SpyderBot Documentation
Observations, Evidence, and Provenance
This page explains how SpyderBot talks about structured observations, the evidence you can inspect, and the provenance that explains where a result came from—without exposing proprietary scoring, sampling, classification, or orchestration methods.
Observation
An observation is a structured record derived from an eligible source. Its meaning depends on the data class:
- an AI-response observation may describe a mention, recommendation, citation, sentiment, entity, or position;
- a Machine Traffic observation may describe a request, URL, status, actor, or policy outcome;
- a human referral observation may describe a session, event, goal, order, or revenue outcome;
- an audit observation may describe the tested condition of a URL or entity at a particular time.
Mixing data classes without saying so produces invalid conclusions. A bot request cannot prove a citation. A citation cannot prove a visit. A referral session cannot identify a prompt without direct deterministic evidence.
Evidence
Evidence is the source material and metadata that allow a user to inspect or audit a result. Depending on the product, evidence may include:
- prompt and raw response;
- extracted observation;
- citation and source;
- request or event record;
- session or attribution record;
- audit finding;
- validation state;
- method or taxonomy version;
- time, source, and scope metadata.
Summary visibility and evidence access may have different workspace, plan, sharing, retention, or source requirements. Access to a summary does not always grant access to underlying evidence.
Provenance
Provenance explains where a result came from. Useful provenance may include:
- source product;
- run, report, observatory, or audit identifier;
- prompt, URL, brand, entity, or website;
- AI system, search surface, or connected source;
- event or response time;
- reporting period and timezone;
- eligibility or validation state;
- method version;
- observed, calculated, estimated, verified, classified, inferred, unavailable, or insufficient state.
Evidence supports an interpretation. It does not reveal a provider's private reasoning or establish causality by itself.