> ## Documentation Index
> Fetch the complete documentation index at: https://docs.parmanasystems.com/llms.txt
> Use this file to discover all available pages before exploring further.
# The CLAIMS.md Discipline
> Every capability claim on this site maps to a line in docs/CLAIMS.md, which maps to a source file or test. How to read it, and how to check any claim yourself.
This page explains a discipline this site follows. For the rule as written for
someone editing this site, see [Contributing](/contributing).
## What it is
[`docs/CLAIMS.md`](https://github.com/parmana-ai/parmana/blob/main/docs/CLAIMS.md) is the
authoritative register of every technical claim Parmana makes about itself. This site
mirrors it; it never overrides it. If a page here ever contradicts CLAIMS.md, the page is
wrong.
## Why it exists
Documentation drifts from code by default, not by malice. A page written honestly on day
one silently becomes a lie the moment the code underneath it changes, unless something
forces a reconciliation. CLAIMS.md is that forcing function: every claim carries a status
tag and a citation, so staleness is checkable rather than assumed away. This site's own
Phase 1 through 4 history is the proof this works in practice, three separate rounds of
this documentation effort found real, live discrepancies (the gateway wiring change, a
mischaracterized replay engine, an undocumented API route) precisely because every claim
here is written against a checkable source, not against memory of how the system used to
behave.
## How to read it
CLAIMS.md sorts claims into three tiers, and this site uses the identical vocabulary
throughout:
* **`[AVAILABLE]`**: built and tested in the current repo. Cites the file(s) and/or test(s)
that prove it.
* **`[PARTIAL]`**: exists, but incomplete, or real and tested yet disconnected from the main
pipeline (see [@parmana/receipt](/reference/receipt) and [@parmana/replay](/reference/replay)
for two concrete examples this site found). States exactly what works and what doesn't.
* **`[ROADMAP]`**: designed, not yet built. Says so in the first sentence, never documented
as if it ships.
CLAIMS.md itself is organized around a promotion lifecycle: a claim moves from idea to
implementation to automated tests to audit to documented proof before it's allowed to become
a public, unscoped claim. Section 5, "Claims We Intentionally Do Not Make," is not a gap
list, it's a permanent boundary, see [What Parmana does not claim](/trust-and-claims/what-we-dont-claim).
## The invitation
Every code block, command, and captured output on this site was run against the real repo,
not written from memory of what the code should do. You don't have to take that on faith:
* **Clone the repo and run the same commands.** [Quickstart](/quickstart) and every
[guide](/guides/write-your-first-policy) are copy-pasteable in order, from a clean clone.
* **Read the cited file at the cited line.** Every `[AVAILABLE]` claim on this site names a
real path, `packages//src/.ts`, check it exists and says what the page claims.
* **Run the tests.** `npm test` (327 tests as of commit `651497a`) and `npm run examples`
(all 60 tutorials plus three scenarios) are both real, both green, both runnable by you.
* **Verify a trust record independently, without trusting this site or Parmana's runtime at
all**, see [Verify a trust record independently](/guides/verify-independently), that guide
exists specifically to make this claim checkable in the most literal way possible: with the
server turned off.
## Next
The honest limits, sourced from CLAIMS.md's own refused claims.
The same discipline, written for someone adding a page to this site.