~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.
Python academiaanthropicarxivbravedeep-researchencryptionhome-automationhomeserverlocallocal-deep-researchlocal-llmmistralollamaopenaipubmedresearchresearch-toolretrieval-augmented-generationsearxngself-hosted
/**
* Tests for components/annotation_surface.js — applyWhenReady re-applies
* highlights across the results page's asynchronous render.
*
* The results page (results.js) first swaps results.html's
* `.ldr-loading-spinner` for its OWN placeholder
* (`<i class="fas fa-spinner fa-pulse"> Loading research results…`) and
* only THEN, after the report fetch, replaces the container's innerHTML
* with the real report. A one-shot readiness check keyed on
* `.ldr-loading-spinner` would see the placeholder as "ready", anchor
* nothing, and never re-run — so highlights would silently never appear
* on the first view of a research report. This drives the real
* LDRAnnotationSurface.init() through that exact sequence.
*/
let SafeLoggerErrors;
beforeAll(() => {
window.__VITEST_TEST__ = true;
globalThis.SafeLogger = {
error: (...a) => SafeLoggerErrors.push(a),
warn: () => {},
info: () => {},
debug: () => {},
};
// The surface now calls safeFetchWithAuth (auth-aware wrapper added on
// main); delegate to the fetch mock each test installs.
globalThis.safeFetchWithAuth = (...args) =>
(globalThis.safeFetch || globalThis.fetch)(...args);
});
afterAll(() => {
delete window.__VITEST_TEST__;
});
beforeEach(async () => {
SafeLoggerErrors = [];
document.body.innerHTML = '';
// Fresh module each test so init()'s document-level listeners don't stack.
vi.resetModules();
await import('@js/components/annotation_surface.js');
});
function mockAnnotationsFetch(annotations) {
globalThis.fetch = vi.fn(() =>
Promise.resolve({
ok: true,
json: () => Promise.resolve({ success: true, annotations }),
})
);
}
function flush() {
return new Promise((r) => setTimeout(r, 0));
}
describe('applyWhenReady across the placeholder → content swap', () => {
it('anchors the highlight only after the real report replaces the placeholder', async () => {
// Container starts with results.js's loading placeholder (NOT the
// .ldr-loading-spinner the naive heuristic looked for).
const container = document.createElement('div');
container.id = 'results-content';
// eslint-disable-next-line no-unsanitized/property -- test fixture
container.innerHTML =
'<div class="text-center my-5"><i class="fas fa-spinner fa-pulse"></i>' +
'<p class="mt-3">Loading research results...</p></div>';
document.body.appendChild(container);
mockAnnotationsFetch([
{ note_id: 'n1', quote: 'attention mechanism', prefix: '', suffix: '' },
]);
const base = '/notes/api/research/res-1';
window.LDRAnnotationSurface.init({
containerId: 'results-content',
endpoints: {
list: `${base}/annotations`,
create: `${base}/annotations`,
deleteFor: (id) => `${base}/annotations/${id}`,
},
});
await flush(); // annotations fetch resolves
// Placeholder is still showing → nothing anchored yet, and we must
// NOT have given up (no "could not be anchored" error logged).
expect(container.querySelectorAll('mark.ldr-research-annotation').length).toBe(0);
expect(
SafeLoggerErrors.some((e) => String(e[0]).includes('could not be anchored'))
).toBe(false);
// results.js now renders the real report (wholesale innerHTML swap).
// eslint-disable-next-line no-unsanitized/property -- test fixture
container.innerHTML =
'<h2>Findings</h2><p>We propose a new attention mechanism for this.</p>';
await flush(); // MutationObserver fires → re-apply
const marks = container.querySelectorAll('mark.ldr-research-annotation');
expect(marks.length).toBe(1);
expect(marks[0].textContent).toBe('attention mechanism');
});
it('applies highlights to a real report that merely begins with the word "Loading"', async () => {
const container = document.createElement('div');
container.id = 'results-content';
// A genuine (long) report whose text happens to START with "Loading".
// Pre-fix, looksLikeLoading() treated ANY leading "Loading"/"Searching"
// as a placeholder and suppressed every highlight on the page forever;
// the fix only treats a SHORT (<120 char) such body as a placeholder.
// eslint-disable-next-line no-unsanitized/property -- test fixture
container.innerHTML =
'<p>Loading historical context first: the paper introduces a novel ' +
'attention mechanism that substantially reduces the compute cost of ' +
'long-context inference across a wide range of downstream tasks.</p>';
document.body.appendChild(container);
mockAnnotationsFetch([
{ note_id: 'n1', quote: 'attention mechanism', prefix: '', suffix: '' },
]);
const base = '/notes/api/research/res-1';
window.LDRAnnotationSurface.init({
containerId: 'results-content',
endpoints: {
list: `${base}/annotations`,
create: `${base}/annotations`,
deleteFor: (id) => `${base}/annotations/${id}`,
},
});
await flush();
const marks = container.querySelectorAll('mark.ldr-research-annotation');
expect(marks.length).toBe(1);
expect(marks[0].textContent).toBe('attention mechanism');
});
it('keeps re-applying until every DISTINCT annotation anchors, not just until the mark count is reached', async () => {
const container = document.createElement('div');
container.id = 'results-content';
// Annotation A's quote spans a <br> → TWO marks for ONE annotation.
// Annotation B's quote is absent from this first render.
// eslint-disable-next-line no-unsanitized/property -- test fixture
container.innerHTML =
'<p>reduces cost substantially, but<br>shared ownership must be tracked</p>';
document.body.appendChild(container);
mockAnnotationsFetch([
{ note_id: 'A', quote: 'substantially, but shared ownership', prefix: '', suffix: '' },
{ note_id: 'B', quote: 'governance model', prefix: '', suffix: '' },
]);
const base = '/notes/api/research/res-1';
window.LDRAnnotationSurface.init({
containerId: 'results-content',
endpoints: {
list: `${base}/annotations`,
create: `${base}/annotations`,
deleteFor: (id) => `${base}/annotations/${id}`,
},
});
await flush();
// A anchored as 2 marks. Pre-fix, a mark COUNT of 2 >= total 2 declared
// the set complete and tore down the observer — even though B had not
// anchored — so B's later content would never be picked up.
// B's content now arrives (wholesale re-render, as results.js does).
// eslint-disable-next-line no-unsanitized/property -- test fixture
container.innerHTML =
'<p>reduces cost substantially, but<br>shared ownership must be tracked</p>' +
'<p>The governance model is shared.</p>';
await flush();
// With the distinct-annotation completion metric the observer was still
// active and re-anchored BOTH. Pre-fix it had already stopped, so
// neither the swapped-away A nor the newly-present B is anchored.
const ids = new Set(
[...container.querySelectorAll('mark.ldr-research-annotation')].map(
(m) => m.dataset.noteId
)
);
expect(ids.has('B')).toBe(true);
expect(ids.has('A')).toBe(true);
});
});