Evidence for 182.C.40.10 (documentation matches deployed behaviour) and
182.C.39.03 (the API guide's captured responses). Four responses from a BFF
this session started, not from a schema.
Method#
apps/oshun/bff/src/server.ts under tsx, on :4010, against the local Postgres
and the staged CPU tools. Secrets were sourced through a throwaway /tmp script
that was deleted at the end; none is in this directory, and the captures were
grepped for key material before committing. The server was torn down and :4010
confirmed free.
Tokens were minted with createDevAuthToken, which is the dev-only
dev.<base64url> form the BFF accepts outside production.
What it corrects#
api-guide.md opened with "Nothing here is mounted" and
creator-workflow.md with "None of this is reachable today". Both were true
when written and stopped being true on 2026-09-11, when the surface was mounted
— and more so today, with five adapters bound. A reader of either document would
have concluded the feature does not run. Both now say what runs and point here.
The boot line#
boot-line.txt, verbatim:
human-video api: 32 endpoints live; bound: preset catalogue; answering 503 until
bound: project drafts, idempotency records, upload sessions (no session table),
identity packs (no pack table), audio takes (no take table), jobs, candidates,
reviews (queue and decisions), releases, erasure, project listing
The three "(no … table)" notes are the finding worth keeping. The tables
exist — human_video_upload_session, human_video_identity_pack and
human_video_audio_take were applied to oshun_dev this session — and the
ports were still unbound, because the committed generated Prisma client did
not carry the models. That is createHumanVideoRuntime treating a stale
client as no database, deliberately: binding a missing delegate would give
undefined.findUnique on the first request, which is a 500 that tells nobody
anything.
Correction, later the same day. This section said the environment's prisma CLI "is 6.19 and cannot regenerate the v7 schema". That was wrong. The root
pnpm exec prismaresolves to nothing;libs/oshun/persistence's ownnode_modules/.bin/prismais 7.8.0 and regenerates in about 1.5 seconds. With the client regenerated all three ports bind, along with drafts, idempotency records and jobs — seven in total. The capture below is what the surface did BEFORE that, and it is kept because the scope measurements in it are still exact. See../2026-09-12-jobs-adapter-live/for the boot line as it stands.
The four captures#
| file | request | status |
|---|---|---|
presets-200.json |
GET /presets, hv:* |
200 — the catalogue, served with no database |
identity-packs-503.json |
GET /identity-packs, hv:* |
503 hv.internal.store_unavailable, detail naming the operation |
presets-403-domain.json |
GET /presets, domain:* |
403 |
boot-line.txt |
— | what a 503 means on this deployment |
The scope boundary, measured#
182.C.40-relevant, because it is the one thing an entitlement change will
touch. Four more requests to the same endpoint, varying only the token:
| scopes | status |
|---|---|
| (none) | 401 |
hv:* |
200 |
hv:projects:* |
200 |
domain:* |
403 |
hv:jobs:* |
403 |
* |
403 |
The wildcard covers the namespace and stops there. domain:* is issued to every
signed-in customer, and letting it reach hv:assets:read would hand all of them
the tenant's identity packs — which is listing the faces a project was built
from. A bare * would be every scope on the platform.
This had only been unit-tested before today; these four requests are the first time it was exercised against a running server.