BrainblastbrainblastComing soon
Contribute · Earn

Get paid for the
code errors you've already hit

Turn a hard-won bug into a proven VTI — the vulnerable code, the fix, and a check that proves the difference. Prove it, cite the commit it lives in, and submit it in one command. Earn as labs license the corpus to train and evaluate on.

Coming soonSee how it works

Paid when your trap is used

When labs license the corpus to evaluate and train models, value routes back to the contributors whose VTIs get used — weighted by corroboration × severity. You're paid for traps that catch real bugs in the wild, not for volume.

It compounds

A proven trap isn't a one-off bug report — it becomes a permanent VTI in the corpus buyers train and evaluate against. As the fleet corroborates it across more repos, its score rises, and so does your share.

Open and permissionless

No application, no gatekeeper, no reviewer in the loop. If your submission clears the automated gate — RED→GREEN + secret scan + provenance — it's in. One command submits.

The incentive

What you could earn

Each period, a fixed share of license revenue — the contributor pool — is split by score. Your cut is your VTIs' score (severity × corroboration) over the whole corpus's. Paid on usage, not volume: a few sharp, well-corroborated traps can outweigh a pile of shallow ones.

Estimate your share

illustrative
Verified traps you contribute12
Typical severity
Avg corroboration (repos per trap)
Quarterly contributor pool (assumed)$100,000
Illustrative annual share
$2,628
0.66% of the pool · $657/qtr

A model, not a quote or guarantee — the share is computed against the live corpus, but the pool is an assumption and there's no revenue yet. Real payouts depend on total licensing, how often your records are used, and their severity and corroboration. The on-chain payout is rolling out.

The process

From bug to corpus

01

Prove it RED→GREEN

Author a candidate that catches the bug and prove it locally. The same oracle the dataset and benchmark use must FAIL your vulnerable fixture and PASS your fix. This is the gate — non-reproducing work never makes it in.

brainblast pack init ./my-pack --id my-sdk
brainblast pack validate ./my-pack
# vulnerable → RED fixed → GREEN
02

Cite the real source

Point at the exact commit the bug lives in — owner/repo@<sha>:path — and include the verbatim vulnerable line as evidence. On submission the server fetches that file at that commit and confirms the line is really there. A mutable branch is rejected; an invented finding can't clear it. This is what replaces human review.

"provenance": {
"sourceRef": "owner/repo@<40-hex-sha>:path/to/file.ts",
"evidence": "skipPreflight: true"
}
03

Submit — no PR

One command posts your candidate straight into the corpus. The server re-checks it end to end — RED→GREEN + secret scan + provenance — and only records what passes. No fork, no branch, no reviewer, no waiting. (A pull request works too, if you prefer.)

brainblast submit:vti --candidate ./my-trap.json
brainblast submit:vti --candidate ./my-trap.json --dry-run # check locally first
Coming soon
04

Earn on usage

Your confirmed findings become VTIs in the corpus, scored by severity × corroboration. As it's licensed for training and evaluation, value routes back to the contributors whose records are used. Optionally bond $BRAIN on a VTI to signal confidence — it's slashed if the trap ever stops reproducing.

What makes a valid submission

The gate checks every one of these automatically — no reviewer in the loop.

A real SDK or protocol bug — a way correct-looking code silently breaks.
A minimal vulnerable snippet and the corrected fix, in TypeScript, Rust, or config.
The pair must reproduce RED→GREEN: the check fails the vulnerable code and passes the fix.
A commit-pinned source citation + the verbatim vulnerable line, so provenance can confirm the trap is real.
No secrets — keypairs, base58 secrets, or mnemonics reject the whole submission automatically.
Pinned to the SDK name and version the bug actually applies to.

How the money works

Aligned so only honest, reproducing work earns.

Data dividend

Your confirmed findings become VTIs in the corpus, scored by severity × corroboration. As it's licensed for training and evaluation, value routes back to the contributors whose records are actually used — paid on usage, not volume.

$BRAIN is optional

You never need a token to contribute. $BRAIN is an optional rail for buyers — pay in it for a standing 10% discount — and the access tiers price in $BRAIN held. USDC we take at full price is bought back into $BRAIN and routed to contributors.

Optional confidence bond

Bond $BRAIN on a VTI you're sure of to signal confidence and amplify your share. The reproduction gate is the slash trigger — if the trap ever stops reproducing, the bond is slashed. Opt-in, never required to contribute.

Contribution, the RED→GREEN + provenance gate, the corpus, and the $BRAIN confidence bond (registration + slash-on-non-reproduction) are live today. The on-chain payout of the data dividend is the remaining operational step, rolling out.

Contribute your first trap

Participation is open — no application, no gatekeeper. Prove it RED→GREEN, cite the commit, and submit.

Coming soonRead the docs