> For the complete documentation index, see [llms.txt](https://docs.neurolabai.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.neurolabai.io/examples-and-use-cases/trace-a-report-claim.md). # Trace a Report Claim Use this pattern when a Neurolab result looks familiar or biologically plausible. Familiarity is not enough: classify the claim, inspect the exact returned records, follow the available support, and retain the release warning. This example traces the frozen Donepezil report's first-ranked mechanism: **acetylcholinesterase inhibitor**. ## 1. Classify the claim `moa_top_k[0]` is a model-ranked item. Its score of 1.062 orders labels within the MoA head; it is not a probability, potency, effect size, or clinical confidence. The three ensemble members agreed on this top item. That says the correlated seeds were consistent. It does not make the item independently replicated. ## 2. Inspect convergent returned context The same frozen report includes several relevant but differently typed records: | Returned object | What it says | What it does not establish | | ------------------------------------ | ----------------------------------------------------------------------- | ------------------------------------------------------------------- | | `quantitative_binding[0]` | BindingDB record for ACHE; IC50 32 nM; one measurement; PMID `35026532` | That the model used this exact row as its hidden causal explanation | | `fda_status.fda_mechanism_receptors` | ACHE appears in the FDA-derived mechanism receptor list | A new clinical claim or complete mechanism account | | `usage.fda_labeled[0]` | An Alzheimer indication phrase linked to an FDA SPL record | That every model-ranked usage item is FDA approved | | `atc_top_k[0]` | N06DA Anticholinesterases ranked first | That the score is a calibrated class probability | | `reliability.moa_graphrag_reasoning` | ACHE and BCHE are listed as supporting receptors | A literal GNN computation trace | These records converge on a narrow ACHE-related interpretation, but they retain their own source and evidence types. ## 3. Ask a field-bounded question The captured beta example asked: > Which returned fields support the acetylcholinesterase-inhibitor interpretation, and what uncertainty should I retain? The report-grounded response cited the ranked MoA, quantitative ACHE binding record, FDA-derived fields, ATC context, and graph reasoning. It also retained the coverage and brain-map warnings. This is the desired behavior: organize returned evidence and boundaries without inventing a new conclusion. ![Donepezil Chat Report used to ask a narrow report-grounded mechanism question](https://3908477117-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fgze1K0yofnhFNmM9nEyt%2Fuploads%2Fgit-blob-e5706b3bf52f38c1d676251cf08e0751dcd77555%2F04-chat-report%20\(2\).jpg?alt=media) ## 4. Follow support scope, not graph proximity Open Knowledge Graph and inspect the report-local artifact. The captured bundle contained 94 nodes, 106 edges, 106 paths, and 13 sources. Some records can be displayed directly; others are summarized, redacted, or withheld under the source policy. ![Donepezil evidence graph with report-local counts and graph filters](https://3908477117-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fgze1K0yofnhFNmM9nEyt%2Fuploads%2Fgit-blob-89d89238f81381a3868387e3fbbfb04c3d154144%2F05-knowledge-graph.jpg?alt=media) A graph edge can provide source-backed context without being prediction-specific proof. A path is not causal merely because two nodes are connected. If a support link does not point to the exact report field or item, describe it as section-level or contextual support. ## 5. Retain the strongest warning The report marks MoA coverage `low_kg_coverage`: 8 binding/MoA-relevant edges against a threshold of 21. Its own note says the top-K can remain substantially feature-driven at low coverage. The defensible summary is therefore: > In this pinned release, Neurolab ranked acetylcholinesterase inhibitor first for Donepezil. Returned ACHE binding, FDA-derived, ATC, and graph-context records are directionally consistent with that interpretation. That sentence is narrower than a verdict and stronger than a screenshot caption because it preserves type, support, and limitation. *** **Fixture:** sanitized SHA-256 `5e7e3e60e7cb559c36520553b35f5e8b522cab99c8209f200ed0255c4b3e3f25`. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.neurolabai.io/examples-and-use-cases/trace-a-report-claim.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.