Calibrate,
or wait another day?
Two explanations match the same protein endpoint: a slow clock under deep suppression, or a fast clock under weak suppression. The design comparison on the main page asks what a next measurement is worth. This page works two synthetic cases end to end, recomputing every number in your browser from the same code the workbench runs. In the first, tightening the baseline calibration separates the hypotheses and another day of waiting does not. In the second, the ranking reverses. For each case the sampled gap is decomposed into the terms the two levers actually touch, priced in each other's units, and then mapped across kinetics so the crossover can be located rather than asserted. The last section takes your own numbers.
Computing both cases…
Where each change wins
A single case can only show that the ranking is contingent. To see what it is contingent on, the same comparison runs across a grid of slower half-lives and follow-up times, with the faster hypothesis held at a fixed ratio and the sustained-RNA centers rebuilt in every cell so both explanations keep reproducing the same anchor. The letter in each cell names the change with the larger sampled gap: C for tighter baseline calibration, W for waiting one more day.
Computing the grid…
WINNER BY KINETICS AND FOLLOW-UP TIME
The crossover is a dimensionless one
C · calibration gives the larger sampled gapW · waiting gives the larger sampled gap* neither change reaches separation
Shading depth is the margin between the two changes in percentage points; the letter, not the color, carries the result. Cells labeled n/a are kinetics the declared anchor cannot produce at all.
What decides the ranking
Under a sustained-step hypothesis, protein approaches the plateau r + (1 − r)·2−t/H → r, so the distance between the two central trajectories stops growing once both have reached their plateaus. A later readout can only add whatever separation the curves still have left to generate. When the effective half-lives are short relative to the follow-up time, that remaining motion is small, and the gap between the sampled ranges converges to a limit set by the shared baseline uncertainty, the follow-up error, and the biological bounds. Baseline calibration acts on that limit directly; time does not.
When the half-lives are long relative to the follow-up time, the trajectories are still separating quickly. A day of extra time then buys more predicted separation than halving the baseline log-width, which is what the shipped default shows. The mechanism is the same in both cases; only the position of the follow-up on the two clocks differs.
Test your own case
Change any of the declared assumptions and the whole comparison is rebuilt: matched sustained-RNA centers, sampled ranges, separation budget, timing diagnostics, exchange rates, and the wait curve. Hypotheses that the anchor cannot produce are refused with a reason rather than approximated, which is itself informative: a protein cannot reach the declared endpoint by the anchor time if its half-life is too long, no matter how complete the suppression.
How the threshold is computed
For each case, the page scans the retained fraction f of the baseline log-width, replacing ×/÷F with ×/÷Ff at fixed center, and bisects for the largest f at which the sampled gap at the current follow-up time is still positive. It reports the sampled gap immediately on either side of that fraction so the crossing is inspectable. Bisection assumes the gap is monotone in f over the scan, which the tests check against a coarse grid; it does not assume anything about the biology beyond the declared bounds.
How the exchange rates are computed
The equivalent wait is the earliest scanned readout time whose sampled gap matches the gap the calibration change reaches at the current follow-up; it is reported as unreachable when no time inside the declared horizon gets there, rather than extrapolated past it. The equivalent calibration is the largest retained log-width fraction whose sampled gap at the current time matches what the delay reaches, found by bisection on the same monotonicity assumption the threshold scan uses. The replicate lines invert a single assumed scaling, n = n0/f 2, which is what a one-over-root-n baseline log half-width implies; they inherit nothing else from the model and carry no error rate.
Limits
- Two parameter sets, not a survey. A different anchor, threshold, error correlation, or biological bound can move the crossing. The regime grid varies two parameters at a time, not jointly, and its crossovers are bracketed by adjacent cells rather than solved. Use the design comparison and the baseline/correlation map on the main page to explore your own assumptions.
- The budget is an identity, not an inference. Decomposing the sampled gap into a central separation and two reaches is exact bookkeeping over the same profile. It explains where a gap comes from; it does not make the gap a statement about probability.
- The replicate arithmetic is width arithmetic. One over root n is an assumption imposed here, not a property of any assay. Systematic baseline error does not average down, so the numbers are a floor on effort, never a promise of it.
- Time is not wasted, it is bounded. In the plateau case, waiting still narrows the overlap slightly. The point is that it cannot close it under these bounds, not that a later measurement is worthless.
- No cost model. A calibration experiment, a repeat assay, and a day of cell culture have different costs and failure modes. The tool does not weigh them.
- Software verification is not biological validation. Tests confirm the code computes what the equations state. They do not establish that any real protein behaves as a single constant-loss pool.
Continue with the design comparison on the main page, or read the derivation and assumptions. All presets are synthetic; no experimental, employer, or confidential data are included.