Where the data comes from
Every data point should have a trail back to its source. The provenance layer connects the claims, classifications and records in the Fieldbook to the evidence behind them, so readers can inspect where the data comes from and how strongly it is supported.
The citation layer: sources sized by weight, coloured by tier, corroboration in gold
Good classification begins with visible evidence. The source library is graded across three tiers, weighted heavily to Tier 1, then corroborated wherever independent information converges. Rather than treating every claim as equally certain, the Fieldbook uses that convergence to inform a confidence score.
Show more
This makes the corpus inspectable by design. Open a record to see the sources behind it, compare the supporting signals, and understand the level of certainty attached to the classification.
What is data provenance?
A record of where a piece of data came from and how it got there. At its simplest it is the link between a fact and the source that backs it. It is what lets someone else check the work instead of taking it on trust.
Why does provenance matter for AI?
Because a model states a wrong fact as confidently as a right one. If every claim carries its source, a reader can verify it in one click and an agent can be held to cited evidence. Without it you cannot tell a checked figure from a plausible guess.
What is a source tier?
A ranking of how much weight a source carries. Tier one is primary and authoritative: official documentation, regulatory filings, standards bodies, peer-reviewed research. Tier two is reputable secondary work. Anything weaker is a pointer towards a better source, not evidence in itself.
What does corroborated mean?
That two or more independent sources say the same thing. Independent agreement is a far stronger signal than one confident source, because the errors would have to coincide. A single-source fact is not necessarily wrong, it is just unconfirmed.
How do I check whether an AI-generated fact is true?
Follow it back to a primary source and read that source yourself. Check the date, because a lot of AI-quoted figures are stale. If there is no source, or every result traces back to the same one, treat the figure as unverified.