goguides.com trust favicon wikipedia.org trust favicon mozilla.org trust favicon
GoGuides publishes independent web evidence, provenance, and machine-readable records.
Log In Navigation
The web evidence problem

AI does not need another truth oracle. It needs better evidence.

The open web is easy to publish to and difficult to evaluate consistently. Pages change, identities drift, copied material spreads, source context disappears, and confident language can look stronger than the evidence behind it. Generative systems make that evidence problem more visible, but they did not create it.

The hard problem is not simply “true or false”

A machine evaluating a public source may need to answer several different questions: What URL was actually observed? When was it observed? What changed? Who controls the domain? Is the current page consistent with earlier observations? Does one stored source address a specific claim? Are two exact pages materially similar under the available evidence?

Those questions require different evidence. Collapsing them into one universal “trusted” or “untrusted” label hides important uncertainty.

Seven failure modes better evidence can help expose

1. Missing provenance

A claim can circulate without a stable record of where it came from, what scope it describes, or when it was observed.

2. Identity ambiguity

A site can make claims about ownership, authority, or affiliation that are difficult to separate from independently observed evidence.

3. Silent content drift

A page can change after it was first retrieved. Without historical observations, a consumer may not know that the evidence moved.

4. Synthetic scale

Large volumes of similar pages can create the appearance of independent support even when the available evidence shows strong content relationships.

5. Citation mismatch

A cited page may exist but still fail to support the particular claim attributed to it.

6. Observation gaps

A system may have too little evidence to answer confidently. Treating missing evidence as positive evidence creates false certainty.

7. Interpretation drift

A derived score or model judgment can be mistaken for raw observation unless the evidence classes stay explicit.

What GoGuides contributes

Capability Evidence contribution Limit
Evaluate A GoGuides assessment tied to a defined normalized scope. Not a universal certificate that the source is factually correct.
Web State V3 Stored objective web-state evidence for supported dimensions. Not a guarantee that the live site still matches the stored observation at every moment.
Change Intelligence Deterministic comparison of concrete stored observations. Does not invent publisher motive or semantic importance.
Verification Evidence that an accepted method demonstrated control or authorization. Does not make every statement on the verified site true.
Claim Support Bounded analysis of whether one fixed stored source addresses a claim. Not a universal fact-checking oracle.
Content Relationships Conservative comparison of two exact URLs under stored evidence. Does not prove common ownership, coordination, deception, or intent.

What this does not mean

GoGuides does not “block hallucinations”

An outside model controls its own generation, retrieval, citation, and safety behavior. Better provenance and source evidence can reduce uncertainty; they cannot force another system to produce a correct answer.

Verification is not truth certification

Proof of domain control answers an identity question. It does not prove every factual claim published by the domain.

Change is not automatically tampering

Pages change for legitimate reasons. Deterministic change evidence tells a consumer what changed under the supported comparison contract, not why.

Similarity is not conspiracy

Content relationships can be evidence of similarity without supporting a conclusion about intent, ownership, or coordinated behavior.

The useful architecture is Evidence → Interpretation → Decision

GoGuides' role is to make evidence and GoGuides-derived interpretation inspectable. The final source-use decision remains with the consumer. That separation is what makes the evidence useful to systems that may have different ranking, citation, retrieval, safety, or eligibility policies.