Evidence-to-decision mapping.
Evidence becomes decision-relevant only when its route is explicit: what was observed, which claim it supports or challenges, which criterion changes, which option is affected and whether that change can alter the authorized action.
Mapping explains how evidence can change a choice.
It is not a bibliography, evidence dump or confidence score. It is a traceable representation of the inferential and evaluative path between evidence and an authorized decision.
Evidence-to-decision mapping is the explicit linkage of sources, observations, claims, uncertainty and applicability to decision criteria, option consequences, thresholds and conclusions. It records where evidence supports, contradicts, qualifies or fails to resolve a decision-relevant proposition.
Do not collapse observation, inference and commitment.
Each layer answers a different question. Skipping one creates an unsupported jump that a reader or retrieval system cannot audit.
Where from?
Origin, method, date, author and custody.
What was found?
Reported or measured state before interpretation.
What follows?
Bounded proposition derived from observations.
Where valid?
Population, context, horizon and assumptions.
What changes?
Decision property affected by the claim.
Which candidate?
Consequence, feasibility or risk revised.
Does it cross?
Threshold, ranking, escalation or action.
Every evidence edge needs a declared logical role.
“Related to” is too weak. The map should state whether the evidence increases support, challenges the claim, narrows its domain, establishes context or leaves a material gap.
| Role | Logical effect | Required statement | Example | Common misuse |
|---|---|---|---|---|
| SUPPORTS | Increases warranted confidence in a bounded claim. | What claim, under which conditions, and by how much? | Verified load test supports recovery under declared volume. | Treating one supporting source as conclusive. |
| CONTRADICTS | Provides incompatible observation or explanation. | Which part conflicts: effect, magnitude, population or mechanism? | Incident history contradicts vendor reliability claim. | Discarding conflict as an outlier without basis. |
| QUALIFIES | Restricts scope, strength or transferability. | Where does the original claim stop applying? | Performance holds only below 100 concurrent users. | Repeating the headline without the condition. |
| CONTEXTUALIZES | Changes interpretation without directly testing the claim. | Which environmental or baseline condition matters? | Market growth changes the significance of current share. | Presenting context as causal proof. |
| CORROBORATES | Provides independent convergence on the same bounded proposition. | How independent are source, method and data? | Separate telemetry and audit log align on event sequence. | Counting duplicated reports as independent evidence. |
| UNKNOWN | Marks a material proposition without adequate support. | What decision could change if resolved? | No verified evidence of migration effort. | Converting absence into zero risk. |
Trace different evidence classes into an option consequence.
Select a scenario. The console shows the minimum mapping record from evidence object to decision implication.
SELECT MAP
One claim may affect several options—and one option may depend on several claims.
The matrix exposes coverage, contradictions and decision-critical blanks without pretending that the number of sources equals evidential strength.
| Decision claim | Criterion | Option A | Option B | Option C | Evidence state | Decision implication |
|---|---|---|---|---|---|---|
| Peak-load recovery stays below 60 minutes. | Continuity | CONTRADICT | SUPPORT | CONDITIONAL | Test + incident record; moderate conflict. | Option A fails unless redesigned. |
| EU residency is available for all required data. | Compliance gate | PASS | PASS | FAIL | Contract schedule verified. | Option C removed before scoring. |
| Migration completes inside 12 weeks. | Time to value | ESTIMATE | OBSERVED | UNKNOWN | Different evidence quality by option. | Preserve uncertainty; no common point estimate. |
| Lifecycle cost remains below approved envelope. | Affordability | SUPPORT | RANGE | SUPPORT | Costs use common 24-month boundary. | Option B sensitive to integration overrun. |
| Exit can occur without material service interruption. | Reversibility | WEAK | STRONG | MIXED | Contract and architecture evidence differ. | Raises evidence burden for Option A. |
Prioritize evidence by its ability to change the action.
A decision hinge is a proposition whose plausible states lead to different choices, thresholds or escalation paths. High uncertainty is not automatically important if resolving it cannot change the decision.
LOW UNCERTAINTY
CONTROLHIGH IMPACT
HIGH UNCERTAINTY
INVESTIGATELOW IMPACT
LOW UNCERTAINTY
RECORDLOW IMPACT
HIGH UNCERTAINTY
DO NOT OVER-RESEARCH
HINGE
PRIORITY
Evidence can be sufficient for action without being complete.
Sufficiency is decision-relative. It depends on stakes, reversibility, cost of delay, available alternatives and the consequence of error—not on achieving universal certainty.
| Decision condition | Evidence requirement | Reason | Acceptable action | Required control |
|---|---|---|---|---|
| Low exposure + highly reversible | Enough to justify a bounded test. | Errors can be detected and corrected cheaply. | Pilot or staged experiment. | Stop condition and observation plan. |
| High exposure + irreversible | Stronger direct evidence and explicit residual uncertainty. | Error cannot be easily repaired. | Commit only after hard gates and escalation. | Independent review and sensitivity. |
| High time-to-harm | Rapid evidence sufficient to distinguish safer immediate actions. | Waiting may dominate evidential benefit. | Precautionary reversible containment. | Continuous update and narrow scope. |
| High value of delay | Evidence expected to materially change the preferred option. | Inquiry is worth more than immediate commitment. | Pause or buy information. | Deadline and decision-hinge question. |
| Evidence unlikely to resolve | Transparent range and robust option test. | Further research may not reduce uncertainty. | Choose action resilient across plausible states. | Monitoring and reopening trigger. |
Conflicting evidence is a map problem before it is a voting problem.
Do not count sources for each side. Diagnose why the findings differ and whether the conflict concerns data, method, population, time, mechanism or interpretation.
Same claim?
Verify that sources address the same proposition, option, population and outcome.
Same observation?
Compare sampling, instrument, baseline, missingness and analytical method.
Same domain?
Inspect geography, segment, system state, volume and operating conditions.
Same validity window?
Determine whether one source reflects an earlier state or evidence decay.
Residual conflict?
Preserve unresolved disagreement and test decision sensitivity to each account.
Evidence changes decisions through different mechanisms.
Select a domain. Each example identifies the claim, criterion, option effect and action rule—not just the source.
Traceability must survive every transformation.
A decision record should preserve not only the final source URL but the sequence of collection, extraction, interpretation, synthesis and approval.
Source identity
Creator, document, system or observation point.
Collection
Method, timestamp, version and custody.
Atomic finding
Exact passage, value or observed event.
Bounded claim
Inference, assumptions and applicability.
Evidence state
Support, conflict, gaps and certainty.
Decision record
Judgment, owner, action and review trigger.
RAG must retrieve the evidence edge—not an orphaned passage.
Store claim direction, applicability, contradiction, criterion link and decision implication beside provenance. Retrieval without qualifiers can turn a true local finding into a false general answer.
Carry the inference boundary with the claim.
This schema improves traceability and conflict retrieval. It does not validate the source or guarantee correct model output.
{
"decision_id": "DEC-06-001",
"evidence_id": "EV-017",
"source": {"uri": "record URI", "observed_at": "YYYY-MM-DD"},
"observation": "recovery completed in 47 minutes",
"claim_id": "CLM-REC-03",
"claim": "priority service recovers below 60 minutes",
"role": "supports",
"applicability": {"population": "EU production", "condition": "peak load"},
"certainty": "moderate",
"criterion_id": "CRIT-CONT-01",
"option_effect": {"option_id": "OPT-B", "status": "passes"},
"decision_hinge": true,
"conflicts_with": ["EV-009"],
"action_if_invalidated": "pause and redesign"
}Evidence-to-decision mapping, clarified.
Operational answers to the most common traceability errors.
Does stronger evidence automatically determine the decision?
No. Evidence changes warranted beliefs about consequences and conditions. The decision also depends on objectives, values, constraints, distributional effects, authority, uncertainty, reversibility and cost of error.
What is a decision-relevant claim?
A bounded proposition whose credible states can change option feasibility, ordering, timing, escalation or commitment. A fact can be true and still be decision-irrelevant if it cannot affect the authorized choice.
How is evidence mapping different from citation?
A citation identifies a source. Mapping declares the logical role of a source or finding, the claim it affects, its applicability, uncertainty, conflicts and the resulting consequence for a criterion or option.
Can one source support several claims?
Yes, but each claim edge should be recorded separately. The same passage may support one claim, qualify another and provide only context for a third. Reusing a source does not create independent corroboration.
What should happen when evidence conflicts?
Verify claim identity, compare methods, populations, contexts and dates, assess independence and preserve residual conflict. Then test whether the decision changes under each credible account.
When is evidence sufficient?
When residual uncertainty is acceptable relative to decision stakes, reversibility, timing and cost of error, and when further inquiry is unlikely to change the action enough to justify delay. Sufficiency is not certainty.
What is a decision hinge?
A claim or uncertain condition whose plausible states lead to different actions or materially different option status. Decision hinges guide proportionate evidence collection.
How should missing evidence be represented?
As an explicit unknown tied to the claim, option and potential decision effect. Do not encode missing evidence as zero impact, neutral evidence or assumed support.
Continue through the complete Decision Intelligence system.
Evidence mapping connects criteria and option consequences before trade-offs, uncertainty, commitment and review.
Decision Intelligence
Structure choices through objectives, alternatives, evidence, uncertainty, trade-offs, commitment and review.
Evidence does not jump to action. The map must show every bridge.
Preserve the observation. Bound the claim. Test applicability. Declare uncertainty. Connect the criterion. Identify the option consequence. Expose the value judgment. Then record who authorized what, under which conditions and with which review trigger.