Case Similarity Rubric¶
Scoring rubric — instantiates Nearest-Exemplar Response Reuse
A fixed, weighted scoring sheet that grades how well one candidate exemplar fits the new case and flags the mismatches that should veto reuse regardless of the score.
A Case Similarity Rubric takes a candidate exemplar that has already been surfaced and grades it. It is not a search engine and it does not rank the whole store; it is the auditable scoring instrument applied to a specific new-case-versus-prior-case pairing. Its two moving parts are a set of weighted criteria that roll a comparison up into a fit-and-quality grade, and a set of blocking flags — the mismatches that, if tripped, veto reuse no matter how high the numeric score. That second part is what keeps a rubric from being a mere average: a candidate can score 85 on the weighted sheet and still be disqualified because one decisive difference is present. Everything about it is legible — anyone can read which criteria carried weight and which flag tripped — which is exactly what a learned metric cannot offer and exactly why the rubric exists alongside one.
Example¶
An insurance claims team is setting the reserve on a new bodily-injury claim by comparison to a settled prior claim that a retrieval step surfaced as similar. The rubric lists weighted criteria: injury severity (30%), liability clarity (25%), jurisdiction (20%), treatment type (15%), claimant age band (10%). The adjuster scores the surfaced prior claim against the new one and it comes out to 78/100 — a strong fit on paper. Consistency and predictability are the goal: two adjusters comparing the same pair should reach the same grade.
But the rubric also carries a blocking flag: different injury category (e.g., soft-tissue vs. surgical) → do not reuse the reserve figure. The prior claim resolved without surgery; this one involves a planned operation. The flag trips. Despite the 78, the rubric's verdict is that the prior claim may be reused as a comparable for liability assessment but not for the reserve amount, and it records the tripped flag as the reason. The adjuster sets the reserve by a different route, and the file shows an outside auditor exactly why the tempting 78-point match was not used.
How it works¶
- Score against fixed criteria. Each pre-agreed criterion is rated and combined by its weight into a single fit-and-quality grade — the same criteria every time, so grades are comparable across cases and assessors.
- Apply blocking flags. A separate list of decisive mismatches is checked independently of the score; any one of them vetoes reuse (or narrows it to part of the response) regardless of how high the grade.
- Record the reasons. The output is not just a number but the criterion breakdown and any tripped flag, so the grade is an auditable argument rather than a verdict.
- Grade one pairing, not the field. The rubric assumes a candidate has been surfaced; it evaluates that pairing rather than searching for it.
Tuning parameters¶
- Criterion weights — how much each dimension counts. Reweighting is the main lever, and the main attack surface: weights can be tuned until a favored candidate passes.
- Blocking-flag set — which single mismatches are disqualifying. A rich flag set catches decisive differences the weighted score would drown out, but too many flags disqualify nearly everything.
- Pass threshold — the fit grade below which reuse is refused even absent a flag.
- Aggregation rule — whether the score is a straight weighted sum or a veto-dominant combination where flags override arithmetic.
- Recalibration cadence — how often criteria and weights are revisited against real outcomes.
When it helps, and when it misleads¶
Its strength is consistency and auditability: it forces "how similar is close enough?" out of intuition and onto a sheet anyone can read, teach, and contest. By pricing decisive differences as blocking flags, it directly resists surface-match overreach — the seductive candidate that shares surface features but differs on the one dimension that matters.[n1]
Its failure mode is the classic one for any scoring instrument: rigidity and false precision, a tidy number standing in for a judgment it cannot really make, plus staleness as criteria drift from reality and gaming as assessors learn which weights to push. The signature misuse is adding up the weighted score and ignoring the tripped blocking flag — treating an 85 as a green light when a disqualifying mismatch is sitting right there. The guarding discipline is to make blocking flags veto-dominant over the arithmetic, to recalibrate the criteria against actual outcomes rather than freezing them, and to treat the grade as a structured argument to be checked, not a verdict to be obeyed.
How it implements the components¶
case_quality_label— grades how good the candidate match is against fixed weighted criteria, producing an auditable fit-and-quality score for the specific pairing.countercase_and_exception_probe— the blocking-flag list: decisive mismatches that veto reuse regardless of the numeric grade.
It does not compute an automated similarity_metric — that's Similarity Search over Case Embeddings; the rubric's weights are hand-set and legible rather than learned, and it grades one surfaced pairing rather than ranking the store. It does not elicit an expert's tacit recall via query_case_profile — that's Expert Case Recall Checklist, which interrogates the assessor's own memory rather than scoring an assigned candidate — and it does not persist outcome_feedback_update_record (Exemplar Feedback Registry).
Related¶
- Instantiates: Nearest-Exemplar Response Reuse — this is the archetype's auditable fit-scoring instrument.
- Sibling mechanisms: Case-Based Reasoning System · K-Nearest-Neighbor Case Matcher · Similarity Search over Case Embeddings · Precedent Matching Workflow · Incident Playbook Lookup · Expert Case Recall Checklist · Exemplar Feedback Registry
Editorial Notes¶
Form Classification¶
Form family: Assessment, Review & Assurance
Rationale: A fixed, weighted scoring sheet that grades how well one candidate exemplar fits the new case and flags the mismatches that should veto reuse regardless of the score, making its operative form a bounded evaluation of existing evidence or work that produces a finding or disposition.
Independent corroboration: The frozen evidence defines Case Similarity Rubric as 'A fixed, weighted scoring sheet that grades how well one candidate exemplar fits the new case and flags the mismatches that should veto reuse regardless of the score', so its operative form is Assessment, Review & Assurance.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Cognitive Science
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Structure-mapping and analogy research distinguishes relational similarity from superficial feature overlap when transferring a prior case.
Related originating lineages:
- Computer Science & Software Engineering — Case-based reasoning operationalizes feature and relation matching for retrieval.
- Education & Pedagogy — Rubric design makes comparison criteria explicit and reviewable for learners and evaluators.
- Statistics & Experimental Design — Weighted distances and similarity measures contribute formal aggregation across features.
Review resolution: Cognitive science is primary because structural alignment and analogy distinguish deep relational similarity from surface resemblance. Computer science, education, and statistics contribute case retrieval, explicit assessment criteria, and weighted distance, making the rubric a cross-disciplinary synthesis.
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¶
[n1] The distinction between surface similarity (shared features and vocabulary) and structural similarity (shared relational form), central to Gentner's structure-mapping account of analogy. Blocking flags are how a rubric encodes "these look alike on the surface but differ where it structurally counts." ↩