> For the complete documentation index, see [llms.txt](https://jordan-7.gitbook.io/jordan-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://jordan-7.gitbook.io/jordan-docs/internal/mirror-test-audit.md).

# Mirror-test audit

**Date:** 2026-07-27 **Trigger:** the black-OBS-tile regression shipped green because the gate tested a *copy* of the overlay's reveal logic rather than the overlay.

## What counts as a mirror

A gate case that reimplements our own logic and then tests the reimplementation. The copy can drift from the original, at which point the test still passes and means nothing.

Three things that are **not** mirrors, and were excluded:

* **Black-box tests against a real server.** Most gates spawn `server.js` (or hit an already-running one, as the tempo gates do) and drive it over HTTP/WS or a real browser page. That is the strongest form, not the weakest — no imports needed. An early pass of this audit wrongly flagged all 3 tempo gates (48 cases) on "no project imports"; they drive real mainnet flows.
* **Test doubles for EXTERNAL dependencies.** `MockFrameSource`, `MockCodeChecker`, `StubSettlement`, the fake Privy provider. Standing in for Twitch or a chain is correct; there is nothing of ours being copied.
* **Comments describing a feature** ("gates mirror the MegaChat gate") or an external system's behaviour (Arc's ERC-20 mirroring native balance).

## Result

**Total gate/verify cases across the suite: \~700. Genuine mirrors: 19 (2.7%), in 3 files.** Ranked by what a stale copy would cost:

| # | Where                          | Cases | Mirrors                                                                        | Risk if it drifts                                                  | Status                   |
| - | ------------------------------ | ----- | ------------------------------------------------------------------------------ | ------------------------------------------------------------------ | ------------------------ |
| 1 | `_gate-mpp-clientpath.mjs:144` | 11    | `normalizeTempoTx`, `WALLET_ONLY_METHODS` from `web/lib/join-page.ts`          | **Payment path.** A stale copy passes while real payments break.   | **Drift detector added** |
| 2 | `_gate-lazy-connect.mjs` D+E   | 6     | the overlay's reveal gate — comment said "Mirror of the overlay's reveal gate" | **This is the one that already bit us** (black OBS tile).          | **CONVERTED**            |
| 3 | `_gate-lk-phase3.mjs:170,214`  | 2     | the client's meter pause/resume on reconnect                                   | Billing continues while disconnected, or stops when it should not. | Filed                    |
| 4 | `_gate-tempo-phase3.mjs:182`   | \~1   | client tick behaviour inside a mainnet parity sweep                            | Lower — the surrounding flow is real and would fail loudly.        | Filed                    |

## What was done

**#2 converted.** The D+E block no longer calls `page.setContent('<div id="stage"></div>')` and reimplement the gate. It loads `/overlay` and calls the page's own `markStingerRevealPoint` / `lkAttach` / `maybeReveal` against its own `lkReveal` map and real DOM. Drift is now impossible by construction. Case count went 6 → 7 (gained: the real tile actually drops `.lk-holding`, which the mirror could not see because it had no DOM). Cost: the block now needs a server, which it did not before. Worth it.

**#1 given a drift detector rather than a conversion.** Converting it means driving the real `ensureMppSession` with a funded wallet on a live chain, which is exactly why it was copied. Instead `_verify-mirror-drift.mjs` compares the copy against the original directly, so the day someone edits one and not the other, a test fails instead of quietly meaning nothing. Currently identical. Negative-tested: removing one method from the copy fails the check and names it.

**#3 and #4 filed, not converted.** Both are small, both sit inside gates whose surrounding flow is real, and #4's mainnet sweep would fail loudly on real divergence anyway. Converting them was not worth the churn tonight.

## The honest caveat

A drift detector is weaker than a conversion. It proves the copy still matches today's original; it does not prove either one is correct. #1 remains the riskiest untested-for-real path in the suite, and the real fix is a browser-driven payment gate against a funded test wallet. That needs a wallet that does not exist here.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://jordan-7.gitbook.io/jordan-docs/internal/mirror-test-audit.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.