Amplification Curves → Cq

1. 01Raw amplification curves pasted as a Cycle column plus one column per reaction, fitted reaction by reaction, with the fitted sigmoid drawn over its own observations so the fit is checked by eye
2. 02Eight sigmoidal models — l4, l5, l6, l7 on the log-logistic scale and b4, b5, b6, b7 on the raw scale — adding Richards asymmetry, then linear baseline drift, then a quadratic term
3. 03Automatic model choice by nested F-test, likelihood ratio, or AICc weights across all eight, with the comparison table shown rather than a silent winner
4. 04All seven quantification-cycle definitions — cpD2, cpD1, maxE, expR, Cy0, CQ and maxRatio — plus a cycle shift, because they are different quantities that land on different cycles for the same curve
5. 05Per-reaction amplification efficiency, the fitted coefficients, residual variance, R² and AICc beside every Cq
6. 06Four independent efficiency estimators side by side — sigmoidal fit, window-of-linearity, exponential-phase fit and LRE — with the window each one chose, and every failure named instead of blanked
7. 07Kinetic outlier detection across the plate, by efficiency Z-score (Bar) or by cpD1 − cpD2 shape (Tichopad)
8. 08A warning naming the reactions whose baseline crosses zero, where a spline efficiency is genuinely undefined rather than merely large
9. 09CSV export of the Cq table, the model comparison and the all-methods table

Every estimator is a port of the R package qpcR, and the optimiser under it is a port of MINPACK’s trust-region Levenberg–Marquardt — the same code nlsLM runs — rather than a textbook damped Gauss–Newton, which on the harder models found a different minimum on 12 of 52 curves. Validated byte-for-byte against real qpcR over all 52 curves in its own datasets, 2,860 assertions. The panel runs on @ports/discover, @ports/eds, @ports/melt, @ports/qpcr, @ports/rdml, @ports/workflow.

Everything runs client-side. Files are parsed in the browser and projects are saved on this machine, so an unpublished sequence is never uploaded.

Which Cq definition should I use?

cpD2, the second-derivative maximum, is the default and what most real-time PCR work reports. The other six are offered because they ask different questions and land on different cycles for the same curve, so compare a plate under one of them and never a mixture.

Do I need a standard curve to get an efficiency?

No. Efficiency is read from the shape of each amplification curve itself, four independent ways at once, so a single reaction has an efficiency. A dilution-series standard curve is still the right check on an assay, and it lives one tab away.

Can I load an RDML file or an instrument export?

Yes. The Import tab reads RDML v1.0–1.4 archives and bare XML, lists every run in the file, and sends a run’s curves straight here; a pasted CSV or TSV table still works exactly as before.

Can I get a confidence interval on a Cq?

No. Bootstrap and Monte-Carlo error propagation over the fitted model are not built, so the panel reports a Cq with no interval rather than an interval it invented.

What happens to a well that never amplified?

The estimator that cannot run on it says so by name. Two of the upstream methods would instead return a number computed from a window at negative cycles, or index past the end of the data; both refuse here, because a plausible number from a failed fit is worse than a stated failure.