Kenneth ChengMathematical modeling

Editorial project overview

Assimilating What You Can Observe

How much state and held-out peak performance can a known aggregate-and-delay observation operator recover, and does its latent uncertainty pass predeclared calibration gates?

Reproducible studyUpdated 30 Aug 2026
Technical figure from Assimilating What You Can Observe
A reviewed research figure synchronized from the private technical workspace. The original aspect ratio and labels are preserved.

Editorial overview

The literature gate returned REFRAME because epidemic ensemble filtering, reporting delay and observation-function mismatch are established topics. The defensible contribution is a frozen, failure-preserving replication-extension audit—not a new EnKF, dengue model or field validation.

Read Assimilating What You Can Observe for the observation model, matched four-arm protocol, point and calibration diagnostics, held-out peak comparison, preserved coverage failure, reproduction hashes and evidence boundary.

Key findings

  • The correct aggregate-and-delay arm reduces infected-state RMSE from 264.255 open loop and 126.335 under the wrong operator to 36.090.
  • It predicts the held-out day-62 total-infected peak with zero-day timing error and 0.743 percent intensity error, compared with five days and 4.726 percent for the wrong operator.
  • Its nominal 90 percent latent-state interval covers only 0.562963, below the frozen 0.70 minimum, so the machine verdict is PARTIAL rather than fully supported.

Limitations

  • The evidence uses one deterministic six-patch synthetic SIR truth, one ensemble design and one jointly misspecified comparator.
  • The study does not isolate grouping, delay and reporting-fraction effects or compare alternative filters, smoothing, inflation or localisation.
  • No dengue records, patients, districts, clinical decisions, operational forecasts or public-health recommendations are involved.

Technical record

Detailed source, calculations, generated figures, and reproduction instructions are maintained in a private technical workspace. Public articles contain only manually reviewed interpretation and approved figures.

Version history

2026-08-30 — Curated overview reviewed against repository evidence.