Skip to content

Generation Cadence Review

Cadence review — instantiates Variation–Selection–Retention Engine Design

Checks whether the selection loop is turning at the right tempo — fast enough to adapt, slow enough that each generation is judged on signal rather than noise — and re-sizes the generation unit, coupled to the variation supply, when it is not.

A Generation Cadence Review examines the tempo of a selection loop — how long one generation lasts, how many variants make up a cohort, and how often selection is applied — and re-sizes the generation unit when the tempo is wrong. Its distinguishing concern is neither which variant wins nor what "fitness" means, but the loop's clock: a cadence too fast judges each generation on noise and burns through variation before it pays off, while a cadence too slow cannot track a moving target. The review's defining move is to weigh adaptation speed against evaluation reliability, and to check that fresh variation is entering fast enough to justify whatever pace the loop is set to run.

Example

A public crop-breeding program selects its best lines each field season — one generation per year. A cadence review asks whether that clock is right. Measured against the breeder's equation, genetic gain accrues per unit time in proportion to selection accuracy and intensity and inversely to the generation interval, so a year-long cycle is throttling annual progress. The review recommends adding an off-season nursery (a form of speed breeding) to fit ≈two generations into the calendar year — but flags the catch: halving the interval roughly doubles the demand for new crosses, or the faster loop will simply re-select the same handful of lines and converge. It also checks the other direction — that a single season is still long enough to measure a line's true performance rather than one year's weather. The output is a re-sized generation unit: faster, but coupled to a matching rise in variation input so the extra speed buys adaptation instead of premature uniformity.

How it works

  • Name the generation unit. Fix what one round actually is — what counts as a generation of variants and what event closes it — so "a cycle" stops being ambiguous.
  • Weigh speed against reliability. Judge the interval and cohort size against both the rate the environment moves and the exposure each variant needs before its result is trustworthy.
  • Check variation replenishment. Inventory where new variants come from and how fast each source refills, and test whether that rate keeps up with the proposed cadence.
  • Re-size, don't just re-clock. Recommend a new interval, cohort size, or selection frequency — paired with the variation input needed to sustain it.

Tuning parameters

  • Generation interval — how long a cycle runs. Short intervals adapt faster but judge on noisier, thinner evidence and burn diversity; long ones are reliable but sluggish.
  • Cohort size — how many variants are evaluated together each round. Larger cohorts sharpen selection accuracy at higher cost per generation.
  • Selection frequency — selecting every round versus pooling several rounds before a cut, trading responsiveness against stability.
  • Evaluation window — how much of the cycle is spent measuring before judging, guarding against calling a winner on too little signal.
  • Variation coupling — how tightly a faster clock is tied to a required rise in variation input; decoupling them is what turns speed into convergence.

When it helps, and when it misleads

Its strength is preventing the two tempo pathologies at once: a loop run too fast that thrashes on noise and exhausts its variation, and one run too slow that keeps selecting for conditions that have already moved on. By naming the generation unit explicitly, it also stops teams conflating an operational cycle ("we ship weekly") with a true generation of variants.

Its failure mode is tuning tempo in isolation. Speeding the clock without raising the variation supply does not accelerate adaptation — it accelerates convergence to a local optimum and drains the diversity reserve, so the loop only gets faster at getting stuck. A cadence set to the average rate of change also misses bursty shifts that arrive between reviews. The classic misuse is running it backwards — marshalling the review to ratify a schedule already chosen for other reasons rather than to test whether that schedule is the right generation. The discipline is to justify the cadence from the environment's real rate of change and the evidence each variant needs, and to move cadence and variation input together.

How it implements the components

  • selection_cadence_and_generation_unit — its core output: an explicit definition of the generation unit and a setting for the interval, cohort size, and selection frequency the loop runs at.
  • variation_source_inventory — it compiles the catalogue of variation sources and their refill rates specifically to test whether new variation arrives fast enough for the chosen tempo.

It does not select which variant persists (that's Champion–Challenger Rotation), preserve a diversity floor when a faster clock burns variance (that's Retention / Pruning Protocol and Variance Floor Trigger), or judge what "fitness" should be (that's Multi-Pressure Tradeoff Matrix).

Notes

Cadence and diversity are coupled: every increase in selection speed raises the rate at which variance is spent, so a cadence change should never be made without a matching decision about variation input and the diversity floor. This review sizes the clock; keeping that clock from draining the reserve is shared with Variance Floor Trigger and Retention / Pruning Protocol.

References

The breeder's equation — response to selection R = h²S, and in its per-year form scaling inversely with the generation interval — is the quantitative statement of why cadence matters: shortening the interval raises gain per unit time only if selection accuracy and variation are maintained, which is the exact trade the review weighs.