Small, inspectable models for assumptions that can change an assay's interpretation. Part of Data-Rich, Insight-Poor · 99 Small Problems · No. 02: Useful models for assumptions with expensive ambitions.
- ADC productive-delivery and pharmacodynamic model: prescribed entry, routing, processing, competing loss, payload escape, finite-window exposure, and an optional engagement/recovery hypothesis.
- Paired-antigen audit: same-cell coexpression, exact quadrants, marginal-only bounds, and a mechanism-evidence panel.
- Single-antigen distribution audit: empirical or synthetic distributions, antigen-low tails, and explicitly hypothetical response mappings.
All three run in the browser, open in dark mode, and require no installation. Demonstrations are synthetic; entered observations are processed locally without an analysis server or persistent storage. Hosting and external fonts still involve normal network requests. Do not enter confidential data.
In the delivery model, load Equal AUC, different time profiles. Review the two-hour entry pulse and 24-hour window, then acknowledge the prescribed free-payload assumption before running the optional PD hypothesis.
The default synthetic cases have approximately equal 24-hour cytosolic AUC and nearly equal peak downstream signals, but a roughly 20.2-fold B/A endpoint-signal ratio. Compare Slow engagement, same K and No signal recovery without changing delivery inputs. Inspect absolute signals, peaks, the assay-clock audit, and residual exposure rather than interpreting a single ratio as efficacy.
Matching changes B's entry amplitude only. Equal finite-window AUC is not equal administered dose, lifetime exposure, target engagement, viability, or therapeutic benefit. Export JSON to preserve parameters and assumptions alongside CSV/PNG outputs.
- Transport: calculation code, equations, units, assumptions and worked examples.
- PD: engagement/recovery code, endpoint interpretation code, mathematical derivations and conceptual references.
- Paired antigen: code, mathematics.
- Single antigen: code, mathematics and references.
- Scientific context: mechanistic-modeling reading guide.
The delivery model includes competing intact-lysosomal loss, separate released-payload losses, mass-balance checks, processing-versus-loss maps for steady arrival and finite-window AUC, pulse/chase inputs, profile normalization, and inspection/export controls. The optional PD layer shares its assumptions between A and B and does not feed back into payload transport.
These tools are deliberately not a single integrated therapeutic predictor. Antigen expression does not determine the prescribed entry flux. The maps are sensitivity calculations, not probability maps; input ranges are not evidence-based uncertainty distributions. The PD signal is arbitrary, not cell killing. Parameters are not fitted to a therapeutic construct, and passing numerical tests is not biological validation.
Node.js 22 or later is used in CI. No npm dependency installation is required for the app, worker build, or numerical tests.
node scripts/build-worker.mjs
node --test tests/*.test.mjs
python -m http.server 8080 --directory webOpen http://localhost:8080/productive-delivery.html. The generated worker bundles are committed so the static app also works directly after checkout. CI rebuilds and verifies all four bundles before deploying web/ to GitHub Pages.
v0.2.0-alpha publishes the paired-antigen and delivery/PD companions. Computational revisions remain separately identified as coexpression-1, delivery-9, and pd-2; the delivery component's historical 0.2.6-example identifier is not the repository release version. The original v0.1.0-alpha is unchanged.
See release notes and deployment guidance. GitHub Pages follows main; release tags provide fixed snapshots. The accompanying publication article, cover, and private editorial material are not included in this release.
Original code and documentation are under the MIT License. Cited publications and external fonts retain their own rights and licenses. No product-specific sequences, construct designs, proprietary assays, confidential observations, or fitted therapeutic parameter sets are supplied; this is not a clinical tool or a legal clearance.