FROM HERMES WATCH TO THE HERMES CORPUS
How a clip captured by Hermes Watch becomes a case audited by Hermes.
Hermes Watch produces evidence: it points at a livestream or a screen region and saves clips, locally, with the seconds before the trigger. Hermes Intake submits a report for analysis: it takes the witness account, the clip, and the surrounding facts, then runs the rule catalog against satellites, aircraft, weather, and astronomy. Today the bridge between them is manual. V2 will make it one click.
What each tool is responsible for
- Hermes Watch is a Chrome extension. It records what you point it at, on your machine, in IndexedDB. It does not upload anything by itself. Pairing to a Hermes account is wired but disabled in V1.
- Hermes Intake is the 12-step witness form at /intake. It captures who, when, where, what-it-did, how-it-ended, and an optional file at Step 11 of 12. On submission it generates a case ID, runs the rule catalog, and writes a permanent case page at
/case/<id>.
The V1 bridge: manual handoff
If you are a tester who captured something with Hermes Watch and want it analyzed by Hermes, this is the workflow today.
- In Hermes Watch, export the clip. Open the extension, find the case in the cases list, and choose Export. Watch writes the clip to your Downloads folder. The clip is the evidence you will upload.
- Note the capture details before you start the form. Hermes Watch records the source URL or screen region, the trigger time, and a rolling buffer of the seconds before. Write down the date, the time (with timezone), and where on the screen or in the sky the object was — you will retype these into the intake form. V1 does not auto-fill them.
- Open /intake and start a report. The form asks one question per screen. Answer them as if you had not used Watch — describe what you saw in your own words. The form does not know about Watch, and that is fine.
- At Step 11 of 12 ("Video or Photo"), upload the clip. The file picker accepts images and videos. Drag in the file Watch wrote to your Downloads folder. The form will check provenance metadata (camera make/model, timestamp, GPS) and assign a tier (A through D) describing what the file gives Hermes to work with.
- Finish the form and write down the case ID. The success screen shows a permanent URL at
/case/<id>. That is your case page. The rule catalog has already run; the verdict will appear there.
What does not work yet (V2)
- One-click upload from Watch. The extension's pairing UI is visible but the upload button is disabled. There is no API call from Watch to
/api/submitin V1. - Auto-fill of intake fields from Watch metadata. Watch records the trigger time, the source region, and rolling-buffer position; intake does not consume that sidecar. You retype the facts.
- JSON evidence packages. The intake form's Step 11 picker accepts images and videos (jpg, png, heic, mp4, mov, webm, etc., up to 2 GB per file). It does not accept a Watch JSON sidecar. If you have one, keep it on your machine for your own records; Hermes is not reading it yet.
What this means for testers
The corpus expects Watch-captured clips to enter through the same Step 11 upload that any other clip uses. The provenance tier system was built with Watch in mind — clips exported directly from a desktop recording session keep their original metadata and tend to land in tier A or B, which is the highest-value evidence Hermes can analyze. Clips that have been re-encoded, screen-captured a second time, or sent through messaging apps before reaching Step 11 will land in tier C or D, with written reasons. Tier describes the file, not the person.
If you have feedback on the bridge — especially friction points where you had to re-enter something Watch already knew — that is exactly what the V2 work needs. Send it through the intake form's feedback field, or open an issue on the project's tracker.
Posture
Hermes Watch produces evidence. Hermes Intake submits it for analysis. The handoff is manual today and we say so. The pieces are real; the seam is honest about being a seam.