last verified · 2026-07-16

Evidence & glass

Two words describe how Lumis earns your trust: evidence and glass.

Evidence means a “done” always comes with something you can open and check — a diff, a test run with its output, a screenshot, a decision record. Glass means you can always see who is doing what, why, with what evidence, and what still needs you. Neither one is a slogan. Both are things the product actually does, on every piece of work, every time.

“Tests pass” is a claim. A captured exit code, with the output that produced it, is a receipt. Lumis only moves work forward on receipts.

What counts as a receipt

A receipt is generated by a real run, not typed by an agent. When you look at a piece of work, you can expect to find receipts like these attached to it:

A diff — the actual lines that changed, not a description of them.
A test run — the command that ran, its full output, and its exit code.
A screenshot — for anything visual, captured from the real page.
A review decision — who reviewed it, and what they approved or sent back.
A cost record — what the work actually spent, when spend is involved.

If a step has no receipt, the product says so plainly instead of hiding the gap. It never shows a claim as if it were a receipt.

What “in glass” means day to day

At any point you can look at a piece of work and answer, in a few seconds: who is working on it, why it exists, what they’ve done so far, what proved it works, and whether anything is waiting on you. There’s no separate status meeting and no “let me check with the team” — the same evidence that convinces the reviewer is the evidence you see.

A worked example: one ticket, start to finish

Here’s what that looks like end to end, using a real shape of ticket: a reported broken link in a docs page footer.

Plan

An agent picks up the report and writes a short plan before touching anything: what’s broken, what the fix is, how it’ll be checked.


Ticket #482 — "Setup guide" link in footer returns 404
Plan: the href points at a page that was renamed. Update the href to the
current route. Verify by running the link-check tests and loading the page.

Build

The agent works on its own branch and opens a diff — never a direct edit to the main branch.


--- a/components/Footer.tsx
+++ b/components/Footer.tsx
@@ -12,7 +12,7 @@
-        <a href="/docs/setup">Setup guide</a>
+        <a href="/docs/set-up-a-machine">Setup guide</a>

Review

A second agent — never the one who wrote the diff — reviews it against the ticket’s plan before it can move any further.


Review by: reviewer agent
Verdict: approved
Note: href now matches the live route; no other footer links affected.

Test run

The change runs against the real test suite. The output is captured verbatim, not summarized.


$ pnpm vitest run tests/links.spec.ts

 ✓ tests/links.spec.ts (3 tests) 412ms
   ✓ footer links resolve to a real route
   ✓ no external link opens without rel="noopener"
   ✓ footer renders on every page

 Test Files  1 passed (1)
      Tests  3 passed (3)
   Duration  1.18s

exit code: 0

Gate decision

The ticket moves to Done only because two things are both true and both on record: the reviewer approved it, and the test run exited 0. Either one missing holds the ticket where it is.

Every step above is a real artifact, not an agent’s summary of one. That’s what “in glass” buys you: you don’t have to take the crew’s word for “done” — you can open the run that proved it.

This is the same chain you’ll find behind any piece of work in Lumis: a plan, a diff, an independent review, a captured test run, and a gate decision that only fires when the evidence backs it. Nothing about it is hidden, and nothing about it is a description standing in for the real thing.