How every score on this site is produced, and what we refuse to do to get one.
A privacy claim that cannot be checked is a slogan. We publish the method so a stranger can catch us lying.
| Dimension | Weight | What a 10 requires |
|---|---|---|
| Telemetry and egress | 25 | Published destination allowlist; cold idle produces zero unlisted packets, independently confirmed. |
| Verifiability | 20 | Reproducible build verified on the scored platform. |
| Server-side retention | 15 | Subpoena answer is two timestamps or an empty set. |
| Ownership, jurisdiction, business model | 12 | Nonprofit or cooperative, no ad model, strong process protections. |
| Local data exposure | 10 | All artifacts encrypted at rest under a user-presence key. |
| Dependency surface | 8 | Small audited dependency tree, signed and pinned update channel. |
| User control and exit | 5 | Usable export, real deletion, no identity hostage-taking. |
| Incident record | 5 | Flaws found, disclosed honestly, fixed fast. |
| Tier | Meaning |
|---|---|
| T1 | We observed it, saved the artifact, and the procedure is re-runnable. |
| T2 | A primary source anyone can open: a vendor repository, a court filing, a regulator advisory. |
| T3 | Credible second-hand: reporting, an academic paper, an independent audit summary. |
| T4 | A claim with no artifact. Never allowed to move a score. It appears in an appendix, labelled, and that is all. |
These are the only two places where a score is limited by something other than evidence. Both exist because the alternative is pretending to know something we cannot.
no reproducible build verifiability <= 3 experimental, unverified verifiability <= 6 verified, bit-identical verifiability <= 10
Reading the source proves what the code does. It does not prove the binary running on your machine is that code. Reproducible builds are the only bridge, and nothing else is.
Closed source is not penalised. It starts lower and climbs a public ladder. The ceiling is the rung you are standing on.
L0 unattested -> 3 nothing but your word L1 declared -> 5 SBOM + signed hash + published destination allowlist L2 observed -> 7 someone other than you watched this exact binary L3 reproducible -> 9 anyone can rebuild it and get the same bytes L4 contested -> 10 someone tried to break it and published the attempt
A court filing lifts the ceiling entirely, because it needs no vendor cooperation.
a vendor running our harness produces a DECLARATION, not an OBSERVATION.
A self-run test is a claim about oneself, so it caps at L1. But L3 and L4 survive a self-run, because a public rebuild recipe and a published adversarial review can be re-executed by anyone. That is the whole escape hatch: reproducibility converts a claim into a check.
| Run | What it does |
|---|---|
| D1 | Cold idle. Fresh install, no account, one hour. Capture every packet. |
| D2 | Normal use, 24 hours. Capture all egress. |
| D3 | Device access trace: audio, camera, location, clipboard, contacts, screen. |
| D4 | Update cycle. Capture the channel, verify the signature, check the pinning. |
The vendor publishes a destination allowlist. Every destination observed in D1 and D2 must appear on it. Anything unlisted is a defect, regardless of what it turns out to be.
It cannot prove the absence of a backdoor. Rice's theorem makes non-trivial behavioural properties of arbitrary programs undecidable, and a backdoor triggered by remote configuration or by message content is invisible to static inspection.
So the honest claim is not purity. It is this: we can prove what a specific build did while we watched, and we can make a false statement expensive to sign. An attestation raises the cost of a backdoor and shortens its shelf life. It is not a certificate of innocence, and anyone selling you one is selling you nothing.
brainix will be scored by this instrument before it is advertised by it. "Zero telemetry" will mean a captured packet log with nothing in it, published, dated, and re-runnable, or we do not get to say it.