Ruleset Versioning

Licklider versions computational behavior independently from superficial product presentation.

What versioning covers

Versioning should be read as covering the rules that can affect a computed result.

That includes behavior such as:

• method-selection policy
• statistical formulas and derived quantities
• interval or correction logic
• other computational rules that can change the meaning of the same input

Reproducibility guarantee

The intended guarantee is version-aware reproducibility.

In practical terms:

• the same input under the same relevant version context should reproduce the same computational behavior
• different version contexts may legitimately produce different outputs
• audit and snapshot metadata exist to preserve which context actually produced the result

What reproducibility does not mean

Reproducibility does not mean that all future versions must mimic all past behavior forever.

Instead, it means the product should preserve a clear record of:

• which versioned rules produced a result
• whether a later release introduced a breaking computational change
• how to interpret an older artifact in light of a newer release

Relationship to artifacts

Version context should travel with reproducibility-oriented metadata.

That is why versioning should be read together with:

• audit metadata
• snapshot metadata
• artifact lineage context

For those payload-oriented details, see Snapshot and Lineage Schema.

Reader guidance

When comparing results across time, do not ask only whether the UI looked the same. Ask whether the same versioned computational rules were in effect.

If the answer is no, treat the difference as a versioned interpretive change rather than as accidental noise.