← return to sky

World Interpreter Engine

A minimal engine that turns real-world signals into meaningful, replayable events and human-readable timelines.

signal interpretation EXPERIMENTAL

World Interpreter Engine starts from a practical frustration: raw data is often technically correct and still useless.

Electricity usage, time ranges, sensor readings — none of them are stories on their own. The engine maps raw signals into events, then turns those events into a timeline that can be read like memory.

[event]   power_draw 1.4kW · zone:kitchen · 18:42–18:51
[home]    Someone cooked.
[rover]   Engine bay drew current for 9 minutes.
[cyber]   Synth-grease flowed through the kitchen subroutine.

Same event, three different narrators. The interpreter is swappable.

Why it exists

Because 120W does not mean much alone.

“Someone cooked”, “the AC ran too long”, “something abnormal happened” — those are closer to how a person actually understands a place. Numbers want to be facts. Events want to be stories. The engine is the layer that decides which event in the firehose deserves to become a sentence.

Shape

The engine has one job: render(events, interpreter) → string[].

events is a typed array — actor, device, zone, duration, signal value. interpreter is a template map keyed on the event type with placeholder slots ({actor}, {device}, {zone}, {duration_min}). Swap the interpreter, change the voice. The events do not move.

Three presets ship with the demo: Home (everyday domestic narration), Rover (Mars-mission control), Cyberpunk (gritty replay of the same data). The point is not the presets. The point is that a single signal stream can carry many readings, depending on whose lens is over it.

Where it sits in the ecosystem

This was the precursor.

The pattern it sketches — signal → interpreter → narrative — is exactly the pattern DreamLink later picked up at production scale, with GroundSense for baseline-aware drift detection and DreamFlow for the output grammar. World Interpreter Engine is the toy version, the demo where the architecture is visible without the runtime weight.

HomeLog’s timeline borrows the same intuition. The interpreter view is what eventually lets a household timeline read like memory instead of a list of relay flips.

What it says about the builder

I keep wanting to see machine output the way I see a person retelling their day — selectively, in the right voice, with the boring parts left out.

This little engine is where I first wrote that idea down in code.