Explore ten animated programs. Here, choosing a scenario creates a new source with those values. For programs that accept changing inputs, use the task examples.

Try a Quineling.
See how it works.

Pick a creature and a prepared situation, then run its task. Follow the calculation through its body and make a copy that checks its answer against the original. Each Quineling can also produce its complete source—the code needed to rebuild it.

01 — 10

Inspect the calculation.
Recover the source.

01 / TASK

It does useful work.

A program is a network of operations: read values, calculate a score, choose a route, or allocate a budget. Each operation runs after its inputs are ready.

Lanternkeeper combines lamp readings, checks inspection evidence and permission, and decides whether to simulate a repair.

02 / BODY

You can see its structure.

Operations become colored organs. Dependencies—the places one operation uses another’s result—become connecting filaments. Click a program line to find its organ.

A chosen body family supplies the silhouette. Program depth and branching shape its folded ribbons. Compression and depth make the folds luminous; inspection reveals the colored organs and connections.

03 / QUINE

It outputs its own source.

A quine is a program that outputs its own source code. Each Quineling computes its task result and constructs its complete source as a separate output.

Reproduce builds and runs a fresh program from that emitted source, then verifies that the source and task results match.

Everything runs in your browser: a JavaScript interpreter executes the program, and a canvas draws its body. Motion displays structure; the Run controls execute tasks. Programs have up to 64 task operations and 1–8 task cycles.

Watch a thought take shape.

Public Midnight.city artwork illustrates this authored task. Follow the graph’s local calculation into its equations and animated body.

MIDNIGHT.CITY AGENT ART
Pixel agent artwork retrieved from Midnight.city
Declared task · local simulation
EXECUTABLE GRAPH
Complete task graph (JSON)
QUINELINGSame source. Same renderer.
01 / A declared thought

Operation type → organ color & frequencyConnections → organ size & filamentsDependency depth → position in the bodyTask cycles → return rings

Reading the formula: θ is the angle around an organ; φ is its animation phase; r is the distance from its center. The frequency counts the repeating pattern around the organ.

Following the thought changes the view. Run this task computes the result and emits its source.

Agent artwork from Midnight.city · Asset provenance. The bubble describes a task paired with an explicitly authored program.

Select Lanternkeeper in the collection, choose its default-confirmed-fault scenario, set Task cycles to 1, then return to the walkthrough. Follow one calculation from input to decision, then verify the program’s reproduction.

  1. Find a calculation. In the walkthrough, select faultScore. It combines readings [1, 1, 0.5] with weights [4, 2, 2]: (4 + 2 + 1) / 8 = 0.875. The connector points to its mint-colored organ with a 16-cycle primary ripple.
  2. Run and read the result. Press Run task + emit source below. The score passes the 0.625 threshold; supporting evidence and permission allow a simulated repair. The program also emits its full source.
  3. Build the healthy-lamp scenario. Choose healthy-lamp-guard-false under Task scenario. This writes different lamp readings into the source. Run that program to see a score of 0.125 and a skipped repair. Replay step → follows recorded operations without running them again.
  4. Verify a fresh generation. Press Reproduce. A fresh program is built from the emitted source and executed. Its source and task results must match its parent’s.
  5. Recover the code from data. Try either recovery control. The complete source is reconstructed from wave samples or exact RGB records and checked by execution.
LOADING
EXECUTABLE SPECIMENGENERATION 0
Select a colored organ to inspect its operation.

Choose colors for operation types, a value from a recorded run, or a plain body. Color data used to recover source is a separate encoding.

Shape the motion

Repeating cycle returns to the same pattern. Evolving harmonics combines frequencies so the pattern changes over time.

Strength of the body’s expansion and contraction. Zero removes this part of the motion.

Strength of the wave traveling across the body. Zero removes this part of the motion.

Changes how far parts of the gesture lag behind one another.

Recover the complete program.

The body shows program structure. Its genome stores the complete source in two independently recoverable forms: wave data and exact RGB color records. Either can rebuild the program, including its task inputs and body design.

Wave data: each band stores 32 coefficients that encode source bytes in a cosine wave. Recovery uses 65 numerical samples from every band to reconstruct those coefficients and the source.

RGB data: each exact color record stores a source byte. This encoding is separate from the organ colors, which identify operation types. Recovery uses the underlying data, not a screenshot of the creature.

View one section of the encoded source. Recovery uses the data from all sections.

32 coefficients per band · 65 numerical samples · exact byte colors

These checks recover the source and run the recovered program to verify it.

Loading source…
DOWNLOAD
Inspect complete program source

A little more about the creatures

Why give a program a body?

A body gives you a visual index into a program: select an organ, inspect its operation, and follow its inputs to the result. Compare a lamp-repair decision, a route, or a budget across prepared scenarios, then reproduce the exact program that made the decision.

What does Living Thoughts mean?

Living Thoughts describes declared tasks taking the form of animated mathematical lifeforms. The thoughts shown here are explicitly authored descriptions. These programs compute bounded tasks and reproduce their source; the animation gives their structure a living form.

Does every program have a unique-looking creature?

The full source is preserved in its genome. The body is a designed visual summary, so different programs can look similar. Operation colors, frequencies, connections, and the chosen family explain the anatomy; genome recovery checks exact source identity.

How does the source reproduce?

The program quotes its task and a smaller constructor body as data. Running the constructor rebuilds the complete expression and emits it. Reproduce parses that output and runs a fresh instance. The generation counter records verified copies with matching source and results.

Read the constructor and encoding specification →