Models Prepare and review private custom models without live training by default. A careful path to try private custom models. Practice mode does not spend real training money. Training Lab / model studio surface at /models. Also called: training / models studioMore details
Developer details
Model studio
Practice turning reviewed learning into a personal model. This path is a local practice lab — not paid training and not a public trainer.
- Review a practice dataset before a practice job can use it.
- Reserve a practice credit before submitting a practice job.
- Practice jobs stay on this machine when the local Lab is on. Real GPU training and billed spend stay off.
More details
Whether the local training Lab fixture path is enabled.
- Disabled is the safe default.
- Enabled only talks to a local loopback Lab — never a public trainer.
More details
Whether the local practice Lab path is turned on. Off is normal for most people.
Developer details
Loopback-gated training Lab surface mode under posture+env double-lock.
Also called: training surface enabled
Whether practice Lab job submit is allowed right now.
- Shows on or off from the double-lock — not from env alone.
- On still means fixture dry-run only — not GPU train or billed spend.
More details
On only when both the product posture lock and the local Lab env lock are on. Off is the safe default.
Developer details
Lab-submit posture ∧ Training Lab env double-lock (LAB-LIVE Wave 1).
Also called: lab submit wire
Network calls for this page are intentionally blocked.
- Private learner data stays on the practice path until gates open.
More details
Off (locked) is the safe default for previews. Unlocking network is an operator choice for live reads.
Developer details
Shown when the surface is inert or gated.
Also called: network disabled
Lab submit: off — needs product authorization and a local Lab. Env alone cannot open Lab submit.
Mode: off. The panels below are the preview path until a local Lab is on with product authorization.
Dataset
Register a fixture dataset for review
Datasets move through registered → pending_review → approved | rejected before any training job may reference them. Rejections show a bounded machine-readable reason code — never free text.
Dataset state: unregistered
Billing (mock)
Reserve a training credit before submitting
Job submission requires an active credit reservation from the mock billing adapter. A missing or expired reservation keeps the submit button disabled in the DOM and is re-checked on the server.
Reservation state: none
Training job
Submit a fixture training job
Enabled only when the dataset is approved AND an active credit reservation exists. The local Lab fake worker runs the fixture job — no real training, no model files, no user content.
Job state: not submitted
- Blocked: code: dataset_not_approved.
- Blocked: code: no_active_reservation.
Provenance
Content-free provenance record
Succeeded jobs emit exactly one content-free provenance record: ids, hashes, and timestamps only. Explicit deletion wipes provenance; TTL expiry retains provenance while removing content.
No completed job yet.
Guarantees
What this page will not do
- No real training spend on this practice path.
- No private learning content sent to a public trainer on this practice path.
- No passwords or secrets shown on this page.
More details
- No MuseHub reads or writes from this preview.
- Only practice credit reservations — not real billing.
- Practice jobs emit fixture metadata only.
- No durable identity storage in the browser.
- Local-only trainer addresses when enabled.