# Airplane Mode Is Not the Same as Fully Offline

**Public-candidate field note — review required**  
**Participant-observed local-operation evidence; not a security, release, or universal-offline claim**

## The observation

During a running development-alpha Local Workshop, Alex placed the Mac in airplane mode and continued a multi-turn conversation through the application’s pinned local Ollama model.

The narrow observation is useful:

> A running local Workshop continued producing multi-turn responses while the participant reported that the machine remained in airplane mode.

The restraint is equally important. This was a field observation, not an instrumented network-isolation trial.

## Recorded provenance

- Observer and participant: Alex Baron.
- Observation classified: August 21, 2026.
- Application state: already-running development alpha.
- Inference path: local Workshop to localhost Ollama to a pinned local model.
- Model identity and digest were recorded in the private source record.
- Network condition: participant-attested airplane mode, not independently instrumented.
- Raw transcript was preserved with a SHA-256 hash and without cleaning its repeated or clipped sections.

The public derivative does not need to expose the private transcript, local path, or full model identifier to preserve the finding.

## What continued to work

The running conversation remained useful across multiple turns. The local model responded to questions about framework design, organizational comparison, self-description, stewardship, and product curiosity. The application continued displaying its transient-interpretation boundary instead of treating model output as institutional authority.

Disconnecting the participant-observed network condition did not end that already-running local conversation.

## What the observation revealed beyond connectivity

The conversation also exposed weaknesses that matter to local systems:

- present-tense claims about the world still require current sources;
- local execution can improve a privacy or latency boundary without guaranteeing either;
- founder-affirming language requires counterevidence and anti-sycophancy tests;
- comparisons with large organizations must recognize their possible security, evaluation, audit, support, and deployment advantages;
- elegant framework language is not evidence that the framework is correct;
- long context can reinforce the model’s own recurring vocabulary; and
- pricing curiosity is not reliable market research.

These findings kept the observation from becoming an advertisement for the system.

## Why airplane mode is not the claim “fully offline”

This field finding did not test:

- installation or cold start without a network;
- whether the local daemon attempted any network access;
- process-level egress;
- physical network isolation;
- model acquisition, licensing, or updates;
- every device or operating state;
- clean-machine operation; or
- commercial offline readiness.

Airplane mode is one observed condition. “Fully offline” is a system-wide claim requiring a much larger test contract.

## What this demonstrates

This note demonstrates a claim practice as much as a local inference result: record the observed condition, preserve source identity, name the participant-observed boundary, and explicitly refuse the larger claim that the test did not earn.

## Next evidence

A stronger reproduction would freeze application and model versions; record machine and operating-system state; instrument network attempts; test cold start; preserve timing and context depth; distinguish local response generation from current-knowledge claims; and repeat the run under a separately reviewed isolation protocol.

Until then, the finding remains valuable and partial: the running local conversation continued in participant-observed airplane mode. Nothing broader is implied.