D-Star & Operations · For the Tuesday net

Last-Heard But Not Heard: Why XLX102 Went Back to v2.5.3

By Steve Haney (KC4SMH) · September 2026 · xlxreflector.org

If you only need the punchline On 8 August 2026 we moved every reflector to xlxd 2.6.0. On a busy module B, some stations lit last-heard and never played in the radio. On 1 September we put XLX102 and XLX104 back on 2.5.3. Dashboards stayed at v2.4.6. XLX103 and XLXARG remain on 2.6.0 as quieter test beds. The first Tuesday net after the rollback sounded like the old hangout again.

If you check into the ARRGUS Tuesday net on DCS102B, you already know the rhythm: net control, a roll call that moves, a few longer overs, and a dashboard that never sits still. In August that dashboard started lying to us — or at least getting ahead of the audio. This is the story of why, where the software came from, and why we chose the older version on purpose.

Where XLX came from

XLX is not a commercial product. In 2016 Jean-Luc Deltombe (LX3JL) and Luc Engelmann (LX1IQ) published the XLX Multiprotocol Gateway Reflector under the GPL. It was meant to be D-Star infrastructure that could also speak other digital modes — DMR, System Fusion — if you added the right hardware translator.

That original line is still alive. Version 2.5.3 is what most public XLX reflectors still run. It is the software NO9S put on the air here, and it is the software that carried years of Tuesday nets without anyone asking why last-heard and the speaker on their radio disagreed.

The fork, and why 2.6.0 exists

Open-source radio software forks the way a club splits a project: someone wants a new band, a new mode, a new way of linking systems, and they take a copy and keep building. Andy Taylor (MW0MWZ) — the same call many of you know from Pi-Star — forked LX3JL’s xlxd and published version 2.6.0.

The new tree is ambitious in a good way. It adds M17, more ways to peer with other reflectors, and a lot of work aimed at systems that do transcode — turning one digital voice mode into another so a DMR hotspot and a D-Star handheld can share a module. That is real engineering, and it is aimed at a different kind of reflector than ours.

Our array is a D-Star hangout first. We publish DCS, DExtra, and DPlus. We do not run XLX hardware transcoding between modes. (Peanut is a separate path: phone audio is translated to D-Star on its own box in Sarasota, then it arrives on the module as D-Star like everyone else. That did not change.)

What we did in August

On 8 August 2026 we updated XLX102, XLX103, XLX104, and XLXARG to xlxd 2.6.0. The dashboards went to v2.4.6 at the same time — the themed portal you still see, light and dark. We asked people to report anything odd. They did.

The complaint from busy XLX102 B was almost always the same: I see them on last-heard, but I did not hear them. Net control would call a station that had just flashed on the dashboard. The radio stayed quiet. Someone would double, light the list, and never make it onto the air. On a slow ragchew you might shrug. On a directed Tuesday net with a hundred-plus overs, it is obvious.

It was not your hotspot If your ID-51 or ID-52, Pi-Star, or openSPOT was linked to module B and you were hearing some of the net, you were connected. The miss was on the reflector: last-heard updated when a station keyed, and the voice that should have followed sometimes never left.

Why 2.6.0 stumbled here

Picture the reflector as a net control station that has to announce who has the floor before anyone can talk. Version 2.5.3 does that the instant you key up. Version 2.6.0 waits a beat first — it looks for a hardware translator we do not run, because the new code is built for systems that mix modes. On a quiet module that pause is easy to miss. On DCS102B it showed up as:

A callsign on last-heard with no audio behind it, especially on short check-ins.
A second station that keyed over someone else, sat in a short holding pattern, then vanished without playing.
Overs that started late, so the first word or two never made the radio.

None of that is a knock on MW0MWZ’s work. The software was doing what it was written to do. We pointed it at a hangout that does not have the translator it expected, and a busy D-Star net is an unforgiving test bench.

Why 2.5.3 sounds like home again

On 1 September 2026 we put XLX102 and XLX104 back on xlxd 2.5.3. The dashboards stayed at v2.4.6; only the engine under the hood changed. Linked hotspots blinked and came back in seconds. We did it after the Tuesday net, not during it.

2.5.3 puts the “who is talking” header on the wire immediately. If two people key at once, the second station simply does not get through — the same courtesy the radio world has always had, instead of a half-queued over that lights the dashboard and never plays. Last-heard and the speaker line up again.

We left XLX103 (Manitoulin) and XLXARG (the beta box) on 2.6.0 for now. If a later 2.6 build learns to speak D-Star without waiting for a translator we do not own, we will try it there first.

What you should hear now

XLX102 module B is back on 2.5.3. Same-mode still applies: D-Star hears D-Star; a DMR user on the same module will not. Peanut users on B, C, and E still come in as D-Star. If last-heard and your speaker disagree on a Tuesday net, tell admin@xlxreflector.org — that is how we caught this one.

What we learned

Newer is not automatically better on a production hangout. The Tuesday net is the load test we actually care about. We had rolled back to 2.5.3 once before, in February, when an earlier 2.6 try felt off. August was the second pass, with a current dashboard and a lot more operators listening. The reports from module B were the deciding data, not the version number on the portal.

Credit where it is due: LX3JL and LX1IQ gave the D-Star world a reflector it could run for free. MW0MWZ is pushing that code toward a multi-mode future many sysops want. Our job is narrower. Keep DCS102B honest for the people who show up at 9:15 on Tuesday with a handheld and a cup of coffee.