitable benchmark;
  • whether the result belongs to the current contracted path or recovered historical lane.

A useful operating question is now:

Do we know what the instrument would do when the truth is known before we ask it to interpret truth that is unknown?