[jonahgabriel]

OMN-19507: dev-202 verify ports reachable from a runner container

Ticket: OMN-19507 (AC3 prerequisite), epic OMN-19500. Authority: operator RULING 2026-09-25T06:57:54Z in the omni_home rolling ledger (line 3238): the dev-202 overlay binds the docker bridge gateway address, not all interfaces and not the LAN, so fleet runners reach it via host.docker.internal.

Stacked on omnibase_infra#4095. This branch carries #4095's commits until it lands; the change here is the single commit df83c4fe8. Once #4095 lands, that commit is rebased onto dev (a clean replay: #4095's squash has the same tree as its branch head).

What changes

• A per-merge verify job runs in a runner container, where localhost is the container itself. It reads a lane at host.docker.internal, the runner's host-gateway alias, which resolves to the default bridge gateway: 172.17.0.1 on .202 (read 2026-09-25 with docker network inspect bridge; daemon.json there sets no host-gateway-ip).

• docker/docker-compose.dev-202.yml: runtime main 61085, effects 61086 and projection API 61002 are published on loopback AND on 172.17.0.1. Loopback stays because the deploy agent's own health probe reads localhost:<port>. Every other port (Postgres, Valkey, Redpanda, consumers, phoenix) stays loopback-only; a verify job reads those through the docker socket. Nothing binds all interfaces or the LAN address.

• deploy/lane-census/lane-manifest.yaml: comment updated to say the same.

• tests/ci/test_dev_202_lane_boundary_omn19505.py: the loopback-only test is replaced by three tests: (1) every port entry names a host address; (2) every port binds only loopback or the gateway, and every port is on loopback; (3) exactly {61085, 61086, 61002} are on the gateway.

• RED first: with the new tests and the old overlay, test (3) failed (1 failed, 15 passed); with the overlay change all 16 pass. Focused runs: the boundary file plus lane-census manifest parity, sim-202 boundary and host-scope tests, all passing.

• Rendered the compose triple (infra + dev-lane + dev-202, --profile runtime) locally with stub env values: 18 published entries across 28 services; the only non-loopback entries are the three gateway binds above.

• Mechanism measured on .202 with throwaway containers on port 61999/61998 (removed after; 0 containers, 0 networks, 0 listeners left): a container on a user-defined network carrying host.docker.internal:host-gateway got PONG from a gateway-published port, got connection refused from a loopback-only one (reproducing the open finding), and the host's LAN address refused the port.

Not in this PR

• The .202 verify runner registration and the (dev, omnimarket) to dev-202 route row (OMN-19510) come with the bring-up.
• The comment in feat(OMN-19507): the per-merge verify jobs read the lane the deploy agent routed the merge to #4105's config/deploy_lane_routing.yaml that lists this as an open choice is updated in the route-row PR.
• Runtime-affecting: the file only affects a lane that is not yet running; no .201 lane reads it.

Evidence-Ticket: OMN-19507
Evidence-Source: OCC#11238

[github-actions]

⚠️ Hostile Reviewer — DEGRADED (informational)

Critical findings: 0
Major findings: 1
Total findings: 6
Models succeeded: glm-review

Note: Fewer than 2 reviewer models succeeded. Degraded results are informational (OMN-8468/OMN-8524) and do not block merge. Error: cli_review exit 2 (fewer than 2 models succeeded — partial/total outage)

---

Semantics (OMN-17492 — the model finds, thread resolution gates)

Surface	Meaning	Blocks merge?
Review threads	Per-finding, posted by the reviewer	No (informational)
Hostile Review Thread Gate	Deterministic: unresolved hostile-reviewer threads exist	Fails until resolved (not yet a required context)
degraded verdict	Fewer than 2 models succeeded (infra)	No

Powered by omniintelligence.review_pairing.cli_review — multi-model adversarial review: qwen3-review, qwen3-review-b, glm-review (OMN-8468/OMN-8524/OMN-17492)

[github-actions]

Hostile Reviewer — adversarial findings (OMN-17492)

Models succeeded: glm-review
Models failed: codex
New finding threads: 0
Deduped (already posted on this PR): 0
Below quorum (one model only, reported not threaded): 5
Nit-level findings suppressed: 1

The model is the FINDER, never the gate: merge is gated only by the
deterministic Hostile Review Thread Gate, which blocks while
hostile-reviewer threads are unresolved. Resolve each thread after
addressing (or rejecting, with a reply) its finding.

Below quorum: 5 finding(s) raised by one model only (OMN-18479)

These are reported and NOT dropped, but they get no thread and do not block: a single model's finding no other model reproduced is not evidence enough to stop a merge. Read them; act on them if they are right.

• [MAJOR] docker/docker-compose.dev-202.yml, VERIFY-RUNNER REACH comment block (glm-review) — Gateway bind exposes ports to every container on the host, not just the verify runner | 172.17.0.1 is the default bridge gateway, but it is reachable from any container on any Docker network on .202 t
• [MINOR] docker/docker-compose.dev-202.yml; tests/ci/test_dev_202_lane_boundary_omn19505.py (glm-review) — Hardcoded 172.17.0.1 is an environment snapshot, not a contract | The address is correct only while the daemon defines no host-gateway-ip, the default address pool still allocates 172.17.0.0/16 to the
• [MINOR] tests/ci/test_dev_202_lane_boundary_omn19505.py, _bindings and test_dev_202_lane_boundary_every_port_entry_names_a_host_address (glm-review) — Port-entry regex and split parsing admit malformed entries silently | The cross-check regex ([^"]*:\d+) matches any colon-suffixed string, and _bindings calls line.split(":") with fixed unpackin
• [MINOR] tests/ci/test_dev_202_lane_boundary_omn19505.py (glm-review) — GATEWAY_REACHED_PORTS duplicates deploy_lane_routing.yaml with no consistency check | The set {61085, 61086, 61002} is transcribed from config/deploy_lane_routing.yaml into a hardcoded test constant.
• [MINOR] tests/ci/test_dev_202_lane_boundary_omn19505.py (glm-review) — Boundary tests validate compose text only, never the listening sockets | All three port tests assert on the YAML source of the overlay. Docker Compose can render a !override differently than expecte

[jonahgabriel]

Lab proof on .202 at this head, taken before landing (rule 24), under the omni_home ledger HOLD of lane dev-202-bringup for surface 202-compose-dev:

• The dev-202 lane was brought up on .202 from dev dfb949d: a workspace build of 64 s, then compose up. With dev's overlay, which binds loopback only, omnipc2-verify-runner-1 got 000 from http://host.docker.internal:61085/health, :61086/health and :61002/projections. That reproduces the open finding.
• The .202 clone was then checked out at this head, 0d55bc2. Only omninode-runtime, runtime-effects and projection-api were recreated, with up -d --no-deps. ss -ltn then listed each of 61085, 61086 and 61002 on both 127.0.0.1 and 172.17.0.1, and on nothing else.
• From inside the verify runner container, as the runner user, after the runtime turned healthy (restarts 0): 61085 answered 200, 61086 answered 200 and 61002 /projections answered 200.
• From the Mac on the LAN, 192.168.86.202 ports 61085, 61002 and 61436 were closed. The positive control, llama.cpp on :8000, was open.