On this page
- RNX-1040.T001 · Access versus retrieval distinction
- RNX-1040.T002 · Citation versus mention distinction
- RNX-1040.T003 · Recommendation classification
- RNX-1040.T004 · Selection versus transaction distinction
- RNX-1040.T005 · Observation provenance
- RNX-1040.T006 · Question cohort representativeness
- RNX-1040.T101 · Citation destination resolution
- RNX-1040.T102 · Citation claim support
- RNX-1040.T103 · Grounded versus ungrounded experiment separation
- RNX-1040.T104 · Repeated-session sampling control
- RNX-1040.T105 · Entity-match adjudication
- RNX-1040.T106 · Observation temporal alignment
- RNX-1040.T107 · Independent annotation agreement
- RNX-1040.T108 · Observation evidence retention
RNX code library · 14 controls
RNX-1040Observed AI Discovery Events
Read what these checks mean, why they matter and which evidence they concern. Each ID below keeps its canonical place in the RNX hierarchy.
These are free learning summaries, not full technical tests or an industry certification. Specialized checks apply only to the relevant feature. Experimental labels are preserved from the canonical source.
RNX-1040.T001
Access versus retrieval distinction
Experimental — not an established ranking requirement
Call a successful crawl evidence of access, not proof that an AI answer used the page.
Why it matters. Reading and selecting information are separate steps.
Illustration — fictional. A successful page response is called proof that an AI answer used the page.
When it applies. Inspected target: retrieval inference. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.
RNX-1040.T002
Citation versus mention distinction
Experimental — not an established ranking requirement
Separate a brand mention from an explicit source citation.
Why it matters. A named business may appear without any link to its website.
Illustration — fictional. Plain brand mention counts as citation.
When it applies. Inspected target: citation annotation. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.
RNX-1040.T003
Recommendation classification
Experimental — not an established ranking requirement
Count a recommendation only when the answer expresses preference or suitability in context.
Why it matters. A list or passing mention is not automatically an endorsement.
Illustration — fictional. Neutral list counts as endorsement.
When it applies. Inspected target: recommendation annotation. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.
RNX-1040.T004
Selection versus transaction distinction
Experimental — not an established ranking requirement
Separate a suggested choice, an attempted action and a completed purchase.
Why it matters. Talking about buying something is not a transaction.
Illustration — fictional. Suggested purchase counts as sale.
When it applies. Inspected target: action outcome. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.
RNX-1040.T005
Observation provenance
Experimental — not an established ranking requirement
Record which interface produced an AI observation.
Why it matters. An API test should not be presented as a consumer-app result.
Illustration — fictional. A developer API experiment is described as an observation in a consumer search app.
When it applies. Inspected target: observation record. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.
RNX-1040.T006
Question cohort representativeness
Experimental — not an established ranking requirement
Explain how test questions were chosen and what they leave out.
Why it matters. A few favorable prompts do not represent all customer searches.
Illustration — fictional. Tests using only the business's name are used to claim success for general customer searches.
When it applies. Inspected target: question cohort. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.
RNX-1040.T101
Citation destination resolution
Experimental — not an established ranking requirement
Follow an observed AI citation to the actual destination.
Why it matters. A displayed link may redirect or fail to reach the claimed source.
Illustration — fictional. Tracking link ends on competitor page but counts as own citation.
When it applies. Inspected target: citation destination. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.
RNX-1040.T102
Citation claim support
Experimental — not an established ranking requirement
Check that a cited source supports the particular answer claim.
Why it matters. A citation's presence alone does not establish a correct answer.
Illustration — fictional. Citation exists but linked page says opposite stock status.
When it applies. Inspected target: cited claim. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.
RNX-1040.T103
Grounded versus ungrounded experiment separation
Experimental — not an established ranking requirement
Separate tests with retrieval enabled from those without it.
Why it matters. These settings answer different questions about information access.
Illustration — fictional. Answers generated without retrieval are mixed with consumer web-search observations.
When it applies. Inspected target: observation configuration. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.
RNX-1040.T104
Repeated-session sampling control
Experimental — not an established ranking requirement
Control and record repeated-session conditions in AI tests.
Why it matters. Conversation history can affect results and make runs less comparable.
Illustration — fictional. Only best of five generations kept.
When it applies. Inspected target: repeat-run block. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.
RNX-1040.T105
Entity-match adjudication
Experimental — not an established ranking requirement
Match a mentioned entity to the right business using evidence.
Why it matters. A same-name company should not earn your business a recorded mention.
Illustration — fictional. Foreign same-name firm counts as client recommendation.
When it applies. Inspected target: entity mention. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.
RNX-1040.T106
Observation temporal alignment
Experimental — not an established ranking requirement
Compare observations within meaningful time windows.
Why it matters. An answer from before an update cannot establish whether the new content helped.
Illustration — fictional. Model rollout coincides with edit but effect attributed solely to edit.
When it applies. Inspected target: comparison window. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.
RNX-1040.T107
Independent annotation agreement
Experimental — not an established ranking requirement
Check whether independent reviewers agree about sampled answers.
Why it matters. Disagreements reveal where a measurement rule needs clarification.
Illustration — fictional. One reviewer changes borderline labels after viewing desired result.
When it applies. Inspected target: adjudicated sample. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.
RNX-1040.T108
Observation evidence retention
Experimental — not an established ranking requirement
Retain the answer and context needed to review an observation.
Why it matters. A summary alone may hide the wording that supported the finding.
Illustration — fictional. Only final percentage retained without underlying answers.
When it applies. Inspected target: retained answer. Use this check only when that object or feature exists in the review scope; provider-specific checks require that provider's integration. Missing evidence remains unknown.