Skip to content

police sweep 2026-05-31: pagination cursor field + synthesize lifecycle #3

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
onebit0fme wants to merge 1 commit into from
Closed

police sweep 2026-05-31: pagination cursor field + synthesize lifecycle #3

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion api-reference/introduction.mdx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,7 +71,7 @@ A URL in `content` is fetched and its extracted text is scored. Set `Accept: tex
| Base URL | `https://u22a8.ai/v1` |
| Auth | `Authorization: Bearer <key>` on every endpoint |
| Format | JSON request and response; `score` also accepts a raw-text body, and `score`/`compare` can return a `text/plain` view |
| Pagination | Cursor-based; list responses return a `next` cursor |
| Pagination | Cursor-based; list responses return a `next_cursor` field |
| Errors | One error envelope; every response carries a request id |

## Failure modes
Expand Down
5 changes: 3 additions & 2 deletions concepts/model-lifecycle.mdx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,7 @@ A model has a **lifecycle**: a small set of states it moves through from first d
stateDiagram-v2
direction LR
[*] --> draft: create
draft --> busy: train / discover
draft --> busy: train / discover / synthesize
busy --> draft: discovery completes
busy --> ready: training completes
busy --> failed: run fails
Expand All @@ -25,7 +25,7 @@ stateDiagram-v2
| State | Meaning |
|-------|---------|
| `draft` | Created and accepting trait and sample edits. Not yet scorable. |
| `busy` | An asynchronous operation is running — training, discovery, or a background retrain. The model reports which one. |
| `busy` | An asynchronous operation is running — training, discovery, synthesis, or a background retrain. The model reports which one. |
| `ready` | Trained and serving scores against an active [version](/concepts/versions). |
| `failed` | The last run did not complete. Correct the supervision and train again. |
| `archived` | Stopped serving. Samples and version history are retained — archiving is not permanent deletion. |
Expand All @@ -38,6 +38,7 @@ One pattern drives the states: **create → supervise → train → score**, wit
- **Trait and sample edits are draft-only.** Adding, changing, or removing traits and explicit samples applies only while the model is a `draft`. Once trained, those edits are no longer accepted — a different standard means a new model, not an edit to a serving one. ([Feedback](/concepts/supervision) still appends samples to a serving model; it adds to the next retrain rather than editing the trained definition.)
- **Train** moves a draft to `busy`, then to `ready` on success or `failed` on error. This first run is a [cold start](/concepts/training): the model becomes scorable from whatever supervision it has.
- **[Discovery](/concepts/discovery)** runs on a draft, attaches intrinsic traits, and returns the model to `draft` — ready to train.
- **[Synthesis](/concepts/samples)** runs on a draft, generates samples, and returns the model to `draft`.
- **Score** serves the active version of a `ready` model. Scoring stays available throughout a background retrain: the model keeps serving its current version until the new one swaps in atomically, so a request never waits on a run and never sees a half-trained snapshot.
- **[Feedback](/concepts/supervision)** on a score request records labels for the scored content as new samples, accumulating for the next retrain — **online learning** that does not interrupt scoring.

Expand Down
Loading