musashi + ΘΕΛΩ — Tidbyt app

Provenance: ⏳🤖 LLM-generated, pending human review · Layer: deployed network (musashi) via a third-party explorer API · Verified: 2026-09-22 (rendered with pixlet 0.34.0 against live data)

A Pixlet/Starlark Tidbyt app that alternates two frames on the 64×32 display: the musashi testnet's Leios health, then this pool's standing. Modelled on functionally/tidbyte-claude — same left-tile-plus-dense-rows shape, same config surface, same cadence.

What it shows

Frame A — network (~4.4 s)

+--------+-------------+     tile:  certified share of announced EBs,
|        | ep      62  |            colored by health
|  41%   | tip     21s |     ep     epoch at the chain tip
|  CERT  | tps     15  |     tip    age of the tip — the liveness signal
|        | eb      62k |     tps    average transactions per second
| green  | bls  ok 147 |     bls    committee-key coverage across all pools
+--------+-------------+

Frame B — this pool (~4.4 s)

+--------+-------------+     tile:  ticker over a status word, colored
|        | act   ep64  |            EXIT / NOBLS red, PEND / IDLE amber,
| THELO  | stk   4.2%  |            SEAT green
|  PEND  | seat    --  |     act    the epoch the current parameters take effect
|        | blk      0  |     stk    share of active stake (ada before activation)
| amber  | crt      0  |     seat   Leios committee seat, or -- when unseated
+--------+-------------+     blk    blocks this epoch (lifetime before activation)
                             crt    certificates this pool has signed

The tile carries the signal you can read across a room; the rows are for when you walk over. tom-thumb gives 4×6 px glyphs, so the right column holds five rows of about nine characters — which is the whole reason the numbers are abbreviated rather than labelled in full.

Health rules

Signal	Green	Amber	Red
Certified share of announced EBs	≥ 35%	20–35%	< 20%
Tip age	< 60 s	60 s – 5 min	> 5 min, or live: false
Committee keys	all active	some neither active nor upcoming	—
Pool BLS key	active	—	anything else
Pool seat	seated	active but unseated (IDLE)	—
Pool lifecycle	—	PEND before its activation epoch	retiring (EXIT)

The 35% and 20% thresholds come from this repository's own modelling: the 14-slot certification gap strands roughly half of all announced endorser blocks, so a certified share near 40% is the expected steady state, not a problem. See the protocol-parameter note. A pool whose key is registered but not yet effective counts as pending, not missing — that is how any freshly registered pool looks for an epoch.

Data source

kleioscan's public REST API, no key required (const API = '/api' in its own bundle). Five endpoints, all cached 60 s at the Pixlet layer:

Endpoint	Used for
/api/networks	the musashi row: live, tip_time
/api/musashi/leios/summary	with_eb_announcement, eb_certified, eb_size_avg
/api/musashi/leios/throughput	avg_tps
/api/musashi/pools/bls-summary	total, active, upcoming
/api/musashi/blocks?limit=1	epoch_no at the tip
/api/musashi/pools/<pool id>	everything on frame B

Blockfrost does not serve musashi — only cardano-mainnet, cardano-preprod, and cardano-preview exist as Blockfrost hosts — so an explorer API is the only off-node option. Nothing here touches the node, which is deliberate: the app keeps working when the producer is down, and can say so.

Quickstart

cd musashi/tidbyt
nix develop                              # this directory's own flake: pixlet, yq, jq, curl
./scripts/check.sh                       # endpoints, liveness, pool, creds
./scripts/preview.sh                     # http://localhost:8080, live reload
./scripts/render.sh                      # writes out.webp
cp config-example.yaml config.yaml       # add Tidbyt creds
./scripts/deploy.sh                      # one-shot push

Dev shell and daemon

This directory is a flake of its own, adapted from the template's — the repository's research shell has no yq and this app does not need the rest of it, so the app's toolchain lives here and the pinned pixlet version lives in exactly one place.

Output	What
devShells.default	pixlet 0.34.0, yq-go, jq, curl, python3
packages.pixlet	the pinned pixlet, with an install check on its version
packages.container	the push-daemon image, config.yaml baked in

For the always-on push, either run ./scripts/push-loop.sh in the foreground, or build the container and let podman keep it alive:

./scripts/build-container.sh             # nix build --impure, then podman load
podman kube play musashistat.yaml        # push every 600 s
podman logs -f musashistat-tidbyt

build-container.sh passes --impure because the flake reads config.yaml from the environment at build time, so the image contains a Tidbyt credential — keep it local, and rebuild after changing credentials or main.star.

When musashi is respun

The app survives a respin, but check two things. kleioscan keeps the id musashi for the live network and moves the old one aside — there is a musashi-old entry today, from the instance that died 2026-09-10 — so NETWORK = "musashi" should still be right, and check.sh says NOT FOUND if it isn't. The pool id is the cold key's hash and does not change, so POOL_ID stands; the pool's registration does not survive, and frame B will read NOBLS until it is re-registered — which is accurate, and arguably the most useful thing the display can tell you that morning. See block-producer.md § 7.

Known limits

• /leios/fragmentation reports available: false on musashi today, so the mempool-fragmentation panel that would be the most interesting frame of all is not available. Worth revisiting.
• The certified share is a whole-chain window figure, not a recent rate, so it moves slowly and will not show a short outage. /leios/throughput-series would support a sparkline of the recent rate.
• kleioscan is a third party. If it stops, the display goes to API ERR while the node is perfectly healthy — the app reports the explorer's view of the network, not the network.