Skip to main content

Validation Status

This page is the human-facing evidence ledger for the active FCV scientific instrument.

The project keeps distinct evidence lanes:

  1. current contract-backed architecture;
  2. real experiment gate evidence;
  3. observability / external commissioning;
  4. recovered real-data calibration;
  5. substantive estimator output.

None is interchangeable with causal evidence.

Evidence vocabulary

EvidenceEstablishesDoes not establish
Software / synthetic acceptanceBoundary logic behaves on controlled fixtures.Real-source validity or a real effect.
Empirical materialization / QASource-backed product exists with provenance, hashes, coverage, QA.Correct scientific role.
Real experiment gate runDeclared design has real support, coverage, timing, and falsification evidence.Automatic causal validity.
Synthetic observabilityKnown injected truth has characterized recovery behavior.That the real effect exists.
External commissioningApparatus reproduces authoritative external measurements/behavior.A new FCV substantive result.
Estimator resultDeclared estimator produced an estimate after gates.Robust causal truth.

Track A — Current contract-backed architecture

Shared foundations

empirical-data-contracts and spatial-data-foundation provide shared contracts, geography/time, membership, and provenance.

Status: AVAILABLE / REAL EMPIRICAL USE

Current GeoGCDF treatment measurement

AidData GeoGCDF v3.0.1 has a governed commitment-area-period Gold product over GADM 4.1 ADM2.

Key policy:

resolution_policy = exclude_unresolved
unresolved geography projects = 22
unresolved commitment time = 0
excluded unresolved projects = 22

Structural zeros are licensed relative to the resolved eligible project universe. The 22 exclusions remain explicit rather than silently repaired.

Status: REAL MATERIALIZATION GREEN WITH EXPECTED YELLOW RESOLUTION-EXCLUSION QA

ACLED outcome measurement

The source-native sparse Gold remains conservative with absent_row_semantics = unknown.

A separate coverage-certified derivative licenses structural zeros only inside an explicit 48-country / full-period certification surface.

Status: REAL SPARSE + CERTIFIED MEASUREMENTS GREEN

Track B — Real current-artifact E2 experiment

Reference analysis scope is the 47-country GeoGCDF treatment-authorized universe intersected with governed GADM and certified ACLED coverage.

Primary model frame:

38,520 ADM2-period rows
6,420 ADM2 units
47 countries
6 treatment periods
7,667 treated
30,853 controls

Gate ledger

GateStateReal metric
E0 input universeGREEN0 pre/post missing; declared coverage resolved
E1 treatment supportGREEN7,667 treated / 30,853 controls
E2 within-period supportGREEN6 / 6 periods mixed
E3 outcome sparsityGREEN0.9222 zero share
E4 pretreatment balanceGREEN`
E5 pre-outcome placeboGREEN0.0047 outcome SD
E6 0.20-SD recoveryGREEN30 / 30

The positive-reported-amount STRESS treatment also passed all gates.

Status: REAL CURRENT-ARTIFACT E0–E6 PASS

Canonical estimator

effect = +0.54999 fatalities
SE = 0.45819
z = 1.20
effect ≈ +0.0176 outcome SD

This is not strong substantive evidence.

Status: REAL ESTIMATE PRODUCED / SMALL-IMPRECISE CANONICAL SIGNAL

Track C — Current E2 R0–R5 characterization

R0 reference lock

The exact analysis identity and PRIMARY frame were frozen before characterization.

Status: COMPLETE

R1 detector curve

Injected truthCanonical rejection / detection
0.00 SD0.02; CI coverage 0.98
0.02 SD0.25
0.05 SD0.96
0.10 SD1.00
0.20 SD1.00

Status: COMPLETE — EMPIRICAL RESOLUTION BOUNDARY CHARACTERIZED

R2 uncertainty calibration

Predeclared families:

ADM2 cluster null rejection ≈ 0.020
country cluster + t null rejection ≈ 0.000
wild country bootstrap null rejection ≈ 0.045

Point estimates were identical; uncertainty behavior differed. No procedure was selected because it gave a more favorable SE.

Status: COMPLETE — CHARACTERIZED

R3 influence / concentration

No single country or high-influence ADM2 explains the canonical coefficient. Period dependence is materially stronger; omitting 2011–12 flips the fatalities sign.

Status: COMPLETE — GEOGRAPHIC CONCENTRATION LIMITED / TEMPORAL HETEROGENEITY MATERIAL

R4 falsification

t−2 placebo ≈ 0.0039 outcome SD
future-treatment placebo ≈ 0.0034 outcome SD
structured-null p ≈ 0.176

The empirical structured-null probability is calibration evidence only, not a causal randomization-test p-value.

Status: COMPLETE — TIMING CLEAN / CANONICAL NOT EXTREME UNDER STRUCTURED NULL

R5 sparse-outcome family

Outcome representationNatural result
fatalities OLS+0.0176 SD
VAC event-count OLS+0.0549 SD
any-VAC LPM+2.04 pp
VAC event-count PPMLIRR 1.348; AME +0.203 events

The canonical fatalities result is not replaced. Raw coefficients are not cross-model comparable.

Status: COMPLETE — POSITIVE INCIDENCE/COUNT PATTERN CLEARER THAN FATALITY SEVERITY

Track D — External commissioning / positive controls

DHS official-report commissioning

Nigeria 4 / 4 GREEN
Uganda 2 / 2 GREEN
Zambia 2 / 2 GREEN
-------------------
TOTAL 8 / 8 GREEN

Status: EXTERNAL COMMISSIONING PASS

Briggs (2017): DHS wealth measurement

Independent historical DHS reconstruction using HV270, HV005 × HV012, and HV024 recovers published regional wealth-quintile quantities essentially to rounding precision.

Status: STRONG POSITIVE CONTROL

Briggs (2017): full aid-targeting analogue

Published oracle:

log richest +0.72034
log poorest +0.10460
within R² 0.23685
N = 195 / 17 countries

Independent pragmatic analogue:

log richest +2.3275
log poorest +0.9488
within R² 0.1419
N = 65 / 7 countries

Known divergences are explicit: current GADM instead of exact historical region ontology, source-informed AidData parent/child semantics, and a reduced defensible country/region set.

The rich-over-poor ordering survives trimming, equal-country weighting, and every leave-one-country-out run; richest LOO range 1.8398–3.0837.

Status: QUALITATIVE POSITIVE CONTROL / NUMERICAL PARITY NOT CLAIMED / CLOSED

Track E — Recovered/legacy-backed calibration

The recovered WB→ACLED E2 checkpoint remains genuine historical calibration evidence:

24,852 area-periods
4,142 GIDs
4 declared WBad/WBkg cells passed hard gates
0.20-SD injection recovery = 30/30 in every cell

This is recovered-lane evidence, not a substitute for current-artifact characterization.

Current readiness summary

Surface / instrument testCurrent state
DHS HR real-source materializationPASS — 3 RELEASES
DHS official-report commissioningPASS — 8 / 8 GREEN
GeoGCDF current treatment GoldPASS — 22 EXPLICIT EXCLUSIONS
ACLED certified outcome GoldPASS
Current GeoGCDF → ACLED E0–E6PASS — REAL FRAME
R0 exact reference lockCOMPLETE
Full real-frame observability curveCOMPLETE / CHARACTERIZED
Uncertainty calibrationCOMPLETE / CHARACTERIZED
Influence / omission stabilityCOMPLETE / CHARACTERIZED
Timing / structured-null falsificationCOMPLETE / CHARACTERIZED
Sparse-outcome familyCOMPLETE / CHARACTERIZED
Briggs DHS measurementSTRONG POSITIVE CONTROL
Briggs full analogueQUALITATIVE POSITIVE CONTROL / CLOSED
Supported-environment exact numerical rerunARCHIVAL HOUSEKEEPING PENDING
DHS spatial exposure experimentPARTIAL — SCIENTIFIC USE PENDING
Current World Bank spatial-period measurementPARTIAL — MISSING EMPIRICAL MEASUREMENT
Breckner–Sunde 2019DEFERRED — GRID + MONTHLY SUPPORT NEEDED

Next evidence transitions

  1. Reproduce the closed numerical packet in a supported environment for archival acceptance.
  2. Advance DHS spatial-exposure commissioning with explicit displacement, cross-grain linkage, timing, and survey design.
  3. Build a current World Bank spatial-period measurement only when it supplies an independent donor comparison.
  4. Undertake Breckner–Sunde only after regular-grid geography and calendar-month semantics exist truthfully.
  5. Prefer cross-source / convergent-validity benchmarks over more E2 specification variants.

This board states what the instrument has demonstrated. It does not collapse materialization, experiment gates, detectability, uncertainty calibration, external controls, and substantive inference into one claim that “the pipeline works.”

See September 2026 Calibration Wave Closure for the detailed closed packet.