PaperClipAI/.github/workflows/runner-live-evals.yml

name: Runner Live Evals

on:
  schedule:
    - cron: "17 6 * * 0"
  workflow_dispatch:
    inputs:
      candidate:
        description: "Comma-separated live candidate IDs (for example codex-luna)"
        type: string
        required: false
      case:
        description: "Comma-separated workflow case IDs"
        type: string
        required: false
      limit:
        description: "Maximum executions after candidate/case filtering"
        type: string
        required: false

concurrency:
  group: runner-live-evals-${{ github.ref }}
  cancel-in-progress: true

jobs:
  authorize:
    name: Authorize paid campaign
    if: github.event_name != 'schedule' || vars.RUNNER_LIVE_EVALS_NIGHTLY_ENABLED == 'true'
    runs-on: ubuntu-latest
    timeout-minutes: 5
    permissions:
      contents: read
    outputs:
      eval_runner: ${{ steps.runner.outputs.runner }}
    steps:
      - name: Require default branch and allowlisted numeric actor IDs
        env:
          GH_TOKEN: ${{ github.token }}
          REF: ${{ github.ref }}
          DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
          ACTOR: ${{ github.actor }}
          ACTOR_ID: ${{ github.actor_id }}
          TRIGGERING_ACTOR: ${{ github.triggering_actor }}
          ALLOWED_ACTOR_IDS: ${{ vars.RUNNER_E2E_ALLOWED_ACTOR_IDS }}
        run: |
          set -euo pipefail
          if [ "$REF" != "refs/heads/$DEFAULT_BRANCH" ]; then
            echo "Paid runner live evals may run only from the default branch." >&2
            exit 1
          fi
          if ! jq -e 'type == "array" and length > 0 and all(.[]; type == "number" and . > 0 and floor == .)' <<< "${ALLOWED_ACTOR_IDS:-}" >/dev/null; then
            echo "RUNNER_E2E_ALLOWED_ACTOR_IDS must be a non-empty JSON array of numeric GitHub user IDs." >&2
            exit 1
          fi
          triggering_actor_id="$(gh api "users/$TRIGGERING_ACTOR" --jq .id)"
          if [ "$triggering_actor_id" != "$ACTOR_ID" ] && [ "$TRIGGERING_ACTOR" = "$ACTOR" ]; then
            echo "GitHub actor identity contexts disagree; refusing the paid run." >&2
            exit 1
          fi
          candidates=("$triggering_actor_id" "$ACTOR_ID")
          for candidate in "${candidates[@]}"; do
            if ! jq -e --argjson candidate "$candidate" 'index($candidate) != null' <<< "$ALLOWED_ACTOR_IDS" >/dev/null; then
              echo "The initiating GitHub account is not authorized to run paid runner live evals." >&2
              exit 1
            fi
          done

      - name: Select paid eval runner
        id: runner
        env:
          AWS_PAID_RUNNER_ENABLED: ${{ vars.RUNNER_E2E_AWS_ENABLED }}
        run: |
          set -euo pipefail
          github_runner='ubuntu-latest'
          aws_runner='runs-on/fleet=paperclip-public-pr-x64/env=public-ci'

          if [ "$AWS_PAID_RUNNER_ENABLED" = true ]; then
            echo "runner=$aws_runner" >> "$GITHUB_OUTPUT"
            echo '::notice title=Paid eval routing::Using an ephemeral RunsOn Fleet runner'
          else
            echo "runner=$github_runner" >> "$GITHUB_OUTPUT"
            echo '::notice title=Paid eval routing::RUNNER_E2E_AWS_ENABLED is not true; using the proven GitHub-hosted runner'
          fi

  live_matrix:
    name: Balanced provider/model matrix
    needs: authorize
    if: github.event_name != 'schedule' || vars.RUNNER_LIVE_EVALS_NIGHTLY_ENABLED == 'true'
    # The authorize job selects one of two literal, reviewed runner labels;
    # dispatch inputs and repository variables cannot inject an arbitrary label.
    runs-on: ${{ needs.authorize.outputs.eval_runner }}
    timeout-minutes: 180
    permissions:
      contents: read
    environment:
      name: runner-e2e-paid

    steps:
      - name: Reauthorize paid execution before provider access
        env:
          GH_TOKEN: ${{ github.token }}
          REF: ${{ github.ref }}
          DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
          ACTOR_ID: ${{ github.actor_id }}
          TRIGGERING_ACTOR: ${{ github.triggering_actor }}
          ALLOWED_ACTOR_IDS: ${{ vars.RUNNER_E2E_ALLOWED_ACTOR_IDS }}
        run: |
          set -euo pipefail
          test "$REF" = "refs/heads/$DEFAULT_BRANCH"
          jq -e 'type == "array" and length > 0 and all(.[]; type == "number" and . > 0 and floor == .)' <<< "${ALLOWED_ACTOR_IDS:-}" >/dev/null
          triggering_actor_id="$(gh api "users/$TRIGGERING_ACTOR" --jq .id)"
          jq -e --argjson candidate "$triggering_actor_id" 'index($candidate) != null' <<< "$ALLOWED_ACTOR_IDS" >/dev/null
          jq -e --argjson candidate "$ACTOR_ID" 'index($candidate) != null' <<< "$ALLOWED_ACTOR_IDS" >/dev/null

      - name: Checkout repository
        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7

      - name: Generate private eval-repository token
        id: evals_token
        env:
          COMMITPERCLIP_KEY: ${{ secrets.COMMITPERCLIP_KEY }}
          GH_REPO: ${{ github.repository }}
        run: |
          set -euo pipefail
          token="$(node .github/scripts/get-bot-token.mjs)"
          echo "::add-mask::$token"
          echo "value=$token" >> "$GITHUB_OUTPUT"

      - name: Checkout canonical Evalbook reporter
        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
        with:
          repository: paperclipai/paperclip-evals
          ref: 0c8dfea0ef71a73e909b59a5c0484554cbee199b
          path: .paperclip-evals
          token: ${{ steps.evals_token.outputs.value }}
          persist-credentials: false

      - name: Setup pnpm
        uses: pnpm/action-setup@0977fd99725f1db4007ccb2928dbb4e90d06cc86 # v6
        with:
          version: 9.15.4

      - name: Setup Node.js
        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
        with:
          node-version: 24

      - name: Install dependencies
        run: pnpm install --frozen-lockfile

      - name: Restore compatible weekly baseline
        uses: actions/cache@caa296126883cff596d87d8935842f9db880ef25 # v5
        with:
          path: packages/paperclip-runner/.paperclip-local/evals/workflows/history
          key: runner-live-eval-history-${{ github.ref_name }}-${{ github.run_id }}
          restore-keys: |
            runner-live-eval-history-${{ github.ref_name }}-

      - name: Build provider-neutral eval kernel
        run: pnpm --filter @paperclipai/paperclip-eval-kernel build

      - name: Run trend-only live matrix
        env:
          OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
          PAPERCLIP_EVAL_BASELINE_READY: "true"
          PAPERCLIP_EVAL_RUNNER_BUILD: ${{ github.sha }}
          PAPERCLIP_EVAL_MAX_CAMPAIGN_COST_USD: "12"
          PAPERCLIP_EVAL_SCHEDULE_SEED: runner-live-seven-week-v1
          PAPERCLIP_EVAL_CANDIDATE: ${{ inputs.candidate }}
          PAPERCLIP_EVAL_CASE: ${{ inputs.case }}
          PAPERCLIP_EVAL_LIMIT: ${{ inputs.limit }}
          PAPERCLIP_EVALBOOK_PROGRAM: ${{ github.workspace }}/.paperclip-evals/evals/paperclip-runner/tools/eval_program.py
        run: pnpm --filter @paperclipai/paperclip-runner report:runner-live-evals

      - name: Publish job summary
        if: always()
        run: |
          summary=packages/paperclip-runner/.paperclip-local/evals/workflows/github-live-summary.md
          if [ -f "$summary" ]; then
            cat "$summary" >> "$GITHUB_STEP_SUMMARY"
          fi

      - name: Upload safe live eval bundle
        if: always()
        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
        with:
          name: runner-live-evals-${{ github.run_id }}
          path: packages/paperclip-runner/.paperclip-local/evals/workflows/
          retention-days: 30
          if-no-files-found: error