ADR018 TemporalDimensions

Title: Temporal dimensions

• Status: Proposed
• Proposed: D2026-09-07
• Follows: ADR014 Literals
• Followed by: ADR027 DatesAsNames
• Related: ADR025 RelationshipSideBlocks, ADR029 EdgeBlocks, ADR030 DescribeInEdgeBlocks
• Affects: Describe, Dimension, Parser, Ui
• Pairings:
  • ADR025 RelationshipSideBlocks — parallel thinking over the same ground

Context

Prose has no literals (014), and no literals means no dates. So there is currently no way for a universe to say when something happened, or when a fact was true.

The want first showed up in this universe. These records exist as concepts, and knowing when each decision was made would let a consumer draw a timeline of how sprig and prose evolved. Today those dates live in markdown headers and the prose nodes carry none, so the field guide can show the decisions and the edges between them but cannot place them in time.

The naive answer is a date literal. That is refused by 014, and upstream of 014 by 013, so the question was originally framed as whether time earns an exception. If time gets a pass, does that weaken the argument against literals generally?

That framing turns out to be the wrong one, and two modelling exercises are why.

Schitt's Creek, modelled by episode. The useful sentences are "this wasn't introduced until Season 4 Episode 7," and, harder, "show me the universe as of some point — which relationships were true then." That is not a request for a date. It is a request for a fact to be scoped to a stretch of an ordering.

Futurama. How old is Farnsworth? It depends on when you ask, and a date of birth does not answer it. Fry is worse: there is more than one timeline involved, and they do not agree.

Those two are not the same feature, and treating "temporal dimensions" as one feature is what has kept the question stuck.

Decision

The decision this record makes is that this is three questions, not one. It proposes no syntax, and it does not settle whether prose gains a temporal facility at all.

1. Ordering. A sequence of things — episodes inside seasons, releases, stages. Prose already does this. follows and isFollowedBy are relationship pairs like any other, and the field guide already detects ordered runs. This axis needs nothing, and it is worth noticing that the ADR chronology problem that prompted the question is largely this axis: follows edges order these records without any record holding a date.

2. Validity. A concept or an edge is true across part of an ordering. "Not introduced until Season 4 Episode 7" is this. It needs structure prose does not have, but — and this is the part worth recording — it may not need literals. If the timeline is modelled as concepts, Season4Episode7 is an identifier, and a validity interval references identifiers the way every other endpoint does. That would put this axis entirely inside the existing language rules rather than against them.

3. Derived values over time. Farnsworth's age. This requires arithmetic over values, and multiple disagreeing timelines on top of that. This axis is refused already, not by 014 but by 013: it is behaviour, and prose does not carry behaviour. Nothing in this record reopens that.

So the exception question — "does time get a pass on literals" — only arises for axis 3, and axis 3 is refused for a different and older reason. Axes 1 and 2 do not obviously need the exception at all.

Consequences

What stays easy. Ordering works today, with no new syntax and no decision required. A universe that wants a sequence should use a relationship pair and does not need to wait for this record.

What this defers. All syntax. There is no proposed spelling for a validity interval, no decision about whether it attaches to a concept, an edge, or both, and no decision about whether the feature ships at all.

Cost of leaving it open. The prose ADR nodes cannot be placed in time, so the field guide cannot render the timeline that motivated the question. Dates stay in markdown, which means the two ADR surfaces carry different facts — tolerable, and currently the working arrangement.

What to watch, and this is the reason to record now. The pressure is arriving from three unrelated directions at once, and each one looks small on its own: ADR metadata wanting dates, expects wanting cardinality (011), and universe modelling wanting validity. Three requests, none of them calling itself a request for literals. That is exactly the erosion 014 named. Deciding them one at a time, on their individual merits, is how the line moves without anyone moving it.

What would have to be true to accept a temporal facility. At minimum, axis 2 would need a real universe where modelling the timeline as concepts is demonstrably not enough. Until someone tries that and it fails, the cheaper answer has not been ruled out.

Alternatives considered

A date literal. The naive solution, and the one the question started as. Not rejected outright here, because that is 014's call and this record does not revise it. Recorded instead as the option that is only needed for axis 3, which is refused on other grounds — which makes the literal exception a solution to a problem this record does not have.

Model the timeline as concepts. A season dimension, an episode dimension, follows between them. Costs nothing, exists today, and covers axis 1 completely. It is the option that has to be tried and found wanting before anything heavier is justified.

Validity intervals referencing concepts. The interesting middle: an edge that names the stretch of an ordering across which it holds, using identifiers rather than values. Not proposed as syntax, but named as the shape most likely to survive scrutiny, because it is the only version of axis 2 that does not require literals.

Solve for the Futurama case up front. Multiple timelines, derived ages, disagreeing histories. Rejected as a design target. It is a real modelling problem and it is not the problem most universes have; designing for it would buy a general temporal engine at the cost of everything 013 protects.

Keep temporal metadata outside prose entirely. Markdown headers, frontmatter, receipts. This is already what happens for these records, and it works. Worth naming because it is the reason the question can stay open without anything being blocked, and because it is the honest answer for metadata about documents even if prose later gains something for facts inside a universe.

Evidence

The two modelling exercises are the evidence, and they point in different directions, which is the finding. Schitt's Creek by episode wants a fact scoped to a stretch of a sequence and never needs a number. Futurama wants a computed value and cannot be satisfied without one. A single "temporal" feature would have to serve both, and the second one is behaviour.

The ADR chronology case is evidence for axis 1 specifically: these records are already ordered by follows edges, and that ordering is legible in the field guide today without any record carrying a date.