Selection¶
Definition¶
Selection = differential persistence of constraint and information carriers competing for limited exergy.
Requires (Lewontin): variation, heredity, differential persistence. The logic itself is invariant (P11–P13); this chapter covers how it plays out in work systems.
D4. Selection favors whatever persists, not whatever is good for the larger system. [Established] "Good" enters only if the selector's measurement tracks it.
Selectors in economies¶
| Selector | Claims to reward | Actually measures |
|---|---|---|
| Customers | Delivered value | Willingness to pay, retention |
| Capital | Expected future flow | Growth metrics, narratives, comparables |
| Managers | Contribution | Visibility, reports, proxies |
| Labor market | Capability | Titles, years, interview performance |
| Regulators | Safety, compliance | Documentation, audits |
| Competitors | (indirect) Relative efficiency | Share shifts |
D5. Selection is only as good as its metrics. [Framing] Selection on proxies breeds proxy-optimizers. Gaming a metric changes the measurement, not the flow: a category error.
What weakens selection¶
- Slack: losing variants survive anyway.
- Delay: feedback arrives after cause and effect blur.
- Noise: measurement can't separate variants.
- Proxy drift: metric diverges from flow.
- Size buffering: large systems shield internal units.
- No variation: nothing to select between.
- Capture: selectors influenced by what they select.
Multi-level selection¶
D6. Lower-level persistence can undermine higher-level function. [Established in biology; Framing for organizations]
| Biological suppression | Function | Organizational analogue |
|---|---|---|
| Germline separation | Only some cells reproduce | Central control of hiring and budget |
| Apoptosis | Self-destruction when not needed | Sunset clauses, kill reviews |
| Immune policing | Remove cheaters | Audits, zero-based budgeting |
| Shared fate | Cells reproduce only via organism | Equity, team-level rewards |
Timescale mismatch¶
D7. Systems optimize for fast selectors and are destroyed by slow ones. [Hypothesis] Quarterly results, velocity and promotion cycles act fast; capability loss, debt and market shift act late.
Diagnostic¶
- Who is the actual selector?
- What do they actually measure?
- How fast and how accurately does that track real routing?
- What suppression exists against self-serving subsystems?