Method
How claims are proven
Every statement on this site carries a position on the truth ladder and an evidence type. This page explains both, how the capability census works, and how public data is produced.
The truth ladder
JARVIS’s own ledgers describe progress in rungs, and the site uses the same eight. A capability climbs only when new evidence justifies it, and live evidence expires. The rungs are never merged: a capability that passes its tests is tested, not live.
| Rung | Meaning |
|---|---|
| CONCEPT , rung 1 of 8 | Designed or specified. No working code yet. |
| IMPLEMENTED , rung 2 of 8 | Code exists. Nothing has demonstrated that it behaves correctly. |
| TESTED , rung 3 of 8 | Automated tests exercise it in isolation, often against fixtures. |
| INTEGRATED , rung 4 of 8 | Wired into the rest of JARVIS and tested across component boundaries. |
| PRODUCTION-REACHABLE , rung 5 of 8 | The owner can reach it through the real runtime path, not only through a test harness. |
| LIVE-PROVEN , rung 6 of 8 | A bounded probe ran the real code path on a booted JARVIS on the real machine and recorded evidence. The project treats live evidence as expiring after 24 hours, so every live reading here shows the date it was taken. |
| FORMALLY VERIFIED , rung 7 of 8 | Its critical property is checked by a proof or exhaustive verification. |
| RELEASE-QUALIFIED , rung 8 of 8 | Passed every qualification gate for a named release. |
Qualifiers sit beside a rung without replacing it. PARTIAL means part of the capability is below the rung shown; BLOCKED means progress waits on a dependency or an owner action; DEGRADED means it works with a reduced guarantee.
Evidence types
Numbers are labelled by where they came from, so a calculation is never read as a measurement.
| Type | Meaning |
|---|---|
| MEASURED | An instrumented measurement on the real machine. |
| OBSERVED | Seen happening on the real machine and recorded, without a formal instrument. |
| TESTED | Produced by an automated test run. |
| FIXTURE | Produced against synthetic or recorded fixtures, not the live system. |
| CALCULATION | Computed from a model or formula. Not measured. |
| SIMULATION | Produced by a simulation. Not a physical result. |
| ESTIMATE | An estimate, such as a slicer print-time estimate. Not measured. |
| PHYSICAL | A physical-world result, such as a printed part that was inspected. |
| RECORDED | A past recording. It shows what happened then, not what is true now. |
| DOCUMENTED | Stated in a project document or ledger. Not independently re-measured here. |
| NONE | No public evidence yet. |
Things that are never treated as the same
| This | is not | because |
|---|---|---|
| DISPATCHED | VERIFIED | Sending a command proves nothing about its effect. |
| SIMULATED | EXECUTED | A dry run shows what would happen, not what did. |
| HTTP SUCCESS | EFFECT SUCCESS | A 200 response means a request was accepted, not that the world changed. |
| STALE | CURRENT | Evidence expires. An old observation is not a present fact. |
| CAPABILITY | AUTHORITY | Being able to do something is not permission to do it. |
| TESTED | LIVE | Passing tests against fixtures is not working on the real machine. |
| CALCULATED | MEASURED | A formula result is a prediction until an instrument agrees. |
| RECORDED | CURRENT | A demo recording shows one moment, not today. |
The capability census
JARVIS keeps a registry of 342 capabilities in its own ledger. A capability counts as LIVE only when, within the last 24 hours, a bounded probe ran the real registered code path against a real booted JARVIS process on a real port, with real (or verified-clone) stores, returned the correct result, touched no test double, and wrote an evidence file that exists on disk. Anything short of that is recorded under a lower state.
The census on the home page was recorded on 17 September 2026. Because live evidence expires after 24 hours, those live readings are historical by the time you read them. Registry states in the project's own vocabulary. LIVE means a bounded probe exercised the real code path on a booted JARVIS within 24 hours of the reading; the live readings here date from 17 September 2026 and are therefore historical. The registry declares 341 capabilities but holds 342 entries (one entry has an undeclared group); the site shows the 342 actually present.
The number of live capabilities fell from 256 in late July to 9 in September. That drop came from tightening the definition above, not from removing features. The history is charted on JARVIS Bench.
How public data is produced
- Research reads the JARVIS repositories read-only: ledgers, documents, commit history and evidence files. Nothing on this site reads JARVIS at request time.
- Findings are converted into a small public data format (capabilities, benchmarks, timeline, roadmap, changelog) that allows only repository-relative references.
- An automated scan rejects private paths, emails, tokens, key-like strings and personal names before the site builds, and the built site is scanned again.
- A person reviews wording for accuracy: the truth ladder is applied as recorded, and anything uncertain is marked uncertain rather than guessed.
- Media is reviewed for names, paths, tabs, notifications and other private details before it is published, and is labelled as a real capture or as illustrative.
What is deliberately left out
- Anything about the owner’s private data: email, calendar, finances, health, credentials, private memory.
- Security-sensitive internals that would help someone attack the system.
- Voice samples. See the voice page for why.
- Training data and prompts.
Public data exported 22 September 2026 from revision 893913a9c.