Occam-Style Model Selection¶
Method — instantiates Parsimony Filter
Compares candidate models or explanations and favors the one with fewer assumptions when adequacy is otherwise comparable.
Occam-Style Model Selection is the comparison rule you reach for when several candidate explanations, models, or hypotheses all fit the evidence about equally well: it favors the one that rests on the fewest unsupported assumptions or entities, and it holds that choice only until new evidence forces a richer account. Its defining move is that it is comparative and non-destructive — it ranks whole rival candidates against each other under an equal-adequacy condition and discards the losers wholesale; it never reaches inside a candidate to remove a part. Assumption count is the decision variable, but only as a tie-breaker: parsimony earns the vote solely among candidates that already explain the evidence comparably. That "leanest of the equally adequate, provisionally" is what makes it this mechanism and not surgery on a single built thing.
Example¶
A payments service went dark for twenty minutes overnight. By morning the incident channel holds two root-cause stories, and the logs are consistent with both. Story A: a single TLS certificate on the gateway expired at midnight. Story B: a config drift the week before, plus a rare race condition in the retry logic, plus a coincident DNS blip at the upstream provider, lined up to produce the same outage window.
Both explanations "fit" — each accounts for the observed failure and recovery. Occam-Style Model Selection makes the choice explicit rather than intuitive. First it confirms the adequacy is genuinely comparable: both stories reproduce the timeline. Then it enumerates what each requires you to believe that is not yet evidenced — Story A needs one unverified premise (the cert had expired), directly checkable; Story B needs three independent failures to have coincided, none of them independently attested. Equal adequacy, three unsupported assumptions against one: prefer A, and go check the cert. It had expired. The add-back threshold is stated before anyone walks away: if the cert turns out to have been valid, Story B's multi-factor account comes back on the table. The elaborate story felt more thorough, but thoroughness that multiplies unproven entities is exactly what the rule is built to resist.
How it works¶
- Establish adequacy parity. Confirm the candidates actually explain the evidence about equally well; if one fits materially better, adequacy decides and parsimony never gets a vote.
- Enumerate each candidate's assumptions. List the unsupported premises and extra entities each explanation must invoke, and count the load-bearing ones — the count, not eloquence, is the comparison.
- Prefer the leanest of the equals. Among candidates of comparable adequacy, select the one requiring the fewest unsupported assumptions.
- Name the add-back trigger. State what evidence would break the tie the other way and force the richer candidate back, so the choice is explicitly provisional.
Tuning parameters¶
- Adequacy-parity bar — how close two candidates' fit must be before assumption-count is allowed to decide. Set it loose and parsimony overrides real explanatory differences; set it strict and the rule rarely fires.
- Assumption unit — what counts as "an assumption": a premise, an entity, a free parameter, a causal link. The choice decides which candidate looks simpler.
- Tie tolerance — how large an assumption-count gap must be to matter. A hair-thin margin invites over-reading; a wide one ignores genuine parsimony differences.
- Add-back sensitivity — how much new evidence it takes to reinstate the more complex candidate. Twitchy reinstatement chases noise; sticky reinstatement clings to a story the evidence has outgrown.
When it helps, and when it misleads¶
Its strength is disciplining the pull toward explanations that feel rigorous because they are elaborate — the conspiracy-thick account that multiplies unproven entities to cover every detail. As a cheap tie-breaker among adequacy-equivalent candidates, it is the standing corrective that Occam's razor names.[n1]
Its central failure mode is underfitting: penalizing assumptions when the true story really is complex, so that "fewer assumptions" is mistaken for "more likely true" and a genuinely multi-causal reality is flattened into a tidy single cause. The classic misuse is invoking parsimony before adequacy is settled — using "the simpler explanation" to wave away a candidate that actually fits the evidence better. The guarding discipline is the two conditions built into the rule: assumption-count only votes among the equally adequate, and the choice is always provisional, reinstated the moment evidence demands the richer account.
How it implements the components¶
assumption_list— it enumerates the unsupported premises and extra entities each candidate requires; the count of load-bearing ones is the decision variable.adequacy_criterion— it gates the whole comparison on equal adequacy, so parsimony decides only among candidates that already explain the evidence comparably.complexity_reintroduction_threshold— it names, up front, the evidence that would break the tie the other way and force the more complex candidate back.
It runs no necessity_test on the internal parts of one design and keeps no removal_log of what it stripped, because it removes nothing — it discards whole losing candidates; that part-by-part removal from a built product is Feature Pruning, its nearest twin. It also stands up no simple_baseline artifact of its own to beat (that is Simple Baseline Model, which supplies one of the candidates it ranks).
Related¶
- Instantiates: Parsimony Filter — Occam-Style Model Selection is the filter applied as a choice among rival candidates rather than a cut within one.
- Consumes: Simple Baseline Model — the baseline is often the leanest candidate this rule ranks the others against.
- Sibling mechanisms: Feature Pruning · Lean Design Review · Scope Reduction Review · Simple Baseline Model · Assumption Audit · Minimum Viable Explanation · Model Complexity Penalty
Editorial Notes¶
Form Classification¶
Form family: Decision, Gate & Allocation
Rationale: Occam-Style Model Selection operates as a case-specific gate, selection, routing, prioritization, or resource disposition because it compares candidate models or explanations and favors the one with fewer assumptions when adequacy is otherwise comparable.
Independent corroboration: The frozen evidence defines Occam-Style Model Selection as 'Compares candidate models or explanations and favors the one with fewer assumptions when adequacy is otherwise comparable', so its operative form is Decision, Gate & Allocation.
Nearest alternative: Analysis, Modeling & Optimization — Occam-Style Model Selection includes features of an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution, but its defining operation is a case-specific gate, selection, routing, prioritization, or resource disposition.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Philosophy
Origin pattern: Convergent development
Present-day reach: Universal
Rationale: Philosophy of science supplied Occam's razor: among adequately explanatory accounts, prefer the one that multiplies assumptions least.
Related originating lineages:
- Statistics & Experimental Design — Statistical model selection operationalized parsimony through penalties, information criteria, and out-of-sample adequacy.
Review resolution: Both independent reviews agree on primary origin philosophy; reconciliation resolves encyclopedia_synthesis_disagreement. Formative alternate lineages retained: statistics_experimental_design. The broader reach of later applications is kept separate as domain_reach=universal; origin_mode=convergent describes the historical relationship among lineages. Confidence is conservatively reconciled to high, and encyclopedia_synthesis=true preserves the reviewers' boundary judgment.
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¶
Occam-Style Model Selection chooses among explanations; Minimum Viable Explanation states one for a given audience. The first ranks rival hypotheses by assumption count and keeps the winner; the second takes an already-selected account and pares its telling down to what a listener needs. They compose — select the leanest adequate hypothesis, then narrate it minimally — but confusing them turns a selection rule into a communication style.
[n1] Occam's razor (the law of parsimony), attributed to the 14th-century friar William of Ockham: entities should not be multiplied beyond necessity. It is a heuristic for choosing among adequate explanations, not a law of nature — which is why this mechanism applies it only under equal adequacy and only provisionally. ↩