Convergence Logic Rubric¶
Decision rubric — instantiates Recursive Triangulation of Triangulation
Fixes in advance how agreement, conflict, outliers, and missing evidence should move confidence, so convergence is interpreted by rule rather than by mood.
Triangulation is only as trustworthy as the rule used to read it, and most teams have no explicit rule — they treat agreement as proof, conflict as an inconvenience to be argued away, and missing evidence as if it were neutral. The Convergence Logic Rubric writes that rule down before the results are known: for a claim of a given required-confidence level, it states how full agreement, partial agreement, a lone outlier, a direct conflict, and a silent gap should each translate into a confidence move. Its defining feature is predeclaration against the claim — the thresholds are pinned to what the original triangulation was supposed to establish and for what decision, so that the interpretation of convergence cannot be reverse-engineered to flatter whatever the evidence happened to show.
Example¶
An election-forecasting desk aggregates three polls of a close race that all point the same way. The temptation is to call the agreement decisive. Before the polls come in, the desk instead sets a convergence rubric tied to its claim — "estimate the vote share within two points, for a public forecast that must not overclaim." The rubric declares: agreement counts as strong only if the polls use different sampling frames; a single outlier is investigated, not discarded, and if it uses the more independent method it gets extra weight; a conflict wider than the margin of error caps confidence regardless of how many polls agree; and a missing pollster who declined to publish is treated as a possible suppressed null, not as absence.
When the results arrive, two of the three "agreeing" polls turn out to share an online panel vendor, and the rubric's different-frames clause docks their combined weight. The lone outlier used live callers — the more independent frame — so it is up-weighted rather than dismissed. The rubric hands the desk a calibrated confidence, not a false one, and does it by rule the desk cannot now bend.
How it works¶
- Anchor to the claim. State the target proposition and its required confidence level first; the rubric's thresholds are only meaningful relative to what confidence is for.
- Predeclare the mappings. For each pattern — full agreement, partial agreement, outlier, conflict, missing evidence — write the confidence move it produces, before the data is in.
- Weight by independence, not count. Tie the strength of "agreement" to how independent the agreeing streams are, so that redundant streams cannot inflate the move.
- Audit the original against the rubric. Replay how the first-order procedure actually handled its agreements and conflicts, and mark every place it departed from what a defensible rule would require.
Tuning parameters¶
- Confidence-move granularity — coarse bands (support / weaken / neutral) or a fine numeric scale. Fine scales feel precise but invite false precision the evidence cannot bear.
- Outlier policy — investigate-and-reweight versus discard-if-isolated. Reweighting preserves signal but costs effort; discarding is cheap and dangerous.
- Conflict cap — how hard a genuine conflict caps confidence regardless of agreement elsewhere. A hard cap is conservative; a soft one lets weight of numbers override a real contradiction.
- Missing-evidence stance — treat gaps as neutral, as mild warnings, or as suppressed nulls. The more suspicious the stance, the more it guards against selective reporting at the cost of stalling on incomplete data.
When it helps, and when it misleads¶
Its strength is that it converts convergence from a feeling into an auditable calculation, and it neutralizes the two commonest interpretive sins — counting agreement as proof and waving away conflict. Predeclaring the rule is the same discipline that preregistration brings to analysis: fixing the decision rule before the data removes the freedom to interpret results into the answer you wanted.[n1]
Its failure mode is a rubric that is precise but wrong: a rule can be followed faithfully to a confident conclusion if the rule itself encodes a bad theory of evidence, and its very explicitness lends that error an air of rigor. A rubric can also be gamed by an author who tunes the thresholds, under the cover of "predeclaration," to license a foreordained result. The guarding discipline is to set the rubric independently of anyone invested in the outcome and to keep it fixed once the results are unblinded — a rule changed after seeing the data is no longer a rule.
How it implements the components¶
The rubric fills the interpretation-and-claim pair of the archetype's machinery, not the whole menu:
convergence_rule_audit— its core: the explicit, predeclared rule for how agreement, conflict, outliers, and gaps move confidence, and the audit of the original procedure against that rule.triangulation_claim_definition— the rubric's thresholds are pinned to the claim's required confidence level and decision context, so interpretation is calibrated to what confidence is for.
It does not decide when to stop recursing (recursive_stop_rule) — that depth call is Meta-Validation Stop Gate; the rubric says how to read the evidence, not when to stop gathering it.
Related¶
- Instantiates: Recursive Triangulation of Triangulation — the rubric supplies the interpretation rule the whole meta-audit depends on.
- Consumes: Triangulation Dependency Matrix — its independence findings set how much a given agreement is allowed to count.
- Sibling mechanisms: Triangulation Dependency Matrix · Method Provenance Review · Independent Meta-Review Panel · Second-Order Replication Probe · Triangulation Red-Team Review · Meta-Validation Stop Gate · Confidence Scope Update Memo
Editorial Notes¶
Form Classification¶
Form family: Rule, Policy & Commitment
Rationale: Fixes in advance how agreement, conflict, outliers, and missing evidence should move confidence, so convergence is interpreted by rule rather than by mood, making its operative form a standing rule, threshold, contractual commitment, or policy constraint governing future conduct.
Independent corroboration: The frozen evidence defines Convergence Logic Rubric as 'Fixes in advance how agreement, conflict, outliers, and missing evidence should move confidence, so convergence is interpreted by rule rather than by mood', so its operative form is Rule, Policy & Commitment.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Ethnography & Qualitative Methods
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Triangulation methodology established explicit rules for how agreement, conflict, outliers, and missing evidence change warranted confidence.
Related originating lineages:
- Philosophy — Formal epistemology contributes treating evidence states as defeasible confidence updates rather than mood or proof.
- Statistics & Experimental Design — Preanalysis plans supplied precommitment before outcomes are known.
Review resolution: Qualitative triangulation supplies explicit treatment of agreement, divergence, and negative cases; philosophy and preanalysis planning supply defeasible updating rules and precommitment.
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:
- NIH Best Practices for Mixed Methods Research
- National Academies Evidence-Based Practice for Public Health Emergency Preparedness and Response
Notes¶
[n1] Preregistration is the practice of specifying a study's hypotheses and analysis rules before collecting or examining the data, removing the researcher's freedom to choose, after the fact, the analysis that yields the desired result (a defense against "researcher degrees of freedom" and HARKing). The convergence rubric applies the same predeclaration to the interpretation of triangulated evidence. ↩