Opening your Alchemy Labs workspace.
Loading your game…
- 1 · Set up
- 2 · Train
- 3 · Check
- 4 · Run
Know the grounds. Set the goal. Then train.
- Environment & assumptions
- Measurable goal
- Reference strategy
- Training phases
- Checks & limits
Objectives & optimization
Example: Success at least 90%; score at least 5; steps at most 500; prioritize efficiency. Separate clauses with semicolons. Use supported metrics shown in Use values. Choose game rules and the task in World; text does not create new mechanics.
Alchemy Labs
One environment. Repeatable decisions. Visible evidence.
What you teach
Match real element traits, select atom-balanced equations and compare Q with Ksp. Complete 12 shuffled card puzzles with feedback after every choice.
Curated virtual classroom rules, not chemistry discovery or a physical laboratory. The small action neuron imitates authored scores over public traits and engineered features. It cannot predict unknown reactions, biological traits, yields or safety. No chemical quantities, procedures or physical deployment.
- World: try the reference, then control a round yourself.
- Train: fit a small action-scoring network away from the animation.
- Results: compare paired seeds, inspect losses and activate checked weights.
- World: test a new seed. Save project to keep your model.
Alchemy Labs
Watch decisions change the outcome.
Choose a card with A–D in manual mode, or run the reference. Twelve decisions complete a round. Atoms are transformed or separated; they never disappear.
Mixed challenge ready.
Inspect the real periodic table · all 118 elements
Choose an element to inspect its source traits. Scroll the table sideways on a small screen. Missing values stay unknown; predicted states keep the source wording.
Select an element. These are source data, not trained predictions.
Source: PubChem Periodic Table, retrieved 2026-11-06. The source snapshot includes approximate or predicted values. IUPAC data-access reference.
Rules, sources and limits
The equation puzzles check neutral formulas and atom counts. They do not model charges, reaction rate, heat, concentration, yield, hazards or unknown reactions. Balanced equations describe possible transformations; they do not establish that a reaction occurs.
The solubility puzzle supplies Q/Ksp directly. Under its idealized fixed conditions, Q greater than Ksp favours precipitation, Q less than Ksp is unsaturated, and equality is the saturation boundary. No recipe or physical experiment is generated.
OpenStax: reaction classes · OpenStax: Q and Ksp. Explanations are original summaries; element data retain PubChem provenance.
Manual buttons work with You · manual. Step advances the selected policy. One step is one decision, including the reference opponent response where applicable. Reset after a round ends.
Rules and what the network sees
Match real element traits, select atom-balanced equations and compare Q with Ksp. Complete 12 shuffled card puzzles with feedback after every choice.
Curated virtual classroom rules, not chemistry discovery or a physical laboratory. The small action neuron imitates authored scores over public traits and engineered features. It cannot predict unknown reactions, biological traits, yields or safety. No chemical quantities, procedures or physical deployment.
An Alkamy Labs experiment · Explore our labs ↗
From features to an action.
8 learned weights → one score for each legal action.
A single linear neuron scores each action. It receives these engineered features, not a teacher’s chosen action:
- Element family match
- Element state match
- Atoms conserved
- Requested reaction class
- Requested product present
- Unsaturated × no solid
- Saturated × equilibrium
- Supersaturated × precipitate
Ridge least squares fits the weights from synthetic reference scores. Rules supply legal actions, feature extraction and opponent behaviour. This is imitation learning. It does not discover an optimal strategy through self-play.
Curated virtual classroom rules, not chemistry discovery or a physical laboratory. The small action neuron imitates authored scores over public traits and engineered features. It cannot predict unknown reactions, biological traits, yields or safety. No chemical quantities, procedures or physical deployment.
Train. Freeze. Compare.
Active weights stay in place while the candidate is checked.
64 fitting seeds, 12 validation seeds and 24 final seeds. Validation selects between two ridge strengths; final tests cannot change the selected weights. Maximum 25 seconds of computation, 30-second worker timeout and two attempts per training round.
Ready · no provider or API key needed.
Activation gate: ≥80% near-best reference-score agreement, finite compatible weights and final mean score at least the untrained baseline and a checked incumbent, when present. Passing this small test does not guarantee future wins.
Uses fresh teacher examples and the current two-attempt training round. Review each candidate before activation. Run keeps active weights fixed.
Off. Train manually, or start another training round when both attempts are used.
What changed?
Compare the same seeds. Keep the losses.
Train or check saved weights for a measured comparison.
Complete a round or training run to see observations and a controlled next test.