Session synthesis — c-csi-cross-geometry-scaleout (BLOCKED at preflight)

Outcome: sealed partial, 0 runs launched. The session terminated at the mandatory dependency-chain preflight (Step 2.5) before committing any CPU.

Blocker (hard, non-recoverable by replanning)

gis_stats — the zero-CPU PostGIS reachability probe — failed:

PostGIS unavailable: (psycopg.OperationalError) connection failed:
connection to server at "127.0.0.1", port 5434 failed:
FATAL: password authentication failed for user "monad"

Every exp-csi-static run stages its scene via gis export-floor, which reads the floor geometry from PostGIS. With the database unreachable, no run in the planned 150-run Phase-1 fan-out (10 ResPlan floors × 3 seeds × 5 occupancy levels @ 2.4 GHz) can produce a valid CSI artefact. Fanning out would only reproduce the prior zero-run budget-exhaustion partial, so no sim_sweep_launch / sim_run_launch calls were made.

This is the campaign's own pre-declared Prerequisite #1 ("PostGIS reachable from the runner env … Verify before dispatch"). It is unmet. Note the failure is worse than the brief anticipated: the brief flagged connection-refused on :5435; the observed failure is an authentication failure for user monad on port :5434 — i.e. the documented port-drift (:5434↔:5435) and monad-user auth misconfiguration together.

Success criteria — all UNRESOLVED

1. Primary (LOFO inflation-ratio CI): unresolved — no runs. (Was pre-declared under-powered even at full Phase-1 corpus: n=10 floors gives CI half-width ≈ z·σ/√n ≈ 3.0 vs the 1.5 target; the powered design needs ≈40 floors.)
2. Variance split (σ²_floor vs σ²_seed): unresolved — no runs.
3. Comparator honesty (held-out within-floor LOFO baseline): unresolved — no runs.

Prerequisites to unblock (for the next dispatch)

1. Fix PostGIS connectivity on the runner: reconcile the port drift (runner is pointed at :5434; the brief expects :5435) and repair monad-user authentication (HETZNER_S3_* / PostGIS creds in the runner env). Re-verify with gis_stats returning table row counts before re-dispatching.
2. Confirm the 10 Phase-1 ResPlan floors (resplan-{12439-multiroom,1374,147440,7421,16157,12419,10156,10425,6090,8266}-floor-0) are actually staged in the now-reachable DB (gis_floorplans_inventory / gis_experiments).
3. Phase 2 only: stage ≥40 floors and extend replications: for the powered primary criterion.

Re-confirming a known infra block for ~$1 of preflight is the correct outcome here — not a wasted run. Re-dispatch once Prerequisite #1 is green.