Diversity-Floor Rate Boost¶
Diversity-maintenance policy — instantiates Opportunity-Gated Adaptive Diversification
A standing control rule that automatically raises the rate of new-variant generation whenever measured diversity falls below a floor, then relaxes it once variety recovers.
Left to itself, an exploring population collapses: a few early winners attract all the resources, variety drains away, and the space locks in long before its niches are understood. Diversity-Floor Rate Boost is the automatic governor that fights that collapse. It couples a diversity measurement to a variation dial — when diversity drops below a set floor, it turns up the rate at which new and divergent candidates are introduced; when variety recovers, it eases off. Its defining move is that it is a closed feedback loop, not a one-time act: it doesn't seed a population once, it continuously defends a diversity floor by modulating the generation rate, the way a thermostat holds temperature rather than lighting a single fire.
Example¶
A music-streaming service has a recommender that, left alone, feeds each listener an ever-tightening loop of the same three genres — engagement stays high for a month, then quietly craters as the feed becomes claustrophobic. The team wires in a Diversity-Floor Rate Boost. A diversity metric (genre and artist entropy over the last fifty plays) is checked continuously against a floor. While a listener's feed stays varied, the recommender behaves normally. The moment entropy slips under the floor, the policy raises the injection rate of exploratory picks — unfamiliar genres, adjacent artists — until variety climbs back over the line, then relaxes so the feed doesn't feel random.
The result isn't a fixed "10% random" quota, which would over-explore the already-adventurous and under-serve the stuck. It's a responsive rate: high exactly when and where a feed is collapsing, near-zero when it's healthy. Diversity stops being something the team audits quarterly and becomes something the system holds in real time.
How it works¶
- Read a diversity signal. Consume a live measure of how much variety the population currently holds — typically the floor and gap reading from a coverage diagnostic.
- Compare to the floor. Check the signal against the minimum acceptable diversity.
- Modulate the generation rate. Below the floor, raise the divergence rate — more exploratory or mutated candidates per cycle, biased toward the under-covered kinds; above it, decay the rate back toward baseline.
- Close the loop. Re-measure after each adjustment so the boost is self-correcting rather than a fixed override.
The distinctive part is the proportional, self-relaxing response: the rate is an output of the diversity state, not a constant.
Tuning parameters¶
- Floor level — the diversity threshold that triggers a boost. A high floor guarantees variety but taxes exploitation of what already works.
- Boost gain — how sharply the rate rises per unit of shortfall. High gain recovers variety fast but can overshoot into thrash; low gain drifts.
- Decay rate — how quickly the boost relaxes once diversity recovers. Slow decay stabilises; fast decay risks oscillation around the floor.
- Targeting — whether the extra variation is spread evenly or steered toward the specific kinds a coverage check flags as thin.
- Damping — smoothing that prevents the controller from chasing noise in the diversity signal and hunting back and forth across the floor.
When it helps, and when it misleads¶
Its strength is that it prevents premature convergence — the collapse of an exploring population onto a local winner before the space is mapped — automatically and continuously, rather than depending on someone noticing the monoculture in a quarterly review. Turning up variation precisely when diversity is scarce is a well-established anti-convergence tactic in adaptive search.[n1]
Its failure modes are those of any feedback controller. Set the gain too high or the damping too low and it oscillates — flooding the population with variants, over-correcting, starving them, boosting again. It can also defend diversity that is merely nominal: raising the count of kinds while every new variant is feeble, buying entropy at the expense of fitness. And a floor set by habit rather than by the task can force costly exploration long after the space is genuinely understood and consolidation is the right move. The discipline is to tie the floor to real evidence about how well the space is mapped, and to let the selection mechanisms cull the weak variety the boost injects.
How it implements the components¶
Diversity-Floor Rate Boost realises the variation-control side of the archetype — the components that act to keep variety alive:
divergence_rule— it is the rule that sets, and dynamically raises, the rate at which divergent candidates are generated.feedback_and_selection_loop— it closes the loop between measured diversity and generation rate, continuously self-correcting rather than firing once.
It consumes but does not define the floor it defends (diversity_floor — that's Diversity Coverage Matrix), and it does not perform the initial seeding of the candidate population (mutable_candidate_population — that's Experimental Cohort Split and Specialization Cohort Seeding).
Related¶
- Instantiates: Opportunity-Gated Adaptive Diversification — this policy is how the archetype keeps a population from collapsing into monoculture mid-exploration.
- Consumes: Diversity Coverage Matrix supplies the floor and the gap targeting the boost acts on.
- Sibling mechanisms: Diversity Coverage Matrix · Innovation Portfolio Review · Network Mixing Protocol · Experimental Cohort Split · Merge and Deprecation Plan · Multi-Criteria Selection Rubric · Parallel Pilot Trials · Protected Pilot Lane · Stage-Gate Exploration · Lineage–Niche Fit Dashboard · Niche Portfolio Matrix · Opportunity Landscape Mapping · Preserve–Prune–Recombine Review · Saturation and Crowding Review · Specialization Cohort Seeding
Editorial Notes¶
Form Classification¶
Form family: Control, Automation & Runtime
Rationale: Diversity-Floor Rate Boost operates as a live operational control that automatically routes, enforces, adapts, or responds during execution because it a standing control rule that automatically raises the rate of new-variant generation whenever measured diversity falls below a floor, then relaxes it once variety recovers.
Independent corroboration: The frozen evidence defines Diversity-Floor Rate Boost as 'A standing control rule that automatically raises the rate of new-variant generation whenever measured diversity falls below a floor, then relaxes it once variety recovers', so its operative form is Control, Automation & Runtime.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Computer Science & Software Engineering
Origin pattern: Single lineage
Present-day reach: Multi-domain
Rationale: Evolutionary computation cohered adaptive mutation or variation rates that rise when population diversity falls, countering premature convergence.
Related originating lineages:
- Biology & Ecology — Population genetics supplied the variation-selection analogy and diversity measures used by evolutionary algorithms.
- Systems Thinking & Cybernetics — Feedback control supplied the automatic floor-triggered increase and relaxation loop.
Review resolution: Both current reviews place diversity_floor_rate_boost primarily in computer_science; the reconciled classification retains only lineages that materially shaped the mechanism and keeps breadth of origin separate from reach.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] Premature convergence — a search or evolving population settling on a local optimum before the space is adequately explored — is a standard failure in evolutionary computation, and raising the variation rate when population diversity drops is a recognised countermeasure. The rate boost applies the same logic outside code. ↩