Diagnostic Differential¶
Method — instantiates Cautious Pattern Completion
Keeps several rival explanations live and drives toward the one discriminating test that separates them, updating each rival's likelihood as evidence lands.
A Diagnostic Differential takes a slate of rival explanations and does the one thing a list cannot: it narrows the slate by evidence. Its defining move is the search for the discriminating observation — the test whose result would rule one rival out while leaving another standing — and the disciplined updating of each rival's likelihood as those results arrive. Where a list only guarantees that alternatives exist, the differential is the engine that converts alternatives into a shrinking, evidence-weighted ranking, never collapsing to a single answer until an observation actually excludes the competitors. It is a method for choosing among hypotheses, not for generating or merely holding them.
Example¶
A patient arrives in an emergency department with chest pain. The instant pattern-completion is "heart attack," but the attending physician runs a differential instead: acute coronary syndrome, pulmonary embolism, aortic dissection, pericarditis, and musculoskeletal strain all go up as live rivals. The method's whole value is in what it asks next — not "what evidence supports a heart attack?" but "what single test would most separate these?" A D-dimer, an ECG, and a focused history each shift the picture differently: a normal ECG lowers the coronary rival without eliminating it; a widened mediastinum on chest film pushes dissection sharply upward; reproducible pain on palpation raises the musculoskeletal rival.
Each result is recorded as an update, not a verdict — the coronary rival's likelihood drops from "leading" to "still possible," dissection climbs from "unlikely" to "rule out now." The method keeps every rival on the board with an explicit, revised weight until one discriminating result finally excludes the rest. The outcome is not a faster guess; it is a defensible narrowing in which the reason each rival rose or fell is written down and checkable.
How it works¶
The differential is built around separation, and its steps differ from ordinary evidence-gathering in one respect: they optimize for discrimination rather than support.
- Ask what would distinguish, not what would confirm. For each pair of live rivals, name the observation whose result they predict differently. The highest-value next test is the one that splits the field most.
- Order tests by discriminating power and cost. A cheap test that halves the field beats an expensive one that only nudges the leader.
- Update every rival, not just the favorite. Each result revises the whole ranking; a finding that fails to move a rival is itself informative.
- Refuse premature collapse. The method holds the field open until an observation excludes competitors — likelihood shifts alone do not close it.
Tuning parameters¶
- Discrimination threshold — how much a candidate test must separate the rivals before it is worth running. High thresholds save effort but leave the field wide; low thresholds chase small distinctions and burn resources.
- Update granularity — whether likelihoods move in coarse steps (up / down / out) or on a fine scale. Fine updates capture more but invite false precision on evidence that cannot support it.
- Rival-retention floor — how low a rival's likelihood may fall before it is dropped rather than merely down-weighted. A low floor guards against missing the rare cause; a high floor keeps the working set manageable.
- Test sequencing — parallel ordering of discriminating tests versus a strict one-at-a-time cascade. Parallel is faster; sequential lets an early result cancel later, costlier tests.
When it helps, and when it misleads¶
Its strength is that it turns "which explanation?" into a series of answerable, discriminating questions, and it makes the narrowing auditable: anyone can see which observation moved which rival and by how much. It is the natural antidote to premature closure — the diagnostic error of settling on the first plausible explanation and then reading all further evidence as confirmation.[n1]
Its failure modes are subtler than a list's. A differential can quietly become confirmation in disguise when the "discriminating" tests chosen are actually the ones expected to support the favorite; the discipline is dead if every test is one the leading rival passes. It can also anchor on the initial ranking so that later updates are too small to dislodge a first impression, and it can retain so many rivals that no test ever discriminates enough to close. The guarding discipline is to insist each next test could plausibly demote the current leader, and to make the exclusion — not the accumulation of support — the only thing that ends the differential.
How it implements the components¶
missing_evidence_marker— the differential's core output is the named discriminating test: for each pair of live rivals, the specific observation that is not yet available but would separate them.confidence_update_record— every rival carries a likelihood that is explicitly revised as each result lands, and the revision (and its reason) is recorded rather than silently overwritten.
It consumes a slate rather than producing one: it does not enumerate the rivals or bound the raw input — that is candidate_completion and alternative_completion_set, held by Hypothesis List, its nearest twin. Nor does it mount a one-sided disconfirmation_probe against a single leading completion — that is Disconfirming Evidence Search, which attacks one story where the differential separates many.
Related¶
- Instantiates: Cautious Pattern Completion — the differential is the narrowing engine that keeps completion evidence-weighted and revisable.
- Consumes: Hypothesis List supplies the slate of rivals the differential discriminates among.
- Sibling mechanisms: Hypothesis List · Disconfirming Evidence Search · Hallucination Check · Reconstruction Note · Source-Tracing Table · Uncertainty Tagging · Withhold-Conclusion Checkpoint · Assumption Log
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Diagnostic Differential operates as a computation, comparison, model, or analytic representation used to infer, estimate, or choose because it keeps several rival explanations live and drives toward the one discriminating test that separates them, updating each rival's likelihood as evidence lands.
Independent corroboration: The frozen evidence defines Diagnostic Differential as 'Keeps several rival explanations live and drives toward the one discriminating test that separates them, updating each rival's likelihood as evidence lands', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Medicine & Healthcare
Origin pattern: Single lineage
Present-day reach: Multi-domain
Rationale: Clinical medicine cohered the differential as an evidence-updated set of rival diagnoses narrowed by discriminating observations.
Related originating lineages:
- Philosophy — Abductive and eliminative reasoning supplied the general logic of choosing among live hypotheses.
Review resolution: Clinical medicine cohered the differential as an evidence-updated set of rival diagnoses narrowed by discriminating observations. The retained alternate lineages materially shaped the mechanism's form.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] Premature closure — in the diagnostic-error literature (Pat Croskerry and others), the tendency to stop considering alternatives once a plausible diagnosis comes to mind, so that subsequent evidence is fit to the chosen answer rather than used to test it. The differential exists to postpone closure until an observation earns it. ↩