Approximation Target Divergence Mapping

Refine an approximation by mapping where it diverges from the target, then focus improvement effort on the most consequential gaps.

Solution archetype #

55

Problem family

Representation, Classification & Model Misfit

Problem subfamily

Abstraction, Reduction & Approximation Fidelity

The Diagnostic Story

Symptom: Refinement effort accumulates without the approximation getting meaningfully closer to fitting its purpose. One reviewer flags edge-case failures, another demands polish on already-acceptable parts, a third questions the whole architecture — and there is no shared map of which gaps actually matter. Average metrics look acceptable while the cases that count most keep slipping through.

Pivot: Stop treating improvement as undirected iteration and instead build a target-relative gap map: snapshot the current approximation against an explicit target, decompose divergence by dimension and context, and score each gap by how materially it threatens validity, safety, or usefulness. Effort flows to the highest-priority mismatches, while acceptable simplifications are documented and defended.

Resolution: Refinement loops converge faster because every action traces to a scored gap rather than salience or stakeholder pressure. Approximation limits become visible to reviewers and decision-makers, so acceptable deviations are tolerated rather than re-litigated. Persistent divergence triggers architecture or target review early, rather than surfacing as a late rejection.

Reach for this when you hear…

[ML engineering] “Our benchmark score keeps going up but the model still mishandles the exact cases the product team cares about — we're polishing the wrong gaps.”

[urban planning] “Every revision cycle someone adds a new concern and we lose track of which deviations from the original brief we actually signed off on.”

[clinical trials] “The protocol hit all the primary endpoints but the FDA asked why we never documented what the model was allowed to miss — and we didn't have an answer.”

When This Archetype Applies

Partial catalog groundingSome structural conditions are represented by existing abstractions, but no sufficient condition set is fully represented.

Diagnostic problem

An acknowledged approximation has a gap structure that is implicit, unevenly perceived, or poorly prioritized, causing actual refinement to follow local cues rather than the gaps that govern fitness to the target.

What this problem means

A model, prototype, plan, performance, forecast, classification, or artifact is recognized as approximate, but the divergence between the approximation and the target is implicit, unevenly perceived, or poorly prioritized. Teams then refine by habit, salience, stakeholder pressure, available tools, or aesthetic polish rather than by the gap structure that determines whether the approximation is becoming more fit for purpose.

Show the applicability expression

Applicability expression5 distinct conditions

Ungoverned refinement prioritiesandany oneMultidimensional target gapsorUnequal deviation severityorConflicting fit criteriaorLocal-global fit divergence

Algebraic1∧(A∨B∨C∨D)

groundedpartly groundedopen

Equivalent to the 4 condition sets it replaces, with 3 duplicate condition cards removed.

1Required in every casenumbered 1–1

These hold no matter which pattern applies.

1

Ungoverned refinement priorities · open

The condition-specific gap information remains insufficiently visible or prioritized to govern refinement, causing actual refinement to follow local cues instead of the gap structure that determines target fitness. more

and also

4At least one of theselettered A–D

Any single one of these completes the pattern.

A

Multidimensional target gaps · open

The current approximation is known to depart from its target in several distinct ways. more

or — any one satisfies this slot

B

Unequal deviation severity · grounded

Some approximation deviations are benign simplifications while others threaten validity, safety, value, or acceptance. more

or — any one satisfies this slot

C

Conflicting fit criteria · open

Different stakeholders use conflicting criteria for judging whether the approximation is closer to its target. more

or — any one satisfies this slot

D

Local-global fit divergence · grounded

An approximation keeps improving locally without a corresponding improvement in whole-target fit. more

Other requirements and context (2)

Why these sit outside the expression

Application gate — it governs whether applying the archetype is appropriate or material, rather than defining the structural problem itself.

Solution feasibility — it describes whether the intervention can work, not whether the diagnostic problem exists.

• Application gateRefinement effort is scarce and not all gaps can be closed at once.

• Solution feasibilityThe target itself is measurable only through proxies, benchmarks, or qualitative criteria.

Coverage

2 of 5 conditions grounded · 3 open.

Read the methodology · Download the trigger-logic data

Mechanisms / Implementations

• Checkpointed Convergence Review: Re-snapshots the approximation at fixed checkpoints to confirm it is still converging on the target — and to trigger a stop-or-escalate when it is not.
• Refinement Backlog Prioritization: Turns scored gaps into an ordered refinement backlog by weighting each by consequence, tractability, confidence, and whose stake it serves.
• Regression-Guarded Refinement Cycle: Executes prioritized refinements one at a time behind a regression guard, so closing one gap can never silently reopen a gap that was already within tolerance.
• Residual Error Heatmap: Renders the divergence map as a colored field so the eye lands first on where residual error is largest — with a confidence overlay showing how far each cell can be trusted.
• Side-by-Side Target Delta Review: Places the intended target and the current approximation next to each other, dimension by dimension, so every consequential delta becomes visible and nameable.
• Tolerance-Band Gap Scoring: Scores each divergence against its acceptable-error band, separating harmless simplifications from out-of-tolerance gaps that actually need repair.

Related Abstractions

Abstractions this archetype builds on — directly (a source ingredient) or as a related pattern. Links follow the typed catalog namespace.

Built directly on (1)

• Refinement: Iteratively improving a candidate solution toward adequacy through repeated cycles of evaluation and adjustment that narrow the gap to a target, rather than deriving the answer in one shot.

Also references 26 related abstractions

• Approximation: Good-enough representation.
• Boundedness: Values remain within limits.
• Calibration: Aligning a system's output to a trusted reference by measuring deviation, adjusting to reduce it, and monitoring for drift.
• Convergence: Movement toward stable state.
• Cost–Benefit Analysis: Evaluate decisions.
• Data Integrity: Accuracy and consistency preserved.
• Diminishing Incremental Gains: Reduced benefit per unit.
• Divergence-Convergence in the Design Process: Expand then refine ideas.
• Engineering Tolerances: Acceptable variation.
• Error Proofing (Poka-Yoke): Error prevention.

▸ Show 16 more

Variants

Narrower or domain-specific specializations that share this archetype's core structure. Recognized variants are established; candidate variants are provisional.

Metric Residual Divergence Mapping · subtype · recognized

Map quantitative residuals between a model or process output and target values to guide selective refinement.

Qualitative Target-Gap Mapping · domain variant · recognized

Map divergence from a qualitative, interpretive, aesthetic, or experiential target using criteria rather than pure metrics.

Benchmark-Anchored Divergence Mapping · validation variant · candidate

Use benchmark cases or reference examples to decide which approximation-target divergences require refinement.

Regression-Guarded Refinement · risk or failure variant · candidate

Close one target gap while explicitly checking that already acceptable dimensions do not regress.

Editorial Notes

Problem Classification

Classification: Representation, Classification & Model Misfit → Abstraction, Reduction & Approximation Fidelity

Problem kernel: approximation error is unstructured and misprioritized

Rationale: A known approximation lacks an explicit map of where and how it diverges from the target, so refinement follows salience rather than task-relevant fidelity.

Independent corroboration: The earliest necessary condition in the frozen evidence is: A model, prototype, plan, performance, forecast, classification, or artifact is recognized as approximate, but the divergence between the approximation and the target is implicit, unevenly perceived, or poorly prioritized. That is a abstraction reduction and approximation fidelity problem because A compressed, approximate, sparse, or proxy representation drops or weights structure without a traceable account of task-relevant loss.

Review outcome: Independent reviewer agreement; high confidence.