04 — True Dry-Run Feasibility (Workstream C)

Pilot pair chosen

DOT_KG_EXPLAIN (producer) + DOT_KG_EXPLAIN_VERIFY (verifier), from v_process_discovery_dot_family_pairs (pair_group=DOT_KG_EXPLAIN, pair_complete=t). Chosen because: KG is the only family with 18/18 complete pairs; clean producer→verifier semantics; both legs already structurally start+end. (Runner-up DOT_KG_CONVERSE_EXTRACT+_VERIFY rejected: verifier engine hybrid vs producer agent_api — extra surface.)

Live inspection of the pair (dot_tools)

code	paired_dot	trigger_type	engine	script_path	file_path	last_executed	usage_count
DOT_KG_EXPLAIN	DOT_KG_EXPLAIN_VERIFY	on-demand	agent_api	NULL	NULL	NULL	0
DOT_KG_EXPLAIN_VERIFY	DOT_KG_EXPLAIN	dual	agent_api	NULL	NULL	NULL	0

Feasibility findings

• Input requirements: none wired — no fixture, no input contract. (agent_api engine implies an external agent call, not a SQL function.)
• Output destination: undefined — no output_ref convention exists for DOT tools; KG writes would land in universal_edges / knowledge graph (production).
• Mutation behavior: unknown/unbounded — KG DOTs (classify/validate/link/explain) are designed to write graph rows; there is no asserted mutating=false plan mode for DOT_* tools.
• Dry-run flag: none for DOT tools. Plan/verify modes exist only in the IU-command layer (dot_iu_command_run.run_mode ∈ plan/verify/apply) — a different layer that governs IU mutation commands, not DOT execution.
• Test fixture: none.
• Rollback/cleanup: none defined for a DOT run.
• Worker/queue dependency: KG DOTs are not queue-driven; there is no live runner that picks a DOT and executes it. script_path/file_path NULL, usage_count 0 confirm they are declared-but-never-invoked.

Classification: WRAPPER_NEEDED

A true dry-run is impossible today without a wrapper that: (1) mints process_run_id/correlation_id; (2) invokes the DOT in an asserted plan/verify-only mode (no apply, no business write); (3) routes any output to a temp/test namespace; (4) logs evidence_type=DRY_RUN observations; (5) guarantees birth_registry and KG edges unchanged; (6) cleans up. None of (1)–(6) exists.

Decision: DO NOT EXECUTE. 0 production DOT executed this macro.

The SIMULATED_DRY_RUN observation (doc 03) substitutes for runner proof: it proves the ledger + correlation + scoring path end-to-end without touching any DOT. The remaining gap is purely the execution wrapper → next macro (doc 09).