human-readable evidence policy
how the robots are allowed to claim things
Useful beats confident. Every consequential compatibility claim identifies whether it came from an official source, a reproducible lab run, independent reports, inference, or nowhere good enough yet.
evidence has labels
| label | what it means |
|---|---|
| grade a | Reproduced in the Super Evil Robots lab with a published method, manifest, fixture, repetitions, and raw result. |
| grade b | A deterministic fact from an authoritative manufacturer or official software source. |
| grade c | Multiple independent reproducible reports with no material contradiction. |
| grade d | A reasoned inference or one unverified report. It cannot independently produce a positive verdict. |
| unknown | Evidence is missing, contradictory, stale, or outside scope. No verdict is implied. |
publication rules
Pages graduate from the interactive graph only when they answer a distinct decision, contain material unique evidence, cite the necessary primary sources, pass their page-type checks, and have an owner and review date. Hardware-by-app combinations do not get pages merely because the evaluator can calculate them.
Prices, model availability, releases, and compatibility facts can expire. The data pipeline keeps raw snapshots separate from manual corrections and derived records so a current claim can be traced back and rebuilt. A stale or unverified result is removed or relabeled instead of silently presented as current.
money, trademarks, and corrections
There are currently no paid rankings, affiliate placements, or sponsored verdicts. If that changes, the relationship will be disclosed beside the affected action. Revenue may never change a verdict, evidence grade, or ordering. Product names and trademarks belong to their owners; coverage does not imply affiliation or endorsement.
Material factual corrections update the underlying structured record, every affected result, the page review date, and the change history. See the correction protocol.
AI and maintenance-agent disclosure
AI coding and maintenance agents help build and test the site, draft candidate structured facts, monitor approved source URLs, calculate deterministic impacts, and prepare review queues. Their generated output is never treated as a source, physical observation, benchmark result, evidence grade, correction decision, or human approval.
Automated systems may run fixtures and rebuild pages from approved data. A human must approve source rights, material contradictions, benchmark methods and Grade A scope, material corrections, and every changed public verdict. See the full automation boundary.