@absolutejs/commerce-machines

Brother PES (v1 truncated writer; reader follows the PEC pointer of any PES version) and Janome JEF. Bare .pec files decode too. VP3 and XXX are recognised as formats but not decoded or written.

convertMachineFile, exportForMachine, MIME_BY_FORMAT, machineChecklist

and a createMachineRegistry factory for the manifest wiring.

The PES writer produces the truncated version-1 layout (header + PEC block,

no CEmbOne/CSewSeg section). Machines and most software sew from the PEC block; a few editors that only read the design section will show it empty.

Thread names come from the Brother PEC and Janome JEF palettes when a file

carries indices; writers match threads by name and otherwise cycle.

raw-tcp — { host, port? } (default 9100) — Server can reach the printer IP (Zebra, most label printers, RIP spoolers).

ipp — { url, username?, password? } — Server can reach ipp://host:631/ipp/print (CUPS, AirPrint, office printers).

printnode — { apiKey, printerId, title? } — Anywhere — PrintNode's paid cloud service relays to its client on a shop PC.

bridge — { bridgeId, action } — Anywhere — the free bridge agent on the shop LAN executes action (see below).

createTransports({ fetch?, bridge?, timeoutMs? }) returns one

MachineTransport per kind (kind, describe, optional probe, send); pass a custom fetch in tests. sendToMachine, probeMachine and describeTarget dispatch on target.transport.

folder sanitises filenames to their base name and refuses paths containing

... raw-tcp opens one connection per file, streams with back-pressure and closes (10 s timeout; probe just connects). ipp encodes an IPP/1.1 Print-Job (application/ipp, document-format = the export's MIME) and returns the printer's job-id; probe is Get-Printer-Attributes. Basic auth is sent when a username is set. printnode posts raw_base64 (ZPL, EPL, TSPL, DST, EXP, PES, JEF…) or pdf_base64 (PDF) to https://api.printnode.com/printjobs; probe fetches /printers/:id.

transportFieldsFor(kind), TRANSPORT_LABELS and transportHelp(kind) feed

a settings form: fields with type: "text" | "password" | "number", a label, and plain-English help on when to use each and what to ask the shop.

raw-tcp-status — push — Zebra unsolicited alerts (SX / alerts.add) over a held-open socket. — Which order is printing; non-Zebra printers.

http-status — push — The RIP or controller POSTs to a webhook the bridge agent serves. — Anything, if the software cannot notify — then it is manual.

snmp-printer — push — SNMP v1/v2c traps and informs (Host Resources + Printer MIB). — Stitches, pieces, job names.

manual — manual — An operator types the time. — Everything — and it says so.

telemetryKindsFor(provider) suggests the paths worth offering for a machine

from its connections and formats. It is a suggestion for the settings screen, never a claim the machine was tested; manual is always included and is always the safe answer. telemetryDelivery(kind) says whether that path is pushed, watched or manual, and telemetryHelp(kind) explains in plain English what it needs, which machines it suits and what it cannot see. telemetryFieldsFor(kind) + TELEMETRY_LABELS build the form.

parseMachineReport(text, parser, { now? }) reads Tajima and Melco

production reports (Label: value and header-row CSV), generic key/value and JSON, through one field vocabulary; an unexpected layout degrades to the generic reader instead of throwing, and returns null when nothing job-shaped is found. referenceFromJobName pulls 288C8286-L1-1 out of a design name.

readingsToRuns(readings, { idleGapSeconds }) collapses a stream into runs:

consecutive running samples make one run, a non-running sample or a gap longer than idleGapSeconds (default 300) closes it, a run is never credited more than one gap past its last running signal, and a non-running reading carrying elapsedSeconds (a finished production report) becomes a run of its own. For push-only machines that can run for hours between signals, pass a larger idleGapSeconds. eventsToRuns does the same over MachineRunEvent[].

The agent opens the socket and sends its bridge token as the first

authenticate frame, then subscribes to bridgeJobs and bridgeTelemetrySources. Both are scoped to its bridgeId by the collections' authorize/match.

A queued job arrives as an insert diff with its files inline

({ filename, mime, bytesBase64 }). The agent runs it and calls the bridge.report mutation; the job leaves the collection.

Run telemetry goes up in small batches through bridge.telemetry

(MachineRunEvent[], stored via BridgeStore.record). bridge.heartbeat carries { version, platform, hostname, capabilities, printers?, telemetry? } once per connection — with a live socket, presence IS the socket.

BridgeStore is the persistence seam: enqueue, claim, complete,

heartbeat, status, and the optional list, pending, readings, record, records. Everything telemetry-related is optional, so an existing store keeps working.

Legacy fallback: createBridgeHandlers(store, { authenticate }) still

returns poll / report / telemetry handlers to mount on POST /bridge/poll, /bridge/report and /bridge/telemetry for shops whose network blocks WebSockets. The poll response carries sources so a fallback agent needs no extra round-trip. It is supported, but it is not the default path and it is the only place anything is polled.