feat(OMN-7404): shadow-mode routing confidence gate in node_model_router_compute by jonahgabriel · Pull Request #2376 · OmniNode-ai/omnibase_infra

jonahgabriel · 2026-07-21T03:12:48Z

Summary

OMN-7404 — "Task 8: [Phase 2] Deploy routing classifier gate in node_model_router_compute".

Adds RoutingGate as an audit-only, shadow-mode confidence gate, injectable
into HandlerScoreModels (the canonical def-B handler for node_model_router_compute,
OMN-14825). The gate never selects a model or alters the routing decision — it only
computes {confidence, would_flag} for logging, per the plan's own reviewed
"Clarification from review" correction (docs/plans/2026-04-03-learning-infrastructure.md,
Task 8): "The classifier predicts 'is this routing decision likely good?' — a binary
confidence estimator, NOT a model selector... Shadow-mode only in Phase 2."

Scope deviations from the ticket text (and why)

The Linear ticket description (and an earlier, superseded draft of the plan) describes
RoutingGate.recommend() with an override-capable use_classifier flag, wiring into
handler_model_router.py, and a Path(...).exists() check inside the handler. None of
that is implemented as literally written, for concrete reasons:

handler_model_router.py no longer exists. OMN-14825 (merged, PR feat(OMN-14825): regenerate node_model_router_compute to canonical def-B #2360) regenerated
node_model_router_compute to the canonical def-B shape; the current handler is
handler_score_models.py::HandlerScoreModels.handle().
The plan itself was corrected after the ticket was drafted. The "Clarification from
review" block explicitly rules out an override-capable recommend() — audit only,
shadow-mode only. This PR implements the corrected audit() design, not the stale
recommend() draft.
No classifier exists. Plan Task 7 (train + persist RoutingClassifier) was never
built — no pickle artifact, no pandas/sklearn dependency anywhere in this repo. gate=None
is therefore the only path ever exercised in production today, which is why default
behavior must reproduce prior output byte-for-byte (see golden-equivalence proof below).
No file-existence check inside the handler. node_model_router_compute is a
COMPUTE_GENERIC node — CLAUDE.md §7a requires it remain I/O-free and deterministic.
Checking _CLASSIFIER_PATH.exists() inside handle() would violate that. Instead, the
gate is an optional constructor-injected dependency (HandlerScoreModels(gate=...));
loading any real classifier artifact is pushed to whichever caller constructs the gate
(a DI/registry seam), never to the handler.

What changed

src/omnibase_infra/learning/routing/gate.py — RoutingGate.audit(), pure/dependency-free
(no pandas), graceful degradation on classifier=None or classifier exception.
src/omnibase_infra/learning/routing/typed_dict_routing_audit.py — TypedDictRoutingAudit
(kept the union-usage ratchet at 154/154 instead of a dict[str, float | bool | None]
3-way-union value type).
handler_score_models.py — optional gate: RoutingGate | None = None constructor param;
when present, logs a routing_audit: ... line after computing the decision, never mutating it.
contract.yaml — patch bump (1.0.2 → 1.0.3) documenting the addition.
scripts/ci/test_selection_adjacency.yaml + tests/unit/contracts/test_protocol_ownership.py —
registered the new learning module and ProtocolRoutingClassifier in their respective
fail-closed allowlists (governed pre-push selector OMN-13973, protocol-ownership gate INFRA-016).

RED → GREEN evidence

tests/unit/learning/test_routing_gate.py — fails at collection (ImportError) before
gate.py exists; 6/6 pass after.
tests/unit/nodes/test_model_router/test_score_models_routing_gate_omn7404.py — verified RED
via git stash of the handler edit (TypeError: HandlerScoreModels() takes no arguments on
4/5 gate-dependent tests); 5/5 pass after restoring the wiring.
tests/unit/nodes/test_model_router/test_score_models_defb_omn14825.py (pre-existing
golden-equivalence corpus, untouched) stays green — proves gate=None (today's only
production path) is a byte-for-byte no-op.

Test plan

uv run pytest tests/unit/learning/ tests/unit/nodes/test_model_router/ tests/unit/contracts/test_protocol_ownership.py tests/unit/scripts/ci/test_test_selection_loader.py -v — 55/55 pass
uv run ruff format / uv run ruff check --fix clean
uv run mypy src/omnibase_infra/learning src/omnibase_infra/nodes/node_model_router_compute/handlers/handler_score_models.py clean
pre-commit run --all-files (targeted) clean
Full local pre-push suite (governed selector escalated to full suite): uv run pytest tests/ --ignore=tests/integration — 21844 passed, 36 skipped
gh pr checks green (watching)

dod_evidence: RED→GREEN proof above; ticket OMN-7404 cited; PR body documents the corrected/authoritative plan source superseding the stale ticket draft.

Evidence-Ticket: OMN-7404
Evidence-Source: OCC#4593
Evidence-Commit: c7a371a9428394710186c03519f43f522a0b36fd

…ter_compute Adds RoutingGate (audit-only, per the plan's reviewed "Clarification from review" correction: a confidence estimator, never a model selector) as an optional injected dependency on HandlerScoreModels. Defaults to gate=None, which reproduces prior behavior byte-for-byte (proven by the existing OMN-14825 golden-equivalence corpus). No classifier has ever been trained (plan Task 7 unbuilt), so this is the only path exercised in production today. Deviates from the ticket's literal `_CLASSIFIER_PATH.exists()` file-check inside the handler: that pattern is I/O inside a COMPUTE node (forbidden by CLAUDE.md 7a). Classifier loading is pushed to whichever caller constructs the gate (a DI/registry seam), never into the handler itself. The literal `handler_model_router.py` target file no longer exists post OMN-14825 canonical def-B regeneration; this lands on its replacement, handler_score_models.py. RED->GREEN: tests/unit/learning/test_routing_gate.py fails at import before gate.py exists; tests/unit/nodes/test_model_router/test_score_models_routing_gate_omn7404.py fails (TypeError: unexpected keyword argument 'gate') on the pre-wiring handler. Both pass after implementation, and the pre-existing golden equivalence test (test_score_models_defb_omn14825.py) remains green unmodified, proving gate=None is a true no-op.

…n gates Governed pre-push selector (OMN-13973) and protocol-ownership allowlist (INFRA-016) both fail-closed on unregistered new modules/protocols; this satisfies both for the new src/omnibase_infra/learning package added in the prior commit.

coderabbitai · 2026-07-21T03:13:00Z

Warning

Review limit reached

You’ve reached a temporary PR review limit under our Fair Usage Limits Policy.

Your recent review volume is higher than typical usage, so adaptive limits are currently applied.

Next review available in: 30 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 1e6bca14-5241-445b-8e14-17e982463049

📥 Commits

Reviewing files that changed from the base of the PR and between 43935c8 and ce4e601.

📒 Files selected for processing (11)

scripts/ci/test_selection_adjacency.yaml
src/omnibase_infra/learning/__init__.py
src/omnibase_infra/learning/routing/__init__.py
src/omnibase_infra/learning/routing/gate.py
src/omnibase_infra/learning/routing/typed_dict_routing_audit.py
src/omnibase_infra/nodes/node_model_router_compute/contract.yaml
src/omnibase_infra/nodes/node_model_router_compute/handlers/handler_score_models.py
tests/unit/contracts/test_protocol_ownership.py
tests/unit/learning/__init__.py
tests/unit/learning/test_routing_gate.py
tests/unit/nodes/test_model_router/test_score_models_routing_gate_omn7404.py

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests
Commit unit tests in branch jonah/omn-7404-routing-classifier-gate-phase2

_{Comment @coderabbitai help to get the list of available commands.}

Jonah Gray added 2 commits July 20, 2026 19:25

jonahgabriel mentioned this pull request Jul 22, 2026

evidence(OMN-7404): bind omnibase_infra 2376 OCC companion OmniNode-ai/onex_change_control#4589

Merged

jonahgabriel merged commit 4380e8b into dev Jul 22, 2026
252 of 312 checks passed

jonahgabriel deleted the jonah/omn-7404-routing-classifier-gate-phase2 branch July 22, 2026 04:51

Navigation Menu

Repository navigation

feat(OMN-7404): shadow-mode routing confidence gate in node_model_router_compute - #2376

feat(OMN-7404): shadow-mode routing confidence gate in node_model_router_compute#2376
jonahgabriel merged 2 commits into
devfrom
jonah/omn-7404-routing-classifier-gate-phase2

jonahgabriel commented Jul 21, 2026 •

edited

Loading

Uh oh!

coderabbitai Bot commented Jul 21, 2026

Review limit reached

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

jonahgabriel commented Jul 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Scope deviations from the ticket text (and why)

What changed

RED → GREEN evidence

Test plan

Uh oh!

coderabbitai Bot commented Jul 21, 2026

Review limit reached

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

jonahgabriel commented Jul 21, 2026 •

edited

Loading