HydraIssues

Issues / #732
Measure from the heads, not from planted probes: head coverage is the real gap
open unclassified Project: hydrahyperscaler Parent: #714 Reporter: 14 Sep 2026 23:16

Description

RESHAPED 2026-09-15 (owner): the probe belongs in the HEAD, not on a planted probe node.

This is right, and it is what #397 and #402 already specify: a network-quality probe inside the HydraHead client, pre-flight in discoverBody() plus an optional low-rate in-session probe, reported to HydraNeck. The campaign's separate probe-node idea was a bootstrap expedient. The head is the device that consumes the stream, so measuring from it measures the real path with no proxy and no approximation, and it needs no hardware shipped.

It is also already proven: the two venues measured successfully on 2026-09-14 were measured FROM THEIR HEADS. turbo-pancake-76 (cloud-seven) and peppy-dumpling-32 (rupelmonde) are both hydraheadflatscreen nodes.

Head coverage, fleet wide

Venue Heads OS Status Can run a binary today
cloud-seven turbo-pancake-76 macOS online yes, already used
cloud-seven cranky-toaster-86 Omarchy Linux offline yes when online
cloud-seven ipad-head-cederik iOS offline no, needs in-app
cloud-seven quest-head-36afcd - offline no, needs in-app
rupelmonde peppy-dumpling-32 macOS online yes, already used
ad6 cheeky-cactus-86 macOS offline yes when online
sint-niklaas ipad-head-map-35 / 37 / 39 iOS offline no, needs in-app
nerdland none enrolled - - -
gallo-romeins-museum none enrolled - - -

The three real gaps

  1. iPad and Quest heads cannot run a binary, ever. sint-niklaas is iPad only, so it can NEVER be measured by a planted binary and must get the probe inside hydraheadipad. This is #397's scope. It is the strongest argument for the in-head design: one venue is unreachable any other way.
  2. ad6 has a head, cheeky-cactus-86, and it is OFFLINE. That is why the 2026-09-14 ad6 numbers came from cederikmini, a build machine that happens to sit on the venue LAN. Usable, but it is not what a visitor streams to. Bring the head up and re-measure from it.
  3. nerdland and gallo-romeins-museum have NO heads enrolled at all, only bodies (wobbly-llama-92 offline, fluffy-dumpling-87 online). Either their heads were never enrolled in hydracluster, or those venues work differently. Worth answering on its own: a venue with a body and no head is a strange shape, and it means we cannot measure what a visitor there experiences by any method.

What this replaces

Drop the plan to build and ship five portable probe Pis (PLAN 4.3, probe-pi-a to probe-pi-e) as the primary approach. Keep one or two as a diagnostic kit for bisecting a venue LAN, which is what ad6 needs and what its closed Omada neck cannot provide.

The better end state

Once the probe rides in the head client, the campaign stops being an expedition and becomes a query. Every head reports path quality continuously to HydraNeck, and qualifying a region is reading accumulated telemetry rather than mounting a measurement exercise. That is what #402 wanted anyway. The qualification YAML and the per-session record are already the same document (PLAN 4.9).

Related: #397, #402, #706, #714, #724, #733.