Practical guide
Evidence navigation: find the source that answers the claim
Practice source selection, authority and conflict handling with a fictional supplier-renewal evidence map.
Evidence navigation is the work of finding, selecting and connecting information to a specific question. It is not simply collecting links or opening many documents. This original supplier-renewal exercise shows how to build a claim-level evidence map, distinguish source authority from presentation quality and hand unresolved conflicts to the next person without disguising them as facts.
Begin with questions the decision needs answered
Search from the claim outward rather than reading every available file in the order it appears.
A fictional renewal pack contains a sales presentation, a signed service schedule, a recent incident log and an account email. The decision owner wants to know whether overnight support is included. Write that question before searching. A general summary of the supplier may be accurate yet fail to answer the one condition that changes the renewal decision.
Create a small map with the question, candidate source, relevant passage and remaining uncertainty. Keep references precise enough that another reader can open the same passage. The map is useful because it exposes missing support; it should not become a decorative inventory of everything in the folder.
Assign authority to a claim, not a whole document
A source can be useful for one question and inadequate for another.
The sales presentation in this exercise advertises continuous support, while the signed schedule describes business-hour coverage. The presentation helps explain what was marketed; it does not settle what this account purchased. The incident log can show recorded response timings, but it cannot by itself establish a contractual entitlement.
Check dates, account tier, document version and any referenced amendments. If an email mentions an upgraded package without the amendment itself, preserve that gap. Ask the responsible owner to confirm the applicable terms. Do not resolve the disagreement merely by choosing whichever document looks newest or most professionally designed.
Use AI to locate evidence, then inspect it
Treat an extracted passage as a route back to a source, not a substitute for reading its context.
Ask AI to identify passages about support hours, exceptions and response commitments. Require file names and locations where the available tool supports them. Then inspect the passages yourself, including nearby definitions. A sentence about emergency escalation might refer to a different service or exclude the account under review.
Save the passage needed for the decision, not an unexplained output from the assistant. Record when a document was unavailable or difficult to search. If the system returns an unsupported quotation, remove it from the evidence map and locate an actual passage before continuing.
Preserve conflicts instead of averaging them
Conflicting statements often describe different scopes; combining them into a compromise can invent a third claim.
In the fictional pack, do not convert business-hour coverage and continuous-support advertising into an assertion of partial overnight coverage. Write both statements, their sources and the exact unresolved question. The next action might be to request an amendment or ask the account owner to explain which package is active.
Test the map by inserting an older service schedule with more generous terms. A reviewer should be able to see why it does not automatically supersede the signed current schedule. The exercise checks version handling and evidence selection, not whether a participant can memorize a preferred document hierarchy.
Deliver an evidence map someone can challenge
The final map should make support and uncertainty inspectable without requiring the reader to repeat the whole search.
For each material claim, include the wording used in the recommendation, the source location, a short explanation of relevance and any limitation. Group unresolved questions separately from supported findings. This helps the decision owner see whether the renewal can proceed, needs a condition or requires further information.
Ask a second reader to trace the overnight-support statement back to the supplied pack. If they cannot, repair the reference before polishing the memo. Interpret the outcome within the exercise: an accurate map supports a statement about this source-navigation task, not a conclusion that the worker will always find every relevant document.
Sources and scope
NIST addresses AI risk management. Skills England describes workplace AI foundations.
These sources provide background, not endorsement of this exercise. The worked example and suggested review method are original illustrative guidance. They are not customer results, validated benchmarks or evidence of a particular product capability. Adapt the exercise to the task and use qualified review where consequences require it.
Sources: [1] [2]
Sources
- 1.AI RMF Core · NIST
- 2.AI foundation skills for work benchmark · Skills England
Explore this topic
- How to develop and assess contradiction detection in AI-enabled work
Contradiction detection is the ability to recognize when two claims cannot both be relied on in the same way, then determine whether the difference comes from error, timing, scope, definition, authority, or legitimate uncertainty. In AI-enabled work, the skill prevents a model from smoothing incompatible evidence into one confident narrative.
- How to develop and assess data extraction in AI-enabled work
Data extraction is the ability to turn information in documents, messages, tables, or images into a structured record without losing its meaning, provenance, or uncertainty. In AI-enabled work, the difficult part is not producing rows quickly. It is deciding what each field means, locating the right evidence, preserving context, and proving that the resulting record matches the source.
- How to develop and assess document comparison in AI-enabled work
Document comparison is the ability to identify meaningful differences between two or more records, explain why they matter, and preserve enough version and source context for another person to verify the result. In AI-enabled work, this requires more than asking a model to summarize changes. The worker must establish which versions are being compared, separate substantive changes from formatting.
- How to develop and assess evidence sufficiency in AI-enabled work
Evidence sufficiency is the ability to decide whether the available information is enough to support a particular claim or action at the required level of confidence. It is not a fixed quantity of documents. Sufficiency depends on the decision, the importance of the claim, the quality and independence of the evidence, the consequences of error, and the uncertainty that remains.
- How to develop and assess provenance tracking in AI-enabled work
Provenance tracking is the ability to preserve where information came from, what happened to it, who or what performed each transformation, and which version supports the current output. In AI-enabled work, provenance makes a claim or artifact inspectable after text has been summarized, combined, calculated, translated, or regenerated.
- How to develop and assess relevance judgment in AI-enabled work
Relevance judgment is the ability to decide whether information actually bears on the question being answered. It is not the same as finding information, trusting its publisher, or collecting more evidence. In AI-enabled work, the skill is visible when a worker states the decision, tests each item against it, and excludes attractive material that does not resolve the reader's real uncertainty.
- How to develop and assess source authority in AI-enabled work
Source authority is the ability to determine what standing a source has for the particular statement being made. It asks who produced the information, under what mandate or competence, through which process, for what purpose, and in which version. It is not a shortcut that turns an official logo into truth. AI tools often flatten these distinctions. A search result, regulatory text, company help.
- How to develop and assess source discovery in AI-enabled work
Source discovery is the ability to locate the evidence a task actually requires, across the places where that evidence may exist. It is more than entering a broad query and accepting the first plausible results. In AI-enabled work, strong discovery is visible when a worker translates a decision into search targets, uses several retrieval routes, follows evidence lineages, records gaps, and knows.