THE 10% PROJECT
Useful work needs a receipt and a recipient
RoboWars runs a small, bounded, verifiable computation in the browser — only when you ask for it. Here is exactly what it does, and exactly what it does not.
The idea
A lot of projects promise that playing their game funds science. Most of those promises are decoration. This one tries to be small enough to be true: one tightly specified task, an independent verifier, a receipt you can download and check, and a public statement of what has actually run.
No background mining. No GPU usage. No paid model calls. No work happens because the page loaded.
Mission 001 — Pattern Hunter
The task colours each edge of the complete graph on 43 vertices red or blue and tries to eliminate every monochromatic complete subgraph on five vertices. The search starts randomly or resumes from the best local graph, proposes edge flips, tracks the change in the number of forbidden groups, and recounts everything before returning a candidate.
This is related to the Ramsey number R(5,5). A valid colouring on 43 vertices with zero monochromatic K5s would imply a lower bound of 44 and would need careful independent examination. This deliberately small heuristic is an exploratory prototype, not a competitive research solver.
What is actually running today
One worker at a time, targeting a ~350 ms search budget and at most 1,200 candidate changes, terminated after 2 seconds if it has not returned.
It starts only when you click Try a search, or opt into one job after every tenth completed battle. The automatic option is off by default.
Starting a fight or hiding the page terminates the worker. Nothing continues after you close the tab.
This is local CPU computation, not LLM inference. One job per ten battles is not a measurement of ten percent of anything.
Results stay in local browser storage until you download them. Counters show completed jobs, the best conflict count and measured elapsed time — never “discoveries”.
Check the evidence
Download the latest JSON receipt, then run the standalone verifier. It reconstructs the graph, enumerates all 5-vertex subsets and checks the claimed conflict count using an implementation separate from the browser optimizer. It returns a graph SHA256 and an explicit scientific-status label.
python3 scripts/verify-research.py robowars-research-receipt.json
It does not verify the claimed elapsed time, seed history, number of moves or novelty — and it will not tell you that you found something.
A meaningful ten percent
Defining the fraction honestly means defining it in money, with a visible denominator, atomic reservations, provider-invoice reconciliation and separate reporting for reserved, spent, failed, verified and accepted work. Unused research funds should stay visibly unspent rather than being relabelled as impact.
There is no funded research budget configured today, and this site will not pretend otherwise until there is.