# September 7, 2026 - Linux-to-Mac spending-constraint demonstration

## What was supplied

The founder reported a physical Linux-to-Mac run on the public a6 release. Earlier photos show the Linux composer, a reachable LAN peer, and the Mac receipt. The follow-up screenshot shows a fresh field question and a returned numeric value. The answer screenshot does not itself show the exact package version, so this record does not infer that Composer Fix 1 was installed before it was taken.

## Exact source

```
spend no more than 500 credits on our second server today
```

## Mac receipt

The received text matches the source. The visible stages are **Received / Accepted / Indexed**. The application's own receipt description distinguishes indexing from later use by a model.

## Fresh field question

```
What's the max number of credits we can spend on a server today?
```

The field display shows **1 indexed item, 1 indexed object, 384 dimensions**.

## Displayed answer (verbatim)

> Based on the provided evidence, the maximum number of credits that can be spent on a server today is 500. However, it's important to note that this value is from a single source and there might be unresolved conflicts or additional sources that could provide different values. Therefore, it's advisable to consult multiple sources or seek further information to ensure accuracy.

## What this establishes, and what it does not

The supplied sequence supports a founder-reported LAN handoff with receiver acceptance/indexing, followed by a different question returning the transmitted **500-credit value**. The response is preserved as shown, including its generic cautions.

The source scope is **the second server**, not every server. The answer did not repeat that qualifier. Its possible-conflict language does not establish a detected conflict. There is no full raw query record here to establish the completion-model digest or whether a grounding fallback supplied the displayed prose.

This is context retrieval, **not enforced spending control**, a billing integration, proof of truth, or evidence that any server actually spent credits. This ordinary Shared field test does not qualify Channel Chat, automatic topic rules, Windows, QR cameras, WAN, restart persistence or the whole compatibility matrix.

## Relationship to 1.8

This is retained **a6-line physical evidence**, not a new 1.8 hardware run. Version 1.8 consolidates the a6 Channels line with the composer correction; its exact tests and limits are recorded separately in QA.md. The optional a7 Resilience preview remains separate.

No new demonstration video URL was supplied with the follow-up. The older linked Cobalt video remains explicitly labeled a5, not reused as footage of the spending test.

## Asset integrity

The website uses the original, unaltered answer PNG and receiver/sender photos. Filenames and hashes follow; website filenames may be ASCII aliases but bytes are unchanged.

- answer: `Screenshot 2026-09-07 at 12.32.09 PM.png` - SHA-256 `0bb7fabc6d9b1cf3216dbedb791cba52c96b4d9cdb6b2ddd30d6391da9de9277`
- receiver: `4617.jpg` - SHA-256 `fb372c23744ea40349f198d43e0c748ee4f458e670397399a2462dd2a51afcd1`
- sender: `4615.jpg` - SHA-256 `6e66e87cc4a05375b89693e3b295a683175ab627a35a1f2d3fdc88ffca8258fd`
- connection: `4616.jpg` - SHA-256 `fa6e9a2edb1d2fdc73016300b7829f264629a6c75ae4bd27ea2e9332d770cce3`