From a classic Learjet to an open flight lab

What was actually built

1. A telemetry console on NASA’s Open MCT. /lab/console runs NASA Ames’ open-source Open MCT framework, version 4.3.1, unmodified, inside a same-origin frame that is created only when the page opens. Open MCT is visualisation software: it draws whatever telemetry a provider hands it. It does not simulate an aircraft and it is not a source of flight data.
2. One authored synthetic dataset. lp-ft-001: six minutes at 1 Hz of a research-style scenario (a climb, a level-off) with explicit units, ISO 8601 UTC timestamps on a fixed synthetic epoch, a schema version, a scenario id, a provenance block and the label Synthetic educational data. It is generated by a seeded script, so two runs are byte-identical. It contains one hidden discrepancy whose explanation lives in a separate truth file.
3. A typed adapter. lab/adapter.js (with adapter.d.ts) owns dataset selection, the shared time cursor, the time window, event selection and Lear assembly ids, validating every id against the dataset and the Aircraft Atlas pack. The console page is glue around it.
4. The Lear assembly view, embedded. The Aircraft Atlas gained an embed=1 mode (no header, rail or footer) and a small postMessage bridge with an explicit schema. Selecting an event frames its window in Open MCT and selects the assembly it concerns; picking a part in the assembly selects the event that concerns it and moves the window. Both directions are exercised by the end-to-end test.
5. Static fallbacks. A chart (SVG) and a table, both written by the same generator, are on the console page before any script runs and are shown automatically if Open MCT cannot be loaded. The authored explanation is in the truth file and, as a spoiler, at the end of this page.
6. A dated research brief and an explainer. The brief reads NASA’s 4 September 2026 X-59 report precisely; the explainer draws shock formation and coalescence as original illustrative artwork.

Which NASA code is used, and which is not

Why the console runs in its own frame

The documented way to start Open MCT is a script tag and openmct.start(element). Tried directly on a page with the site’s header (an experiment kept as a screenshot in the engineering notes), the shell mounts but escapes its container and overlays the page, and Open MCT’s theme stylesheet, written for a page it owns, restyles html, body (measured: 12 px Helvetica), every a (grey), div (position: relative), p, li, button, select and the headings, so the site’s header, footer and authored panel are redrawn in NASA’s theme. Rather than patch a 320 KB stylesheet or the framework, the console gives Open MCT a same-origin page of its own (/lab/telemetry) inside an iframe with an accessible title, exactly the “supported integration boundary” the framework expects; the host page drives it through the public time API. Open MCT owns that frame’s URL hash; the console page uses none.

The dataset, field by field

File: /lab/data/lp-ft-001.json (about 78 KB). Generator: scripts/gen-lab-dataset.mjs, seed 20260901, PRNG mulberry32, Box–Muller noise, rounding per channel, no clock reads. The values are generic research-scenario values chosen to be readable; they are not Learjet 24 or N24SA parameters, limits or procedures, and the file says so in provenance.valuesNote.

Companions: lp-ft-001.truth.json (the authored answer: fault channel, start, rate, explanation, discriminating evidence, helpful observation, and a verdict per event; fetched only when a visitor asks for the reveal and frozen in memory), lp-ft-001.svg (the static chart), the table stamped into the console page between markers, and lear-24xr-components.json (the pack’s component ids, so the adapter can refuse unknown ids without importing the pack).

The evidence arithmetic

For the current window the console reports, per comparison, the mean of observed − expected, the largest absolute difference and when it occurred, and the first sample above the threshold. These come from lab/evidence.js, pure functions pinned by unit tests on a five-sample fixture; the console renders those numbers and nothing else. No model arithmetic is involved anywhere.

The adapter, with an example

Framework-free, DOM-free, and unit-tested in Node. It validates dataset ids against an allow-list, event ids against the dataset, and component ids against both the dataset and the Lear pack; unknown ids throw and emit error, and the state is left untouched.

import { createAdapter } from '/lab/adapter.js';

const adapter = createAdapter({ load: (url) => fetch(url).then((r) => r.json()) });
adapter.on('window', ({ start, end }) => openmct.time.setBounds({ start: Date.parse(start), end: Date.parse(end) }));
adapter.on('component', (id) => assemblyFrame.contentWindow.postMessage({ v: 1, type: 'atlas:select', componentId: id }, location.origin));

await adapter.loadDataset('lp-ft-001');      // fetches, validates (schema + Lear ids), emits dataset/window/cursor
adapter.selectEvent('egt-diverge');          // window 14:02:50Z to 14:06:00Z, cursor 14:03:20Z, component right-exhaust
adapter.selectComponent('outflow-valve');    // → event cabin-settle, its window and cursor
adapter.setCursor('2026-09-01T14:04:46Z');   // nearest sample; readouts follow
adapter.getState();                          // { datasetId, dataset, cursor, window, eventId, componentId, componentIds }
adapter.selectComponent('flux-capacitor');   // throws: Unknown Lear component id

Contract: loadDataset(id), setCursor(tISO), setWindow(startISO, endISO), selectEvent(id), selectComponent(id), eventForComponent(id), getState(), on(type, fn) with events dataset, cursor, window, event, component, error. Types in lab/adapter.d.ts.

The embed protocol

The assembly view is <iframe title="Lear 24XR assembly view" src="/atlas/lear-24xr?embed=1">. Messages are same-origin only, checked on both sides for origin and source window, and parsed strictly (version, type, exact keys, slug ids up to 64 characters); anything else is dropped:

{ v: 1, type: 'atlas:ready' }                                // atlas → console: listening
{ v: 1, type: 'atlas:selected', componentId: string | null } // atlas → console: the visitor picked or cleared a part
{ v: 1, type: 'atlas:select',   componentId: string }        // console → atlas: select this part (unknown ids ignored)

The sync demonstration

The end-to-end test (tests/e2e/lab.spec.ts, Chromium with software WebGL, desktop and phone viewports) asserts the whole chain on the built site: the landing page shows the static chart and the label without any request to /lab/openmct/; the console requests openmct.js and Open MCT’s shell mounts inside the frame; clicking event 3 moves Open MCT’s bounds to 14:02:50 to 14:06:00 (read back from openmct.time.getBounds()) and the assembly frame’s URL gains component=right-exhaust; picking the outflow valve inside the assembly frame selects event 6 and moves the bounds to 14:01:00 to 14:06:00; the label stays visible in every state; blocking the script route shows the static fallback; and the brief, explainer and this page pass an axe audit.

Reproducible commands

git clone <repository> && cd lloydpilot-home && npm ci
node scripts/gen-lab-dataset.mjs            # writes lab/data/* and the static table; run twice, diff nothing
node scripts/gen-lab-dataset.mjs --check    # exit 1 if the committed files differ from the generator
npm run check && npm test                   # syntax/type checks; vitest incl. tests/unit/lab.test.ts
npm run build                               # copies the pinned Open MCT runtime to dist/lab/openmct/4.3.1/
npm run preview                             # http://127.0.0.1:4173/lab/console
./node_modules/.bin/playwright test --project=desktop-1440 tests/e2e/lab.spec.ts
./node_modules/.bin/playwright test --project=phone-390 tests/e2e/lab.spec.ts

What visitors learn

• That “expected” and “observed” are different kinds of number, and that the interesting information is in their difference over time, not in either alone.
• That a discrepancy has to be explained against alternatives: a hotter engine, a wrong indication, a fuel-scheduling difference, a wrong model. Each predicts something different in the other channels.
• That one sensor disagreeing while everything it should move with stays put points at the measurement chain, and that the next useful step is an independent observation rather than a conclusion.
• That a real telemetry tool (Open MCT) is unglamorous and precise: fixed bounds, a time conductor, a plot per quantity, a table for the numbers. The framework does not know or care that the data is synthetic; the labels beside it are what keep the exercise honest.
• Where the channels live on the aircraft, schematically, through the Aircraft Atlas assembly the console points at for each event.

Spoiler: the authored answer

Sources and notices

1. NASA, Open MCT: repository github.com/nasa/openmct; API.md sections “Starting an Open MCT application”, “Domain Objects and Identifiers”, “Telemetry Providers”, “Telemetry Metadata”, “The Time Conductor”, “Time API”; the packaged index.html bootstrap; npm package openmct@4.3.1, license Apache-2.0, published August 2026. Checked 7 September 2026.
2. Apache License, Version 2.0, §4 (redistribution: licence copy, notices) and §6 (trademarks). apache.org
3. OpenVSP licence: NASA Open Source Agreement v1.3, clauses 3.A to 3.F and 4.A (recorded to explain why the tool was not used). github.com/OpenVSP
4. NASA media usage guidelines, “Images and media”, page updated 13 August 2026: NASA insignia not in the public domain; no implied endorsement. nasa.gov
5. The Lear training model and its sources: the Aircraft Atlas provenance drawer at /atlas/lear-24xr.

Software notices (Open MCT’s licence text and copyright line, and every other library) · Open MCT LICENSE.md · Sources / suggest a correction. Educational; nothing here is operating guidance for any airplane. This independent site is not affiliated with or endorsed by NASA or its Administrator.