[Kukks]

Builds on #85. Three small additions to the executor profile, found while moving a project (Portcullis) from the delegatee to the executor.

Own database. executor-initdb creates an executor database before the executor starts, the same way bucket-sync-initdb does for bucket-sync, and the executor's EXECUTOR_DATABASE_URL points at it. In #85 the executor writes to postgres's default postgres database, which is where the delegatee kept its tables. A stack upgraded in place from the delegatee would hand the executor's migrations a database that already holds the delegatee's tables. I haven't hit a collision, but a database of its own rules it out.

--profile delegatee keeps working. It now resolves to executor instead of failing with "unknown profile", so existing scripts and docs still start the stack.

EXECUTOR_RESERVE passes through (default 15m, the executor's own default, also in .env.defaults). The executor batches a renewal at expiry minus the reserve. With the default ARKD_VTXO_TREE_EXPIRY=180 that is immediate. A project running much longer-lived coins (3072 s) otherwise waits until 15 minutes before expiry, and its renewal tests time out. Setting EXECUTOR_RESERVE=50m renews them as soon as they're due.

Tested

• node --test: 8/8. The new test/executor-profile.test.mjs fails on delegatee -> executor #85 (no alias) and passes here.
• node regtest.mjs start --env <file> --profile ark --profile emulator --profile delegatee, on a separate project with every host port moved: the active profiles resolve to ark, base, emulator, delegatee, executor, executor-initdb exits 0, the executor database exists, /v1/info answers (v0.0.4-rc.1, the fixed regtest cosigner key, server key equal to arkd's signer), the admin port's /v1/admin/health answers 200, and the container gets EXECUTOR_RESERVE from the env file.

[arkana-ai-bot]

Arkana review — PR #87: executor own database, delegatee alias, reserve passthrough

No protocol-critical paths touched; this is regtest tooling only.

---

✅ What looks good

executor-initdb (docker/compose.ark.yml +290–310)
Idempotent shell one-liner (SELECT 1 … | grep -q 1 || CREATE DATABASE) matches the pattern used by bucket-sync-initdb. restart: 'no' is set, condition: service_completed_successfully is correct. Hardcoded postgres password is fine for regtest.

Database isolation (compose.ark.yml +323)
Changing EXECUTOR_DATABASE_URL from …/postgres to …/executor is the right call. The stated motivation (avoid migration collisions with tables left by an older delegatee) is sound.

EXECUTOR_RESERVE passthrough (compose.ark.yml +326, .env.defaults +141–143)
The env var threads through correctly with the executor's own default as fallback.

Alias resolution (lib/profiles.mjs +10–11)
delegatee: ['executor'] correctly chains through resolveProfiles without collision: since the Docker Compose delegatee service was removed in #85, the extra --profile delegatee flag Docker Compose receives is silently ignored, so no double-start, no port 7280 conflict.

---

⚠️ Worth addressing before or shortly after merge

delegatee bleeds into DEFAULT_PROFILES (lib/profiles.mjs +10)

DEFAULT_PROFILES is derived as Object.keys(PROFILE_DEPS).filter(p => p !== 'evm-e2e'). Adding the alias to PROFILE_DEPS causes 'delegatee' to appear in DEFAULT_PROFILES, which is used as the full "start everything" fallback in regtest.mjs:348. In practice this is harmless today (no Docker Compose service matches the profile), but:

• DEFAULT_PROFILES grows by one invisible no-op entry;
• if a future service ever re-uses the delegatee Docker Compose profile name, it would silently start by default.

Consider either filtering it out explicitly:

export const DEFAULT_PROFILES = Object.keys(PROFILE_DEPS)
  .filter(p => p !== 'evm-e2e' && p !== 'delegatee');

or keeping aliases in a separate map so Object.keys(PROFILE_DEPS) stays equal to the set of real tiers.

.env.defaults comment is ambiguous (line ~141)

# Longer than a coin's life (ARKD_VTXO_TREE_EXPIRY) renews it as soon as it is due.

This reads as "if EXECUTOR_RESERVE > ARKD_VTXO_TREE_EXPIRY, renewal fires immediately", which is true but requires the reader to know the arithmetic. The PR body explains it well (3072 s example); a one-liner like # If set above ARKD_VTXO_TREE_EXPIRY the batch fires as soon as VTXOs are created. would save future confusion.

---

🔍 Test coverage

test/executor-profile.test.mjs covers the happy paths. Two gaps worth considering:

• No assertion that resolveProfiles(['delegatee']) does not include any unexpected extras beyond ['ark', 'base', 'delegatee', 'emulator', 'executor'].
• No assertion that DEFAULT_PROFILES does not include 'delegatee' (which would catch the bleed-through described above and lock in the intent).

---

No security findings. No cross-repo API surface changed. Changes are ready to land as-is; the DEFAULT_PROFILES nit can be a follow-up.