Admissibility frontier · Edition 3

Compiled corpus reading projection · Wider corpus

Worldwright calibration identifiability audit

Canonical path
corpus/programmes/worldwright-research/calibration_identifiability_audit_2026-10-23.md
Source SHA-256
7e18367dc303b6a4697a7907401dba47f4fddcd8b8cef36c48472959d29c4c31
Build revision
91a0b922ec4e7119c8f23c140838872041d2d05a
Corpus links
0 inbound · 0 outbound

History and editing → · Raw Markdown →

Route into this page

Why this page is here: it belongs to the wider canonical corpus. No explicit Markdown-link route from an Edition 3 citation was found; enter it through hierarchy or search.

Worldwright calibration identifiability audit

Date: 2026-10-23

The current public-source programme has a useful discovery frame: 248 eligible IFRC events across 91 countries and six hazard classes, day-precision onset for every event, and public report references for 94.8%. Its attempted GLIDE bridge nevertheless reaches only 32.3% existing-or-strong coverage, recovers 61.1% of masked positive controls exactly, and selects a wrong unique serial in 11.1%.

The important correction is that cross-source identity is not the sole calibration blocker. The retained projection contains no locality, intervention channel, dispatch or first-delivery time, baseline deficit or trend, delivered dose, health outcome, relational outcome, selection reason, or concurrent-intervention record. Consequently none of the six principal causal mechanisms—urgency loss, delivery failure/capture, learning threshold/gain, relational persistence, identity salience, or channel contrast—is structurally identifiable from the current ledger.

Even a perfect event crosswalk would therefore not open causal estimation. It would permit safer multi-source enrichment, but action-level timing, baseline, outcome and confounding evidence would still be missing.

Corrected pathway

The next work should proceed in two lanes:

  1. An IFRC-only action-time feasibility pilot can use exact IFRC event and report identifiers to ask whether locality, delivery channel, dispatch, and first-delivery intervals can be extracted under a governed projection. Outcomes and retained narrative remain barred.
  2. A source-provided crosswalk or governed partner identifiers remain necessary for multi-source enrichment. The failed fuzzy matcher must not be loosened.

Only after action-time feasibility should the programme bind baseline deficits, repeated harm outcomes, selection, and concurrent interventions. Aggregate outcome weights are not parameters to estimate from this observational cohort; they require stakeholder-separated elicitation and partial-order analysis.

Machine-readable evidence is in worldwright-calibration-identifiability-audit-local-v1.json, with checks in verifyworldwrightcalibrationidentifiabilityaudit_local.mjs.

Referenced by

Inbound corpus links

No explicit inbound Markdown links were found.

Continues toward

Outgoing corpus links

No explicit outgoing corpus links were found.