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.
Pilot evidence feasibility before promising a full study
Check whether available material can answer the intended question before committing to a larger review, without presenting a small pilot as the final finding.
Read guideCheck who entered or left a longitudinal research set
Separate changes in a retained account population from changes in what that population discusses before interpreting a longitudinal pattern. Preserve membership history.
Read guideSeparate entry-cohort differences from changes over time
Compare accounts or cases grouped by an explicitly documented start event without confusing different entry conditions with a shared temporal change.
Read guideInterpret a change within one public account’s history
Compare an account’s explicit earlier and later statements without assuming a stable identity, constant conditions or a population-wide change. Preserve the narrow case scope.
Read guideSeparate a learning process from a persistent product obstacle
Review whether an author reports improvement with practice while avoiding assumptions that all difficulties disappear or that every repeated issue is a defect.
Read guidePreserve uncertainty in retrospective experience accounts
Distinguish immediate descriptions from memories of earlier events, and keep recalled timing and sequence appropriately qualified without dismissing useful public evidence.
Read guideConsider public presentation without diagnosing motives
Recognize that public accounts are authored for an audience while avoiding unsupported claims that a particular person exaggerated, concealed or performed an experience.
Read guideKeep published experiences separate from all experiences
Recognize that only experiences turned into visible public material enter the corpus, especially when unusual outcomes are more likely to become stories.
Read guideAvoid giving detailed accounts automatic decision priority
Distinguish interpretive richness from frequency and importance so well-documented cases do not eclipse less detailed but relevant evidence.
Read guideCheck whether a code means the same across contexts
Test category meaning across languages, formats or periods before treating coded differences as differences in the underlying phenomenon.
Read guideConsider the costs of false positives and false negatives
Match a decision’s evidence gate to the consequences of acting on a mistaken finding or missing a real issue, without inventing financial or safety estimates.
Read guideDo not infer an uninterrupted process across video cuts
Treat edits as limits on observable continuity when interpreting a demonstration’s sequence, duration or outcome. Keep shown steps distinct from unseen transitions.
Read guideAttribute transcript claims only when speaker identity is clear
Keep narrator, guest, quoted speaker and off-screen voice separate when reviewing video statements, preserving ambiguity rather than assigning every line to the creator.
Read guideDistinguish captions, interface text and creator annotations
Identify the role of visible text before treating it as a product label, spoken claim or independently observed result in a video or image. Preserve source context.
Read guideRequire event context before using an image as incident evidence
Treat an image’s visible contents separately from claims about when, where or why the pictured event occurred. Preserve unsupported context as unknown.
Read guideBound interface findings drawn from a single screenshot
Describe the state visible in an interface image without inventing prior inputs, hidden settings or the actions needed to reach it. Preserve workflow uncertainty.
Read guideKeep uncertain technical words in spoken evidence unresolved
Avoid silently substituting a plausible product term when audio or transcript wording is unclear and the choice would change the research claim. Preserve alternatives.
Read guideSeparate a demonstration from a controlled performance test
Review what a public demonstration actually controls before interpreting its outcome as a repeatable benchmark or a comparison of product capability.
Read guideUse timestamps at the precision the evidence supports
Treat a returned video moment as a navigation aid with an explicit evidence basis rather than inventing frame-level timing or exact process duration.
Read guideSeparate recommended steps from demonstrated actions
Distinguish what a tutorial tells viewers to do from what its author visibly or explicitly reports doing, especially when evaluating a workflow’s practical sequence.
Read guideAvoid identifying a product solely from visual resemblance
Use explicit labels and source context to establish the object in an image or video before assigning a model-specific research claim. Preserve unresolved identification.
Read guideChoose between a frozen corpus and an updating view
Decide whether a research comparison needs a stable evidence set or current coverage, and make the consequences explicit rather than mixing the two modes.
Read guideKeep rankings conditional on the evidence actually available
Describe a ranked research output as an ordering within a defined visible set, not a complete leaderboard across unavailable accounts or sources.
Read guideRecord disclosed commercial context without dismissing evidence
Keep explicit sponsorship or affiliate disclosures separate from judgments about claim accuracy, and avoid guessing undisclosed relationships from tone or popularity.
Read guide