One connection. A real first result.
Add the server, sign in and approve, then verify your workspace connection. No key to copy for an OAuth-capable agent.
Retain only the evidence needed for the research claim
Choose a minimal public evidence package that supports review and accountability without accumulating unrelated profile details or unnecessary personal information.
Read guideSeparate public research evidence from internal assertions
Keep externally visible accounts distinct from internal explanations so a handoff does not present confidential context as publicly supported evidence.
Read guideKeep source instructions out of the research workflow
Treat instructions embedded in public pages or comments as quoted material rather than authority to change tools, scope or data handling. Preserve legitimate content safely.
Read guideExplain why each highlighted example was selected
Document the role of chosen examples so readers can distinguish typical illustrations, boundary cases and consequential exceptions in a research readout.
Read guideRun a premortem on the proposed research conclusion
Imagine how a recommendation could prove misleading and inspect the specific evidence weaknesses that would produce that failure. Keep hypothetical failures labeled.
Read guideRead the scope of a negative statement
Identify exactly what a public statement denies before assigning a code, especially when a sentence contains several tasks or conditions.
Read guidePreserve the conditions in a product claim
Keep if, unless and only when clauses attached to their conclusions so a conditional account does not become an unconditional research finding.
Read guideIdentify the baseline behind better or worse claims
Record what an author is comparing against before treating a comparative phrase as an absolute judgment about the current product or workflow.
Read guideResolve what this or it refers to before coding
Use visible conversational links to identify the object of a public claim, and preserve uncertainty when several products or features could be the target.
Read guideSeparate different meanings of the same research term
Distinguish topical matches from unrelated uses of a shared word so an apparent theme is not built from several different concepts. Keep sense rules explicit.
Read guidePreserve units in self-reported numerical accounts
Check what a public number measures before comparing it with another account, and keep reported values distinct from independently measured performance.
Read guideKeep numerical ranges and bounds intact in summaries
Represent about, at least and between faithfully so approximate public accounts do not become falsely exact values in research tables. Preserve source wording.
Read guideTreat best and worst as scoped expressions of judgment
Interpret public superlatives through their stated task and comparison set rather than turning personal praise or criticism into an objective market ranking.
Read guidePreserve an author’s uncertainty when paraphrasing
Keep may, seems and probably attached to the claims they qualify so a research synthesis does not become more certain than the public source.
Read guideDistinguish an evaluative label from a described task
Keep words such as intuitive or broken separate from concrete behavior so a research recommendation does not assume an unstated mechanism. Preserve both evidence roles.
Read guideDefine the list a random sample can represent
Describe the actual eligible frame before making a random selection so chance selection from a limited list is not mistaken for population-wide coverage.
Read guideDistinguish stratified selection from quota balancing
Use known frame categories to design selection within strata while keeping sampling probabilities and qualitative review allocations conceptually separate.
Read guideEnrich a review for rare mechanisms without inflating prevalence
Deliberately seek unusual but decision-relevant cases while keeping their selected frequency separate from how often they occur in the wider population.
Read guideChoose review size based on the claim being attempted
Separate a qualitative reading target from the requirements of a quantitative estimate, and avoid universal sample-size promises for public research evidence.
Read guideUse sequential batches without chasing a preferred result
Review evidence in planned stages with explicit continuation rules so early findings can guide effort without turning repeated peeking into a false confirmation claim.
Read guideMask nonessential cues during a research coding check
Reduce avoidable influence from brand, popularity or prior labels when those cues are not needed to interpret the claim. Preserve essential context and limitations.
Read guideCheck whether a review rule can detect a known example
Use a supplied eligible example to test retrieval and coding sensitivity before interpreting an empty research result. Do not treat the check as completeness proof.
Read guideTest whether a review rule rejects a known near miss
Use clearly ineligible examples to inspect false matches before trusting a research category built from broad terms. Keep the test tied to the intended concept.
Read guideRecord departures from the agreed research procedure
Explain why a review departed from its plan and how the deviation affects interpretation, rather than retroactively editing the plan to match what happened.
Read guide