# Time, truth and provenance

*Foundations · lesson 4 of 5 · ~10 min*

## What you will learn

Why the standard treats history as load-bearing: the Semantic Timeline, provenance, fingerprints, and what "reconstruct any past state of knowledge" buys you.

## Knowledge without history is rumor

Most systems store only the current state. Ask them "why is this value what it is?", "who asserted it?", "what did we believe last quarter?" and they shrug. For a single app that is survivable. For federated knowledge it is fatal: you are consuming someone else's assertions, and without history you cannot audit, dispute, or trust them.

Vercy's answer is structural, not procedural: **change itself is an entity.**

## The Semantic Timeline

Every mutation of semantic reality is an **Event**: object created, value asserted, mapping changed, contract signed, model version released. Events are append-only; nothing is ever edited in place. The sequence of events is the **Semantic Timeline**, and it gives every Universe three superpowers:

1. **Time travel**: reconstruct the exact state of knowledge at any past moment: which objects existed, which versions of which rules applied.
2. **Attribution**: every assertion traces to who made it, when, on what basis.
3. **Honest revision**: correcting the past means *appending* a correction event, never rewriting history. The mistake and its correction are both permanently visible.

## Two consequences discovered in the wild

The civic case study (a state modelled as a Dimension) found the Timeline indispensable in two places its designers did not anticipate:

- **Courts.** A just legal system judges past acts by the norms *in force at the time*. That is impossible unless norm versions are reconstructible: the Semantic Timeline applied to law.
- **Anti-gaming measurement.** When you measure value, people optimize the measure. The defense is retrospective revaluation: re-scoring past events as consequences become visible. Revaluation *changes the evidence, never the history*: append-only again.

When your model needs either "what did the rules say then?" or "re-assess the past without falsifying it", you need the Timeline. Most serious models need both.

## Provenance and the queryable graph

Provenance generalizes the timeline across derivations: this record was *summarized from* that source, *harvested by* that pipeline, *validated by* that owner. The standard makes the derivation graph queryable, so "show me everything downstream of this retracted source" is a query, not archaeology.

## Fingerprints: identity for meaning

How do you know two copies of a model *mean the same thing*? Bytes differ (formatting, ordering, serialization); meaning may not. The **Semantic Fingerprint** is a reproducible hash over the *normalized semantic structure* of a model: same meaning, same fingerprint, regardless of serialization. It is how packages self-verify, how registries deduplicate, and how drift between master and copy is detected mechanically. One of the production models uses exactly this trick to keep its public website honest: a daily job compares the fingerprint of the live projection against the model and alerts a human on drift.

## Key takeaways

- Change is an entity: append-only Events on a Semantic Timeline.
- History enables time travel, attribution, and honest revision, and real systems (courts, anti-gaming measurement) cannot function without it.
- Provenance makes derivations queryable; fingerprints give meaning a verifiable identity.

## Go deeper

- [Event and the Semantic Timeline](/spec/#04-core-concepts/Event.md)
- [Traceability](/spec/#02-architecture/Traceability.md) · [Provenance graph](/spec/#02-architecture/Provenance-Graph.md)
- [Versioning and the Semantic Fingerprint](/spec/#02-architecture/Versioning.md)

Next: [The standards family](05-family.md)