# JG25: speed and evidence Website source review: 20 October 2026. This is a review of published sources, not a new radio test. ## Current evidence at a glance - **Documented behavior:** the current implementation and operator guide describe controls, services, and the beta-licence process. Documentation is not an over-the-air delivery measurement. - **Calculated and modeled:** payload rates account for framing overhead; file timelines include handshakes, turnarounds, and explicit losses. These are not measured wall-clock goodput through radios. - **Simulated:** the FM comparison fixtures record 90%-packet-delivery thresholds for specific modes and channels. They are not universal minimum SNRs or comparable FSQ text-copy floors. - **Historical recorded audio:** the decoder study recovered its 37 complete JG2400/F0 fixtures byte-exactly. This is not current-build whole-file delivery evidence. - **Illustrative screenshots:** the website images show controls and example exchanges, not current-build radio performance. The documented behavior, model, and simulation were checked against [OpenJSQ main at 6164a59](https://github.com/jgoddard2/OpenJSQ/tree/6164a59939dca3c85e6b7b7a263523641da290fe): the [1.0.8 operator guide](https://github.com/jgoddard2/OpenJSQ/blob/6164a59939dca3c85e6b7b7a263523641da290fe/docs/help/OpenJSQ_Help.html), [mode-throughput model](https://github.com/jgoddard2/OpenJSQ/blob/6164a59939dca3c85e6b7b7a263523641da290fe/docs/MODE_THROUGHPUT.md), and [modem-comparison package](https://github.com/jgoddard2/OpenJSQ/tree/6164a59939dca3c85e6b7b7a263523641da290fe/fixtures/modem_comparison_2026-09). The model and thresholds retain their documented implementation baselines. The upstream [JG25 specification is experimental v0.9](https://github.com/jgoddard2/OpenJSQ/blob/6164a59939dca3c85e6b7b7a263523641da290fe/docs/JG25_SPECIFICATION.md). The discussion of v0.3 and earlier builds below is historical context, not the current specification revision. ## What this site does not establish This source review adds no current-build over-the-air demonstration, whole-file radio goodput measurement, or head-to-head weak-signal result against JS8. A current-build radio session is still needed for those claims. Use the [first-contact guide](/start/) for an operator test, and the [modem tables](/modems/#speeds) for the scoped model and test figures. ## Historical evidence review — 6 September 2026 JG25 is designed to move useful data efficiently through ordinary FM voice radios, including the microphone/pre-emphasis and receive de-emphasis paths. The material reviewed in September supported waveform specifications and scoped decoder results. It does not establish a current-release file-transfer speed across those paths. ## Waveform rate is not file throughput The [native JG25 waveform family](/docs/jg25-specification.html#4-native-jg-waveform-family) describes seven raw rates from 1,200 to 2,400 bit/s. JG2400 therefore means 2,400 raw bits per second. Payload framing, acquisition, error correction, turnarounds, acknowledgements and retries reduce the rate of useful file bytes delivered. The implemented wire format has one bearer envelope per PHY frame, with full acquisition and a protected header on each frame. JSQ services provide their own reliability. The separate compact link layer, 32-frame link ARQ window and multi-frame burst optimization in v0.2 were not adopted; v0.3 preserves them as historical proposals in Appendix C. ## What the recorded-audio study reports The [JG2400/F0 phase-body decoder handoff](/docs/evidence/JG2400_F0_PHASE_HEAD_HANDOFF.md) reports CRC-valid, byte-exact recovery of **37 of 37 complete JG2400/F0 real-radio fixtures**: nine held-out August body-CRC failures and 28 existing successes. All passed on the first timing hypothesis. That result has a specific scope: - It tests the JG2400/F0 body decoder against a recorded corpus, not a fresh over-the-air delivery campaign. - The handoff explicitly says this decoder was not yet called by the runtime at that stage. - One additional fixture had a clipped preamble and was excluded from the 37 complete fixtures. The report separately describes finding and decoding its frame in the untouched continuous recording. - Other modes, FEC profiles, current-release end-to-end reliability and file goodput are not established by the 37/37 result. - The document's later integration-gate checklist describes required future tests, not completed results. This is encouraging decoder research. It should not be advertised as a 100% reception rate or a demonstrated transfer-speed advantage over AX1200. ## What would support a microphone-path speed claim A publishable comparison should use the same payloads and radio path for JG25 and AX1200, and record: - Exact app builds, radio models, bandwidth settings and audio interface. - Transmit entry point (including microphone pre-emphasis) and receive audio exit point. - Mode, FEC, acquisition settings, drive levels and receiver processing. - Payload size, successful whole-file verification, elapsed wall-clock time, retries and failed attempts. - Delivered goodput in useful file bytes per second, plus the number of trials and range of results. Report ordinary successful runs and failures together. Keep synthesized-audio tests and recorded-audio replay results separate from live radio measurements. No comparable current-release goodput measurement was available in that historical review. ## Source provenance For the September review, the protocol snapshots and decoder handoff were read from the OpenJSQ application repository's `claude/shared_files_integration` branch, pinned at commit `09c350bac83dfec0f8a6245256d9f3ec3f01fb80`. The copied decoder handoff is `docs/JG2400_F0_PHASE_HEAD_HANDOFF.md`, Git blob `a3767d84d9b972f703ef540a5da63a1b60d9135f`. Its experiment was based on `claude/jg25_fm` at `500ceb2`; its claims retain that historical scope. The source recording corpus is not included with this website. The historical virtual-audio bench findings for build `fde52e7` reported transmit underruns and corrupted cross-station frames. That session verified sender-side and single-instance UI, but could not establish successful inter-station mail, file or relay delivery. That report is retained as historical evidence; it does not establish whether those faults persist in the current build. Its screenshots remain useful as interface examples only. The populated Chats, Files, Network and Inbox images on the website are screenshots of the running application, captured during an example net between three test stations linked by a virtual audio cable. The displayed traffic, paths and statistics are captioned as examples and are not presented as performance measurements. The website guide and images remain published snapshots. Use the pinned current-source links above to distinguish them from the historical integration branch.