Confidence Update Worksheet¶
Template — instantiates Independent Evidence Triangulation
A structured record of prior confidence, stream-specific likelihoods, dependency discounts, contradictions, and sensitivity that resolves to a single bounded confidence claim.
Once the streams are inventoried and their disagreements investigated, someone still has to say how confident we now are. The Confidence Update Worksheet is the template that does the arithmetic of belief in the open. It records where confidence started, how much each stream should move it, what to subtract when two streams turn out to lean on the same lineage, and what remains unresolved — and it resolves all of that into one bounded confidence statement with its assumptions attached. Its defining trait is that it is the terminal artifact of triangulation: it does not gather evidence or classify agreement, it converts an already-assembled, already-vetted evidence picture into a calibrated posterior that a decision can act on.
Example¶
A public-health team is estimating how severe a new respiratory outbreak is — specifically, whether the hospitalization rate is high enough to trigger surge protocols. They open a worksheet. The prior, from the last comparable season, sits at a moderate estimate. Then each stream enters a row: syndromic surveillance nudges the estimate up but is discounted for reporting lag; a sentinel-hospital cohort pushes it up more and carries the most weight; a household serosurvey pulls it down, suggesting many mild cases the hospital data never saw. Two of the dashboards, the worksheet notes, both draw on the same state reporting pipeline — so their agreement is entered once, not twice, with an explicit dependency discount.
The bottom row is not a point estimate but a bounded claim: "moderate-to-high severity, driven mainly by the sentinel cohort, with the serosurvey as the largest downward pull and reporting lag as the biggest remaining unknown." A sensitivity line shows that if the sentinel cohort were removed the conclusion would soften to "uncertain" — which tells the team that its confidence is resting on one stream and that one more independent hospital signal would be worth more than ten more dashboards.
How it works¶
- Start from an explicit prior so the update is visibly a change in belief, not a fresh verdict conjured from the evidence pile.
- One row per stream, carrying its own weight. Each stream's pull on confidence is a function of relevance, quality, independence, and uncertainty — never its familiarity or the seniority of who ran it.
- Discount shared lineage before summing. Streams flagged as dependent are combined once with an explicit haircut, so correlated evidence cannot masquerade as independent confirmation.
- Run the leave-one-out check. Recompute the bounded confidence with the strongest stream removed; if the conclusion collapses, the worksheet says so.
- Close only when the stopping condition is met — when another stream would add little relative to its cost and the residual uncertainty is tolerable for the decision.
Tuning parameters¶
- Weighting form — ordinal, qualitative, or probabilistic. Numeric weights are auditable but invite false precision over judgments that are really ordinal.
- Dependency-discount severity — how hard correlated streams are penalized. Too gentle double-counts; too harsh throws away real corroboration.
- Confidence expression — a bounded interval, a categorical band, or a probability. Pick the form the decision-maker can act on.
- Stopping threshold — how small an expected information gain justifies closing the update versus commissioning one more stream.
When it helps, and when it misleads¶
Its strength is that it makes belief-updating inspectable: anyone can see which stream carried the day, where the dependency discounts were taken, and how fragile the conclusion is to its strongest input. Done honestly it is disciplined Bayesian bookkeeping,[n1] and the leave-one-out row is its best defense against a confident answer resting on a single stream.
Its failure mode is the tidy total. A worksheet that resolves to a crisp number invites treating the number as the finding rather than as a summary of contestable weights, and it is easily run backwards — filled in to justify a conclusion already chosen, with weights nudged until the bottom row agrees. It also silently rewards streams that are easy to quantify over ones that are merely important. The guarding discipline is to keep every weight and discount defensible on its own line, carry the sensitivity result forward rather than burying it, and treat the output as a bounded claim whose assumptions must hold — not a settled fact.
How it implements the components¶
confidence_calibration_statement— its final row is the bounded, scope-tied confidence claim, stating what is supported, how strongly, and what remains unresolved.weighting_and_non_dominance_rule— each stream's contribution is set by relevance, quality, and independence, and the leave-one-out check enforces that no single stream is allowed to silently dominate.stopping_and_sufficiency_rule— the worksheet closes on an explicit sufficiency condition rather than trailing off, distinguishing "enough for this decision" from "everything obtainable."
It consumes rather than produces the raw evidence picture: evidence_stream_inventory and per-stream quality live in Evidence Stream Matrix, and the support/repetition classification of sources is Cross-Source Corroboration Table's source_diversity_plan — this worksheet only prices what those tables have already catalogued.
Related¶
- Instantiates: Independent Evidence Triangulation — produces the archetype's terminal output, the bounded confidence statement.
- Consumes: Evidence Stream Matrix for the per-stream inputs and Convergence–Divergence Rubric for how much each agreement should count.
- Sibling mechanisms: Evidence Stream Matrix · Cross-Source Corroboration Table · Convergence–Divergence Rubric · Triangulation Audit Trail
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: The worksheet starts from a prior, weights evidence streams by relevance, quality, independence, and uncertainty, discounts shared lineage, and recomputes sensitivity, producing a bounded updated confidence result.
Nearest alternative: Interface, Display & Cue — The worksheet structures user input, but its defining contribution is the evidence-combination and sensitivity calculation rather than the entry surface.
Review outcome: Adjudicated after independent review; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Bayesian statistics established prior-to-posterior updating through likelihood-weighted evidence with dependence explicitly discounted.
Related originating lineages:
- Security Studies & Intelligence Analysis — Structured analytic techniques supply the legible worksheet for mixed evidence, contradictions, and sensitivity.
Review resolution: Both reviewers agree on statistics_experimental_design as primary. Reading the source mechanism confirms that its defining operation belongs to that lineage; the final record retains security_intelligence only where it materially formed the mechanism and keeps present-day application breadth separate from provenance.
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] Bayesian updating — revising a prior probability in proportion to how much more likely the observed evidence is under one hypothesis than another. The worksheet is a deliberately legible, often qualitative, stand-in for that calculus, with the dependency discount standing for the fact that correlated evidence carries less than its face-value likelihood. ↩