The OSINT investigator's workbench. One case, one folder, every tool. Local-first: your media and your investigations never leave your machine.
Fixtures
Sheet fixtures
The pathologies of three real investigation binders, distilled into files a test can read. Synthetic and anonymous on purpose: the binders themselves are case material, so no name, no unit, no place and no URL from them appears here — what is kept is the shape of the mess, which is the only part a parser cares about.
| File | What it is there to break |
|---|---|
binder-semicolon.csv |
A European export: semicolon-separated, comma decimals, a quoted cell holding a comma, dd/MM/yyyy dates, #REF! where a formula died. |
binder-worklist.csv |
A geolocation worklist with no status column at all — its progress is the fill rate of Coordinates, and its cells hold three coordinate formats plus To be found. |
binder-timeline.csv |
An event timeline: a bare hh:mm local time with no date, an hour note in prose, a multi-value equipment cell with a quantity, and mostly empty rows. |
binder-states.csv |
The eight words a real binder used for one status column, so a state vocabulary is tested against words nobody would invent. |
Any test asserting "the app reads the real thing" belongs here rather than against a table written to suit the code.
Map sites
| File | What it is there to break |
|---|---|
map-sites.raw.json |
The observation: every URL nine map sites wrote about themselves after a known gesture, with the window they were driven in. Nothing in it has been parsed. |
map-sites.json |
The same run, parsed by engine/mapsites.py and solved for where each site draws its camera. Apple's stale z, Google's satellite height in metres, Yandex's longitude-first order, and the five sites that draw their centre somewhere other than the middle of the window. |
Read from both sides: tests/test_mapsites.py checks the parse still returns those
views, frontend/src/lib/extensionMapFrame.test.js that the extension's arithmetic
puts them back where the browser had them. docs/MAP_SITES.md is the prose version.
Neither file is written by hand. npm run calibrate:maps drives the sites again and
rewrites the raw one; python scripts/build_map_fixture.py turns that into the other.
The protocol both follow is frontend/calibration/protocol.mjs, and it is itself tested
against a map whose layout is known (frontend/e2e/map-calibration.spec.js).
The Case workspace
| File | What it is there to break |
|---|---|
case-workspace-0.3.1.json |
How the Board, Graph and Timeline read a case left the way a 0.3.1 analyst leaves one: saved readings of each surface, Live and Snapshot, in both Timeline modes, with Media tracks, hides and pins; dated, undated and unplaced Claims; a kept Detect pin and a proof date; Notebook mentions, a promoted Sheet row and galleries. Ids are written type:label and the moment of the build is scrubbed. |
Written by tests/caseworkspace.py, never by hand, and compared by
tests/test_case_workspace_fixture.py; frontend/src/lib/caseWorkspaceFixture.test.js
reopens its saved views through the parsers the three surfaces use. A change that
means to alter a reading rewrites it with AZIMUT_WRITE_GOLDEN=1 uv run pytest tests/test_case_workspace_fixture.py, and the diff is that change's account of what
reads differently.
Search folding
| File | What it is there to break |
|---|---|
fold_cases.json |
What a search folds and what it keeps, one line per case: Latin, Greek and Cyrillic accents, Arabic harakat and the tatweel, Hebrew points go; a Devanagari sign, a dakuten, the Hebrew maqaf and the punctuation of an email, a network or a handle stay. |
Written by hand. Read by tests/test_textfold.py against engine/textfold.py and by
frontend/src/lib/textFold.test.js against lib/textFold.js, so the server's index
and the browser's in-memory search cannot fold one text two ways.