The one-slot context model — time and branches from a single field
2-min readUpdated Jun 18, 2026
The idea
Today a fact is a triple: (subject, relation, object).
Add one slot — the context the fact is asserted in: (subject, relation, object, context).
That's it. No "time" anywhere.
Contexts are just nodes with one optional link
A context can declare after — the context it follows. That single link is the only structure:
- a chain of
afterlinks (d1 ← d2 ← d3 …) is a timeline - a fork (two contexts both
afterd3) is a branch / what-if - no
after(unordered labels) is a set of perspectives
So you never build a time engine. You build a small graph of contexts, and time is just the shape "one long chain."
Projection = pick a leaf + resolve
To read the graph: pick one context as the leaf, walk its after chain back to the root, and
for each (subject, relation) take the value from the latest context in that chain that asserts it
(latest-wins). Everything outside that chain is invisible.
| pick leaf… | you get |
|---|---|
d3 |
the world as of day 3 (time / as-of) |
d5 |
today (newest leaf on the main chain) |
wt |
the "what if we'd kept the tree" branch — a different world, same data |
Same field. Same resolver. Different leaf. Time and branching are the same operation.
The one rule you must pin
When two contexts in the selected chain disagree, the later one wins (later = further from root). That single rule is the whole abstraction:
- chain + later-wins → time
- branch + later-wins → scenarios
- unordered + explicit priority → overlays (draft over approved)
- unordered + keep-both → perspectives (Design vs Eng, side by side)
No clock, no fold, no slider, no as_of baked into the graph. Just contexts, an after link, and
later-wins. See context_demo.py for the birdhouse running both a time projection and a branch.