AI host activation matrix

The production activation matrix is the final technical gate for Codex, Claude Code, Antigravity CLI, Cursor Agent CLI, and Grok Build CLI. One operator command can preflight and run all five live campaigns in sequence. A separate read-only command independently verifies five completed campaigns and emits one public, secret-free summary.

Each host must contribute exactly three unique live trials. All 15 trials must use the same release, source commit, verified package build, caller-owned bot, and guild binding. Every trial must finish setup, apply or resume, independent Activity Evidence validation, and exact disposable-guild baseline restoration in under ten minutes. With 15 fixed trials, the production target of at least 95% requires 15/15 passes because 14/15 is only 93.3%.

The matrix revalidates every private HMAC attestation instead of trusting the per-host public summaries. It then aggregates the original trial records again, so reusing a trial ID, client session, attestation envelope, or Activity Evidence identity across hosts fails closed.

Prerequisites

Run each host’s three-trial campaign with a unique run ID and the same:

• released package and source commit;
• caller-owned Discord bot and disposable guild;
• private activation artifact root;
• exact target and write approvals.

Live campaigns can incur model-provider cost and Discord writes. They require fresh operator approval at execution time. CI fixtures, parser tests, catalog checks, and offline host inspection do not satisfy this gate.

Run all five campaigns

The orchestrator accepts one approval bound to the exact release, matrix run, source commit, guild, and expected bot. Before starting the first campaign it checks the presence and bounded local shape of all four provider API keys, the readability of the local Codex authentication file, and the exact installed version of all five host CLIs. It then derives a unique run ID for each host and runs the hosts sequentially. It stops at the first failure, rejects package-build drift between hosts, and invokes the independent 15-trial verifier only after all five campaigns pass.

The command below makes paid model calls and writes to the disposable Discord guild. Set these values only immediately before an approved run. The bot token and provider keys remain process-local and are not included in the public result.

export DISCORD_TESTBOT_B_TOKEN='<rotated caller-owned bot token>'
export DISCORD_EXPECTED_BOT_ID='<expected bot id>'
export DISCORD_MCP_ACTIVATION_ARTIFACT_ROOT='/private/discord-mcp-activation'
export ANTHROPIC_API_KEY='<Claude Code provider key>'
export GEMINI_API_KEY='<Antigravity provider key>'
export CURSOR_API_KEY='<Cursor provider key>'
export XAI_API_KEY='<Grok provider key>'

pnpm benchmark:activation:matrix -- \
  --release 0.24.0 \
  --run-id five-host-live-001 \
  --source-commit '<40-character source commit>' \
  --guild '<disposable guild id>' \
  --confirmation 'APPROVE_FIVE_HOST_ACTIVATION_MATRIX:0.24.0:five-host-live-001:<guild id>:<expected bot id>:<40-character source commit>' \
  --codex-version '<installed Codex version>' \
  --claude-code-version '<installed Claude Code version>' \
  --antigravity-cli-version '<installed Antigravity CLI version>' \
  --cursor-cli-version '<installed Cursor Agent CLI version>' \
  --grok-cli-version '<installed Grok Build CLI version>'

Codex authentication is read from CODEX_HOME/auth.json, or from the default Codex home when CODEX_HOME is unset. A failed or interrupted matrix is never reported as partial success. Preserve its private artifacts for diagnosis and use a new matrix run ID for a later approved attempt.

Verify the five campaigns

Keep the bot token in the process environment. The verifier uses it only to authenticate the private campaign envelopes and prints neither the token nor private Discord identifiers, paths, or Activity Evidence payloads.

export DISCORD_TESTBOT_B_TOKEN='<caller-owned bot token used by the campaigns>'
export DISCORD_ACTIVATION_GUILD_ID='<disposable guild id>'
export DISCORD_EXPECTED_BOT_ID='<expected bot id>'

pnpm benchmark:activation:matrix:verify -- \
  --artifact-root /private/discord-mcp-activation \
  --expected-release 0.24.0 \
  --expected-commit '<40-character source commit>' \
  --expected-build-digest 'sha256:<64 hex characters>' \
  --codex-run-id codex-activation-live-001 \
  --claude-code-run-id claude-code-activation-live-001 \
  --antigravity-cli-run-id antigravity-cli-activation-live-001 \
  --cursor-cli-run-id cursor-cli-activation-live-001 \
  --grok-cli-run-id grok-cli-activation-live-001

A successful orchestrated or read-only verification result contains a public aggregate that uses discord-mcp.activation-trials-verifier.v1, reports verified: true, host_count: 5, and contains five three-trial host summaries. Any missing, mixed, duplicated, tampered, non-live, unsafe, incomplete, slow, or baseline-drifting campaign produces one generic failure envelope and a nonzero exit code.

The public summary is useful release evidence, but the private envelopes remain caller-owned. Do not upload the artifact root, bot token, guild identifiers, or raw Activity Evidence.