Evidence Ladder Labeling¶
Classification prompt — instantiates Knowledge-Warrant Audit
A label set that sorts claims from direct observation and replicated evidence down through inference, testimony, analogy, and unsupported assumption.
Evidence Ladder Labeling is an ordered rubric: a ranked ladder of warrant types, from the strongest rung (direct observation, replicated measurement) down through inference, testimony, and analogy to the weakest (bare, unsupported assumption). You classify each claim by finding the rung its warrant sits on, and because the rungs are ranked, the label is a strength rating — a claim tagged "analogy" is by construction weaker-warranted than one tagged "replicated evidence." Its defining move is collapsing what kind of warrant and how strong it is into a single ordinal position. It is not a communication convention for readers and not an aggregating table; it is the analyst's rubric for deciding, per claim, where on the strength ladder a warrant belongs.
Example¶
A detective reviewing an open case lays each supporting claim against an evidence ladder before writing the charging recommendation. The claims sort themselves onto rungs. "The suspect's DNA was on the weapon" lands on the top rung — direct forensic measurement, replicable. "A neighbor saw someone of the suspect's build leave at 9pm" drops to eyewitness testimony, a middle rung known to be fallible. "The suspect had a motive" is inference, lower still. "The method matches a burglary across town last year, so it's probably the same person" is flagged analogy — the weakest rung short of pure assumption. And "the suspect seems the type" is bumped to the bottom: unsupported assumption, no warrant at all.
The value is not in re-investigating anything but in seeing the shape of the support: the case looks strong in prose, but the ladder shows one top-rung claim propping up a stack of middle- and low-rung ones. That ordering tells the detective exactly where a defense will push and which rung most needs to be raised — corroborating the testimony, or finding a second top-rung fact — before the case is sound.
How it works¶
- Publish the ranked rungs. Fix an ordered ladder of warrant types — direct observation and replicated evidence at the top, then measurement, expert testimony, inference, analogy, and unsupported assumption at the bottom — with a one-line test for each rung.
- Place each claim on a rung. For every claim, ask which rung's description its warrant actually matches, forcing the honest question rather than the flattering one ("is this really measured, or inferred and dressed up?").
- Read the rung as the strength. Because the ladder is ordered, the rung a claim lands on is simultaneously its warrant type and its warrant strength — no separate scoring step is needed.
- Surface the low rungs. The output is the distribution of claims across rungs, spotlighting which load-bearing claims sit near the bottom.
The distinguishing discipline is ordinality: unlike a flat type-tag, every label carries a "stronger-than / weaker-than" relation, so classifying is rating.
Tuning parameters¶
- Number of rungs — how finely the ladder is cut. More rungs draw sharper distinctions but invite endless argument over adjacent rungs; fewer rungs are decisive but coarse.
- Rung definitions — how strict the test for each rung is. Strict definitions demote optimistic self-classification but can feel punitive; loose ones let weak warrant climb too high.
- Domain calibration — whether the ladder's ordering is generic or tuned to a field (in some domains a single measurement outranks expert consensus; in others the reverse). A tuned ladder is more accurate but non-portable.
- Tie-handling — what to do with a claim resting on two warrant types at once. Scoring it at its weakest rung is conservative; at its strongest is generous and riskier.
When it helps, and when it misleads¶
Its strength is compressing a messy question ("how well-supported is this?") into a single, comparable position on a shared scale, so claims can be ranked against each other at a glance. It formalizes the intuition behind a hierarchy of evidence — the idea, sharpened in evidence-based practice, that not all support is equal and that warrant type predicts reliability.[n1]
It misleads when the ladder's ordering is treated as universal truth rather than a domain heuristic — a lower rung is not always weaker (a single decisive observation can beat a shaky replication), and a rigid ladder can demote a genuinely strong analogy or over-credit a badly-run measurement. It also flattens the content of a warrant into its category, so two claims on the same rung look equal when one is far better executed. The guarding discipline is to treat the rung as a prior, not a verdict — let a well-argued case move a claim off its default rung — and to pair the ladder with attention to how well each individual warrant was actually produced.
How it implements the components¶
Evidence Ladder Labeling realizes the classify-and-rate components — the audit's per-claim strength-assessment core:
warrant_type_taxonomy— the ladder is the taxonomy of warrant types (observation, measurement, testimony, inference, analogy, assumption), applied per claim.support_strength_rating— because the rungs are ordered, placing a claim on a rung simultaneously rates how strong its support is, without a separate scoring pass.
It does not standardize how these ratings are marked for downstream readers (communication_labeling_standard, uncertainty_and_unknown_marker — that is its nearest twin Epistemic Status Labeling; the difference is that the ladder *ranks warrant strength for the analyst, while that standard marks claim kind for the reader), nor inventory the full belief set into a table (claim_inventory_scope — that is Belief-Warrant Matrix).*
Related¶
- Instantiates: Knowledge-Warrant Audit — the ladder is the audit's warrant-classification and strength-rating step compressed into one ordinal label.
- Sibling mechanisms: Assumption Conversion Prompt · Belief-Warrant Matrix · Claim-Confidence Warrant Review · Epistemic Status Labeling · Source-Independence Cross-Check · Unsupported-Certainty Red Flag · Update-Trigger Checkpoint · Warrant-Decay Review
Editorial Notes¶
Form Classification¶
Form family: Rule, Policy & Commitment
Rationale: The mechanism establishes a standing ordered vocabulary and assignment test that governs how every claim's warrant type and strength must be classified.
Nearest alternative: Interface, Display & Cue — Visible rung labels cue readers, but the operative form is the persistent classification standard that determines which label a claim may carry.
Review outcome: Adjudicated after independent review; high confidence.
Origin Attribution¶
Primary origin: Medicine & Healthcare
Origin pattern: Convergent development
Present-day reach: Multi-domain
Rationale: Evidence-based medicine operationalized explicit hierarchies of evidence that rank warrant types, study designs, and expert opinion for practical decisions.
Related originating lineages:
- Law & Governance — Legal traditions materially developed ranked forms and weights of evidence and testimony.
- Philosophy — Ordering observation, testimony, inference, analogy, and assumption by warrant is fundamentally an epistemological classification. The distinction among observation, testimony, inference, analogy, and assumption also reflects the older epistemological analysis of warrants.
Review resolution: Oxford CEBM documents both the history and intended use of evidence levels. Philosophy and law contribute older epistemic and proof hierarchies, but the modern label set most directly descends from medicine.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Researched adjudication after independent review; high confidence.
Sources consulted:
Notes¶
[n1] The hierarchy of evidence — most developed in evidence-based medicine, where study designs are ranked (e.g. systematic reviews and randomized trials above cohort studies above case reports above expert opinion) — encodes the same claim: warrant type is a strong predictor of reliability, so classifying the type largely tells you the strength. ↩