Research can continue indefinitely when primary sources remain unavailable. An evidence sufficiency rubric defines when to stop searching, what conclusion the available material permits, and what would justify reopening the question. It protects an editor from both premature certainty and endless collection of secondary summaries that never resolve the missing evidence.
Primary sources checked on 8 October 2026. Worked examples and sample records below are illustrative; they do not represent product tests or real customer findings.
Define the decision and the required evidence
Write the claim and the decision it would influence. A low-impact background note may need a different threshold from a public allegation or a production integration decision. Identify the primary record that would directly answer the question: a release note, license, model card, original dataset, or official feature document. This keeps the search focused on evidence rather than brand mentions.
Separate facts necessary for the decision from interesting context. If the unresolved issue is whether a feature is available on a particular plan, more articles about the model’s quality will not close it. List the required evidence fields and state which are essential. A stopping rule should be designed before the search becomes emotionally tied to a preferred answer.
Plan a bounded source search
Choose official domains, document types, known product names, former names, and a date window. Include direct navigation through the provider’s documentation and support pages, not just search-engine results. Perplexity’s official Pro Search description presents a discovery and synthesis workflow; discovered sources still need direct inspection before they support a consequential claim.
Set a practical time or query budget and a coverage checklist. For example, inspect the official product page, help center, release notes, and linked technical document. Record failed access attempts and unavailable pages. Repeating the same query with slightly different wording is not necessarily new coverage. The log should show what was genuinely checked.
Grade evidence without manufacturing confidence
Use categories such as directly supported, supported with qualifications, primary record unavailable, conflicting primary records, and secondary-only reporting. Give each category a permitted conclusion. A secondary-only result might support an attributed report, while an unavailable license should leave a redistribution decision unresolved. Avoid assigning a numerical confidence percentage that disguises the reason for uncertainty.
Consider independence. Several articles quoting one missing press release do not provide several independent confirmations. Record the citation chain and identify common sources. If all reports lead to one unavailable statement, say so. The issue is not the number of links but whether the evidence answers the specific question with the required authority and scope.
Specify the stopping conditions
Stop when the required primary evidence is found and inspected, when the planned coverage is complete without resolving the gap, or when a concrete access restriction prevents further authorized work. A stop is a research state, not a verdict that the claim is false. Write the conclusion allowed by that state and the missing item that would change it.
In an illustrative API-support inquiry, the researcher checks current docs, release notes, and the official endpoint reference but finds no statement establishing the requested feature. The defensible output is not confirmed in the inspected primary sources. It is stronger than a vague maybe and more accurate than declaring the feature impossible.
Escalate a question rather than a vague uncertainty
Prepare one targeted question for the source owner: which current document confirms the feature for this model and plan? Include what was checked, relevant dates, and the exact missing condition. If contacting someone is not authorized, retain the question as a next action instead of sending it automatically. The research artifact can still be complete with an explicit unresolved item.
Avoid asking a provider to endorse your broad conclusion when a narrow documentation question will do. A response can clarify access without establishing performance or reliability. Preserve the response’s author, date, and scope if it later arrives. Treat private guidance according to the organization’s disclosure rules rather than assuming it is suitable for public quotation.
Create an evidence sufficiency rubric
The final rubric should contain claim type, minimum evidence, permitted wording, stopping condition, owner, and reopening trigger. Add examples of wording that overstates the evidence and a corrected version. This helps editors apply the rule consistently instead of interpreting stop as abandon or unresolved as false.
Reopen the investigation when a named primary source becomes accessible, the provider publishes a relevant update, or the decision context changes. Do not repeat the same unsuccessful search on every review without a trigger. A good stopping rule preserves the work already done and directs the next effort toward evidence that could actually change the conclusion.
Review and preserve the finished evidence sufficiency rubric
Before handing over the result, perform a review that starts from the evidence rather than the generated prose. Pick a clearly supported item, a disputed item, and an item with missing information. Write the expected treatment of each before reading the proposed conclusions. Then open the cited originals and compare the result with those expectations. Check the surrounding text, headings, dates, and qualifications, because a citation can point to the correct document while supporting a narrower statement.
Ask a colleague to use the finished deliverable without your explanation. They should be able to locate the original records, identify unresolved questions, and understand which statements are direct evidence and which are your interpretation. Revise any place where they must guess. Keep the reviewed output, source manifest, prompt, and correction history together. Record the review date and the event that should trigger another check. When a source changes, preserve the earlier assessment and explain which conclusions were affected rather than silently replacing the evidence.
Do not equate an empty result with the absence of a problem. Check that all intended inputs were included and readable. Explain exclusions and merged records. A useful handover makes incompleteness visible and assigns a next action to the person who can resolve it.
Frequently asked questions
Does stopping mean the claim is false?
No. It means the planned research reached a defined boundary; state what remains unestablished.
Can many secondary sources close a primary-source gap?
Only if the decision permits that evidence type. Repeated citations to one unavailable source do not create independence.
Should every claim use the same threshold?
No. Match the required evidence to the claim and consequences while keeping the rule explicit.
What should trigger another search?
New primary evidence, resolved access, a relevant product update, or a changed decision requirement.
What is a complete unresolved result?
A documented search boundary, permitted conclusion, exact missing evidence, and an owner or trigger for the next action.
