Macro-2 Host-Cron vs Agent-API Weighted Comparison — R2-B2

Date: 2026-06-19 · Workstream: R2-B2-MACRO-2-OWNER-DECISION-PREREQ-CLOSURE-READINESS-2026-06-19 (Deliverable 9 of 45) · Editorial revision: rev1 Class: weighted channel comparison / decision-support · READ-ONLY · NON-ENACTING · NON-AUTHORIZING · RECOMMENDATION_ONLY — NOT AUTHORITY · NO channel selected · NO write performed.

Metadata convention. Editorial revision (rev1) only. AgentData storage revision/content_length authoritative at read time; not pinned here.

Comparison lock. A side-by-side comparison of the two candidate channels as decision-support. The weighting is illustrative; it selects no channel and the value judgment remains the Owner's. RECOMMENDATION_ONLY — NOT AUTHORITY. Any wording that reads as selection is CHANNEL_AUTHORITY_DRIFT → HOLD.

---

0. Status and non-authorization

STATUS: PASS — engineering / decision-support. A weighted comparison of host cron vs agent-api on the criteria an Owner channel decision turns on. It selects nothing.

Engineering PASS ≠ authority PASS. Default disposition: HOLD. No channel selected/wired/promoted; no DB write/DDL/DML; no TD; no blocker resolved; v0.1 not overwritten; v0.2 not authority.

Evidence basis — FRESH_READONLY_EVIDENCE (2026-06-19 RQ-13/14/16/17) + INHERITED_KB_EVIDENCE. Reading discipline: main process, no reader-agents.

---

1. Purpose

Give the Owner a structured comparison of the two candidates (Deliverables 7, 8) without choosing between them.

---

2. Sources / evidence read

Deliverables 7 (host-cron memo), 8 (agent-api memo); the Mega-Gate host-cron-vs-agent-api brief; the fresh RQ-ledger. Read directly, main process.

---

3. Accepted baseline (carried)

Both are candidates; the host-cron-vs-agent-api value judgment is the Owner's (Mega-Gate brief; recommendation-only).

---

4. Weighted comparison (illustrative — Owner reweights)

Score key: ✓✓ strong · ✓ adequate · △ weak (per criterion; not summed into a verdict).

Criterion	host cron	agent-api executor	Note (FRESH 2026-06-19)
Blast radius (lowest = best)	✓✓	✓	host: one cron line; agent-api: contract + master-switch + shared infra
Proven precedent for this lane	✓✓	✓	host: sibling scanner DOTs run on it; agent-api: KG-EXPLAIN pilot only (RQ-16)
Governance alignment (Điều 32)	✓	✓✓	agent-api: contract-bound, observable, Điều-32-promotable
In-DB observability	△	✓✓	host: snapshot-only (CAV-3/4); agent-api: contract + dispatch records
Transactionality with inspect_* writes	△	✓	neither is pg_cron-grade in-DB; agent-api dispatch is contract-traced
Lane isolation	✓✓	△	host: dedicated entry; agent-api: shared with KG lane → must bind per-DOT (X-7)
Extra gates to cross	✓✓	△	host: Điều 32 + S2; agent-api: Điều 32 + S2 + DRY_RUN→REAL_RUN promotion + transient-switch confirm
Read-only first obligation met today	✓✓	✓✓	both readable & fail-closed (RQ-14/16/17)

Reading (recommendation-only): host cron leads on blast radius / precedent / lane isolation / fewer gates; agent-api leads on governance alignment / observability / transactional tracing. There is no dominant winner; the choice is a genuine Owner trade-off. No channel is selected here.

---

5. Result

RECOMMENDATION_ONLY — NOT AUTHORITY: present both candidates with this trade-off; the Owner picks (or asks for a deeper proof in Macro-3). pg_cron/job_queue remain future-gated; manual one-shot rejected. No selection. OWNER_DECISION_REQUIRED.

---

6. Owner-gated future work

Selecting either candidate, and any wiring/binding/promotion, is Owner-gated and forbidden now.

---

7. What remains unresolved

• No winner chosen; the trade-off is the Owner's.
• Transactionality/observability deltas are provable only after a build (write-gated).
• Blockers — all OPEN.

---

8. Ready for GPT/Codex review

Yes — as a weighted comparison.

Core rule: host cron vs agent-api is a real Owner trade-off (blast-radius/precedent vs governance/observability); no dominant winner, no channel selected.

Default disposition: HOLD. Engineering PASS ≠ authority PASS. All blockers remain OPEN.