Show measured gateway latency on the README badge

[kfallah]

Summary

The README badge shows the latest representative gateway p50 request latency from the Experiential-only mock benchmark on main. The value comes from representative_run.gateway.p50_ms in the generated report, not from a difference versus the mock and not from a workflow pass/fail image.

The badge label is gateway latency. The message is formatted like 22.2 ms.

How it updates

1. The latency job writes the full report as gateway-latency.json and the Shields endpoint as shields/gateway-latency.json. Both stay in the uploaded artifact.
2. After a successful push to main on ubuntu-latest, publish-badge publishes gateway-latency.json to the badges branch.
3. Pull requests and workflow_dispatch (including the 32-core runner) still measure and keep the artifact. They do not publish the badge.
4. Publishing does not push to protected main. The badges branch does not trigger gate, latency, or package workflows. GITHUB_TOKEN pushes also do not start new workflow runs.

Seeded public endpoint

The badges branch is live now, seeded from the latest successful main artifact (run 32608210921). That report's representative_run.gateway.p50_ms is 28.365618999998787, which the PR formatter publishes as 28.4 ms. main was not modified.

Verified:

• https://raw.githubusercontent.com/experientiallabs/experiential/badges/gateway-latency.json returns Shields schemaVersion 1 with "message": "28.4 ms"
• https://img.shields.io/endpoint?url=https%3A%2F%2Fraw.githubusercontent.com%2Fexperientiallabs%2Fexperiential%2Fbadges%2Fgateway-latency.json returns an SVG labeled gateway latency: 28.4 ms

README image:

[![gateway latency](https://img.shields.io/endpoint?url=https%3A%2F%2Fraw.githubusercontent.com%2Fexperientiallabs%2Fexperiential%2Fbadges%2Fgateway-latency.json)](https://github.com/experientiallabs/experiential/actions/workflows/gateway-latency.yml?query=branch%3Amain)

Verification

• uv run pytest -q exp/runtime/gateway/latency_badge_test.py exp/runtime/gateway/latency_report_test.py exp/runtime/gateway/latency_measure_test.py
• uv run ruff check . && uv run ruff format --check . && uv run ty check

  

[cursor]

@greptileai Please review this README badge insertion. The latency report itself already merged in #591 and stays Experiential-only.

[greptile-apps]

Greptile Summary

The PR replaces the workflow-status badge with a numeric Shields endpoint showing representative gateway p50 request latency.

• Generates the badge payload from representative_run.gateway.p50_ms.
• Uploads the report and endpoint payload from latency runs.
• Publishes successful main measurements to the badges branch.
• Updates the README and gateway-latency documentation for the numeric badge.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
.github/workflows/gateway-latency.yml	Generates and uploads the nested Shields payload, then publishes successful main-branch measurements to the badges branch.
exp/runtime/gateway/latency_badge.py	Defines the Shields schema and formats representative gateway p50 latency as a one-decimal millisecond value.
exp/runtime/gateway/latency_report.py	Adds optional badge output sourced directly from representative_run.gateway.p50_ms.
exp/runtime/gateway/latency_badge_test.py	Verifies metric selection, endpoint shape, formatting, validation, and README integration.
README.md	Displays the numeric gateway-latency Shields endpoint and links to main-branch workflow runs.
docs/reference/gateway-latency.md	Documents the measured gateway metric, artifact publication flow, stable endpoint, and initial branch-creation behavior.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Push to main] --> B[Latency job]
  B --> C[Generate full latency report]
  B --> D[Generate Shields payload from representative gateway p50]
  C --> E[Upload workflow artifact]
  D --> E
  E --> F[Publish badge job]
  F --> G[badges/gateway-latency.json]
  G --> H[Shields endpoint]
  H --> I[README badge]

Loading

_{Reviews (2): Last reviewed commit: "Show representative gateway p50 on the R..." | Re-trigger Greptile}

[cursor]

@greptileai Please re-review 4438d181. The README badge is now a numeric Shields endpoint for representative p50 gateway-added latency, published only after a successful main benchmark.

[cursor]

@greptileai Please re-review ec5bd88d. The README badge now shows representative gateway p50 request latency from representative_run.gateway.p50_ms, published as badges/gateway-latency.json.