Variance Floor Trigger¶

Threshold monitor — instantiates Variation–Selection–Retention Engine Design

A tripwire that fires when a population's diversity falls toward a floor, forcing fresh variation back in before selection grinds the pool down to a single fragile winner.

Selection is a diversity-spending process: each generation it concentrates the population toward whatever currently scores best, and left unchecked it spends the pool down to a single winner — efficient today, defenceless the moment the environment moves. Variance Floor Trigger is the guardrail against that. It is a tripwire watching a diversity statistic of the live population, set to fire before diversity collapses rather than after. When the measured spread approaches a preset floor, it acts — injecting fresh variation from a maintained reserve, shielding minority variants from pruning, or easing selection pressure until headroom is restored. Its defining feature is that it watches diversity, not fitness: it will deliberately protect variants that are losing right now, because their only value is the option they preserve for a future the current winner cannot survive.

Example¶

A quant fund runs a book of a few dozen trading strategies and lets performance decide capital allocation — winners get more, losers get starved. Through a long trend the allocation quietly concentrates: capital piles into the handful of strategies riding the same momentum factor, and a book that still looks diversified on the org chart becomes, in effect, one crowded bet.

A Variance Floor Trigger watches not returns but the correlation structure of where capital actually sits. When effective diversity drops toward its preset floor, it fires: capping any single factor's share and holding a minimum allocation in uncorrelated strategies that are underperforming right now. Those starved strategies feel like dead weight during the trend — which is exactly the point. When the momentum regime finally breaks, the book still holds variants adapted to the new environment, instead of having spent them all chasing the old one.^[1] The trigger cost some return during the good times; what it bought was survival of the turn.

How it works¶

It continuously measures a diversity statistic of the live population — genotype spread, strategy correlation, coverage of the option space — not the fitness the rest of the loop optimizes. That number is compared against a floor, and as the population approaches it the trigger fires a restorative action drawn from a maintained inventory of variation sources: re-inject mutation or fresh candidates, exempt minority variants from the persistence rule, or relax the selection pressure enough to let laggards survive a while longer. Two things distinguish it from ordinary performance monitoring: it acts on diversity rather than fitness, and it acts pre-emptively, on the approach to the floor, because a diversity collapse is far cheaper to prevent than to reverse once the variants carrying the lost options are already gone.

Tuning parameters¶

Floor level — how high the minimum diversity sits. Higher preserves adaptability but dilutes selection and slows convergence toward any winner; too high and the loop never commits to anything.
Diversity metric — what "diversity" is measured as. The choice determines which monocultures the trigger can even see; a population diverse on the chosen axis can still be a monoculture on the one that matters.
Response action — inject variation, shield minorities, or relax pressure. Each restores headroom differently and at a different cost to current performance.
Hysteresis band — a hard threshold or a band with separate trip and reset points, to keep the trigger from oscillating on and off near the floor.
Reserve depth — how much fresh variation is kept on hand to inject. A thin reserve fires but has little to add; a deep one is insurance that costs to maintain.

When it helps, and when it misleads¶

Its strength is cheap insurance against premature convergence: it keeps the engine from spending its whole diversity budget on the current optimum, so that when the environment shifts the population still holds variants able to meet it. It forces the system to keep options a pure fitness rule would liquidate.

Its failure modes are the mirror image. A floor set too high neuters selection — the loop hedges forever and never converges on anything. A poorly chosen diversity metric hides the real monoculture, so the book looks varied while every variant shares the one exposure that matters. And it is easily gamed: "we must preserve diversity" is a convenient banner for protecting pet variants that deserve to die. The classic misuse is raising the floor to justify keeping favourites, or quietly installing it after a collapse to look prudent in hindsight. The discipline that keeps it honest is to tie the floor to a named adaptability need, choose a metric that tracks the specific fragility you fear, and fix both in advance of the pressure that would erode them.

How it implements the components¶

Variance Floor Trigger realizes the variation-supply and diversity-guardrail side of the archetype's machinery — the components that keep the pool from starving:

variance_floor_or_diversity_reserve — it defines and enforces the floor: the minimum diversity the population is never allowed to fall below, held as a deliberate reserve rather than an accident.
variation_source_inventory — it maintains the stock of variation sources it draws on to re-inject diversity when the floor is breached, so the response is a stocked action and not a wish.

It does not measure fitness or judge the proxy (fitness_metric_or_survival_proxy → Fitness Proxy Audit), decide which variants persist under normal conditions (differential_persistence_rule → Retention / Pruning Protocol), set the loop's cadence (selection_cadence_and_generation_unit → Generation Cadence Review), or keep the ancestry record (retention_and_lineage_memory → Variant Lineage Log) — it reads that record for its diversity signal but does not own it.

Instantiates: Variation–Selection–Retention Engine Design — it is the guardrail that keeps selection from consuming the diversity the engine needs to keep adapting.
Consumes: Variant Lineage Log supplies the current population's spread and lineage concentration — the signal the trigger measures against its floor.
Sibling mechanisms: Variant Lineage Log · Retention / Pruning Protocol · Selection Loop Map · Selection Pressure Sandbox · Fitness Proxy Audit · Generation Cadence Review · Environmental Shift Retest · Multi-Pressure Tradeoff Matrix · Champion–Challenger Rotation · Escape Variant Watchlist · Adverse Adaptation Red Team

Notes¶

Because it acts on diversity rather than fitness, the trigger will sometimes override selection — protecting variants that are losing on the metric everyone else is optimizing. That tension is the mechanism, not a defect: it is the loop buying an option against a future the current winner cannot handle, and it will always look like waste right up until the environment moves.

References¶

[1] In evolutionary search a population that loses its diversity collapses onto a local optimum it can no longer climb out of — premature convergence. Diversity-preserving techniques such as niching, fitness sharing, and migration exist precisely to hold a variance floor; the trading-book analogy is exact in structure, differing only in substrate. ↩