Skip to content

Commit 043e654

Browse files
committed
docs(routing): sync routing reference to RI-05 execution
Profiles now execute on explicit policy/<id> or alias requests; note the live-path evidence surface (tools/image), the capability penalize no-op until a score dimension ships (RI-06+), and the full CLI flag set.
1 parent 6ad6a7c commit 043e654

1 file changed

Lines changed: 19 additions & 12 deletions

File tree

  • docs-site/src/content/docs/reference/configuration

‎docs-site/src/content/docs/reference/configuration/routing.md‎

Lines changed: 19 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -87,10 +87,10 @@ commands, see [Combos](/guides/combos/).
8787

8888
Routing policy profiles are the Router Intelligence selection layer: an explicitly requested
8989
`policy/<id>` (or configured alias) selects among a fixed candidate allowlist using hard capability
90-
requirements and deterministic, explainable scoring. In this release profiles are configuration and
91-
dry-run evaluation only: production requests are not yet routed through them (execution wiring
92-
arrives with RI-05), and existing model ids are **never** routed through a profile implicitly.
93-
Policy ids do not participate in the model resolution order above until execution lands.
90+
requirements and deterministic, explainable scoring. An explicit `policy/<id>` request (or a
91+
configured alias) executes the evaluator and routes the selected candidate. Existing model ids are
92+
**never** routed through a profile implicitly: the `policy/` namespace and profile aliases are the
93+
only entry points, and both are validated against the model resolution order above.
9494

9595
Each key is an id matching `[A-Za-z0-9][A-Za-z0-9._-]{0,63}`, always addressable as `policy/<id>`,
9696
with one optional `alias`. Aliases must be unique and cannot collide with configured providers,
@@ -110,9 +110,15 @@ namespace, or reserved bare native families (`gpt-*`, `o1-*`, `o3-*`, `o4-*`, `c
110110
`structuredOutput`, `localOnly`, `remoteAllowed`, `encryptedCodexTasks`; plus `reasoningEffort` and
111111
`serviceTier` strings.
112112

113-
Request evidence supplied to a dry-run (context window, tools, image input, structured output,
114-
reasoning effort, service tier, encrypted Codex tasks) is evaluated against candidate capabilities
115-
together with the profile `require` block; a candidate must satisfy both to be eligible.
113+
For `unknownEvidence.capability`, `penalize` currently behaves like `allow`: scoring has only a
114+
configured-priority component until a capability score dimension ships (planned with RI-06+), so
115+
`penalize` cannot yet change the selected candidate.
116+
117+
Request evidence is evaluated against candidate capabilities together with the profile `require`
118+
block; a candidate must satisfy both to be eligible. On the live request path the proxy derives
119+
tools and image-input evidence from the request body; context-window size and the remaining
120+
evidence dimensions stay unknown at routing time. Use the dry-run API/CLI to inspect the full
121+
evidence surface for context-sensitive profiles.
116122

117123
The CLI dry-run accepts request-evidence flags but cannot supply candidate capability evidence yet;
118124
candidate evidence is provided through the API (`POST /api/routing-profiles/dry-run`).
@@ -140,9 +146,9 @@ candidate evidence is provided through the API (`POST /api/routing-profiles/dry-
140146
}
141147
```
142148

143-
CLI: `ocx route policy list`, `ocx route policy show <id>`, and
144-
`ocx route policy dry-run <id> --model-context <tokens> --tools`. Dry-run evaluates candidates
145-
without sending any upstream request.
149+
CLI: `ocx route policy list [--json]`, `ocx route policy show <id> [--json]`, and
150+
`ocx route policy dry-run <id> [--model-context <tokens>] [--tools] [--image] [--structured-output] [--json]`.
151+
Dry-run evaluates candidates without sending any upstream request.
146152

147153
### Combos vs policy profiles
148154

@@ -152,8 +158,9 @@ without sending any upstream request.
152158
requirements filter first, then deterministic scoring ranks the survivors.
153159

154160
Both are virtual namespaces with aliases and collision validation; they differ in *how* a candidate
155-
is chosen. Profile scoring will expand with capability (RI-05), health (RI-06), quota (RI-07), and
156-
cost (RI-08) dimensions; per-request trace recording arrives with execution (RI-05).
161+
is chosen. Profile scoring currently uses the configured-priority component only; health (RI-06),
162+
quota (RI-07), and cost (RI-08) score dimensions are planned. Per-request route-decision traces are
163+
recorded when a policy profile executes.
157164

158165
### Catalog eligibility
159166

0 commit comments

Comments
 (0)