Model-Class Revision¶
Representation revision — instantiates Meta-Symbolic Rule Reflection
Reconsiders the whole class of models or representational primitives allowed for a problem when the current class simply cannot express what turns out to matter.
A Model-Class Revision changes not a value inside a model, nor a category within a scheme, but the class of models itself — the representational primitives a problem is allowed to use. Its defining move is expressive: it is triggered when anomalies reveal that the current class cannot say the thing that matters — a linear model that has no way to represent a feedback loop, a single-label scheme that cannot hold a case with two causes, a static-state model with no vocabulary for transitions. The revision proposes a richer class (linear → network, states → transitions, single-label → relational) and checks what that richer class newly makes expressible, and at what cost. It is the deepest of the revisions because it changes what kinds of structure can be represented at all.
Example¶
An ecology group has modeled a lake food system for years as a food chain: algae → zooplankton → small fish → pike, each level eating the one below. The model keeps failing to predict crashes. When a mild season collapses the pike, the small-fish population also falls instead of booming — impossible in a linear chain. The anomaly is not a bad parameter; it is that the chain class has no way to represent an organism that both preys on and competes with another, or a species eaten by two predators. The group runs a model-class revision. They map the current primitives — a strict linear ordering of trophic levels — and name what it structurally cannot express: cross-links, omnivory, indirect effects.
They propose moving to a food-web network class, where any species can link to any other and effects propagate along multiple paths. Before committing, they trace the consequences: the network class newly predicts the small-fish decline (pike loss releases a competitor that suppresses them), but it also demands far more data to parameterize and can no longer be solved by hand. The revision is adopted because the new class expresses the structure the crashes hinge on — with eyes open to the tractability it costs.
How it works¶
- Diagnose an expressive failure, not a fit failure. The trigger is an anomaly the current class cannot represent at all, distinguished from one it merely fits badly — the latter needs a better parameter, not a new class.
- Map the primitives and their ceiling. The revision names the current representational primitives and the exact structure they structurally exclude (no cycles, one label per case, no time).
- Propose a class that lifts the ceiling. It nominates a richer class and specifies the new primitive that carries the missing structure (a link, a transition, a relation).
- Trace what the new class newly implies — and costs. It follows the richer class into the predictions and behaviors it changes, including the tractability, data, and interpretability it spends.
Tuning parameters¶
- Expressive reach — how much more the new class can say; a modest extension or a wholesale change of primitives. More reach captures more structure but costs tractability and interpretability.
- Parsimony pressure — how hard you resist adding expressive power; strict pressure keeps models usable, loose pressure risks a class so rich it fits anything and predicts nothing.
- Tractability floor — the minimum solvability/estimability the new class must retain (closed-form, simulable, estimable-from-available-data). A high floor rules out powerful but unusable classes.
- Migration scope — whether existing results must translate into the new class or can be rebuilt from scratch. Translation preserves continuity but constrains how different the new class can be.
When it helps, and when it misleads¶
Its strength is escaping problems that are unsolvable in principle within the current class — no amount of tuning a food chain will ever predict an indirect effect, because the chain has no vocabulary for one. Tracing the richer class's consequences before adopting it is what keeps the move honest: it earns the new primitives against the specific structure they let you represent.
It misleads through the seduction of expressive power. A richer class can fit the past so well it stops constraining the future — the trap behind "all models are wrong, but some are useful"[1] is that the most elaborate class is often the least useful, buying flexibility with the ability to be wrong. Moving to network or relational primitives can also explode the data appetite and destroy the interpretability that made the model decision-ready. The classic misuse is jumping to a fashionable rich class (a neural or agent model) because the old one felt crude, not because a specific structure demanded it. The guarding discipline is to require a named piece of structure the current class cannot express, and to trace the new class's cost in data and tractability before adopting it.
How it implements the components¶
symbol_system_map— it names the current representational primitives and, crucially, the exact structure they cannot express, as the object under revision.revision_proposal— its output: a nominated richer class of models and the new primitive (link, transition, relation) that carries the missing structure.downstream_consequence_trace— it follows the proposed class into the predictions and behaviors it changes, and the tractability, data, and interpretability it spends.
It does not revise the entities and relations *within a fixed representational system while preserving continuity (legacy_mapping_table, change_control_record) — that within-system work is Ontology Revision Session, whereas this mechanism swaps the class of representation itself.*
Related¶
- Instantiates: Meta-Symbolic Rule Reflection — it revises the class of representational primitives when the current class cannot express relevant structure.
- Sibling mechanisms: Ontology Revision Session · Taxonomy Redesign Workshop · Category Boundary Audit · Edge-Case Walkthrough · Language / Rule Critique · Governance Rule Review · Paradigm Review · Meta-Cognitive Reflection Prompt · Versioned Schema Change
Editorial Notes¶
Form Classification¶
Form family: Intervention, Treatment & Transformation
Rationale: Model-Class Revision operates as a direct treatment or transformation intended to change the target state or representation because it reconsiders the whole class of models or representational primitives allowed for a problem when the current class simply cannot express what turns out to matter.
Independent corroboration: The frozen evidence defines Model-Class Revision as 'Reconsiders the whole class of models or representational primitives allowed for a problem when the current class simply cannot express what turns out to matter', so its operative form is Intervention, Treatment & Transformation.
Nearest alternative: Analysis, Modeling & Optimization — The revision uses expressive analysis, but its defining operation directly replaces the representational primitives and model class allowed for the problem.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Historically ambiguous
Present-day reach: Multi-domain
Rationale: Changing functional/model class after structural misfit is most directly a statistical model-selection operation; philosophy of science and data science supply the explanatory and computational lineages. This establishes statistics_experimental_design as the primary origin lineage rather than merely a domain where the mechanism is now applied.
Related originating lineages:
- Data Science & Analytics — Machine learning operationalizes model-class choice across architectures and hypothesis spaces.
- Philosophy — Revising the representational vocabulary when a theory cannot express relevant phenomena is rooted in philosophy of science and conceptual change.
Review resolution: Authoritative/primary-source research resolves the conflicting primary-origin claims in favor of statistics_experimental_design: Changing functional/model class after structural misfit is most directly a statistical model-selection operation; philosophy of science and data science supply the explanatory and computational lineages. Retained alternate origins (philosophy, data_science) are limited to independently formative or materially shaping lineages supported by the reviewer evidence; downstream adoption alone was not promoted to origin. The breadth of present-day use is recorded separately as domain_reach=multi_domain. origin_mode=historically_ambiguous, confidence=medium, and encyclopedia_synthesis=true reflect the surviving provenance evidence and the encyclopedia's generalization.
Attribution caveat: The mechanism deliberately abstracts across scientific and cognitive versions of representational revision. The generalized formulation deliberately spans statistical model selection and philosophical representation critique.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Researched adjudication after independent review; medium confidence.
Sources consulted:
- NIST/SEMATECH e-Handbook of Statistical Methods — Authoritative statistical reference for decomposition, fitting, calibration, model selection, diagnostics, and validation.
References¶
[1] Box, G. E. P. "Science and Statistics". Journal of the American Statistical Association 71(356), 791–799 (1976). Argues that because all models are wrong, excessive elaboration cannot make one correct and economical models should be preferred. registry ↩