Campaign session

01KWWJ0REMMQ995BQRRZG3K6QS

finished 2026-07-06 20:30:51.348208+00:00 → 2026-07-06 20:33:42.496951+00:00 · 0 runs · supervisor: react-agent

“BLOCKED (infra): no runs executed — PostGIS floor-export unreachable on :5435; all three criteria unresolved, zero CPU/S3 consumed.”

Archive snapshot, as of 1 day ago — the run corpus is rebuilt once a day, so this page is not a live reading. The fleet panel is the live one; it refreshes every 30 s.

Success criteria

Criterion	Resolved
Primary (topology): LOS-vs-NLOS mean CSI occupancy-sensitivity difference, 95% CI excluding 0, \|d\|>=0.2, >=10 LOS + >=10 NLOS links/floor.	no
Horse-race honesty: floor-level area correlation and link-level topology correlation each reported as CI + effective n.	no
Method: real Welch t (Satterthwaite) + exact permutation p at cluster level; single BH-FDR pass over both topology sessions, m stated.	no

Synthesis

Session synthesis — c-csi-topology-los-balanced (PARTIAL / BLOCKED)

Outcome: no simulation executed. All three success criteria UNRESOLVED. Budget consumed: 0.0 CPU-h, 0 S3 bytes.

A single cheap feasibility probe (exp-csi-static, floor a_seminar, seed 0, n_placements=8) was launched before committing the ~400-run balanced-placement batch. It terminated in 207 ms with a pre-simulation SceneStageError: gis export-floor could not reach PostGIS on 127.0.0.1:5435 / localhost:5435 (connection refused). The sionna-csi-runner image resolved and launched cleanly — contrary to the curated note's GHCR-push caveat, image resolution is not the blocker; PostGIS reachability is. Because every exp-csi-static run stages its scene from a PostGIS floor export, no run across any of the ≥20 intended floors can execute in this environment. Launching the remaining seeds/floors would only re-hit the same wall, so the batch was not dispatched.

Per criterion:

C1 (primary topology d≥0.2): unresolved — no link-level corpus produced.
C2 (horse-race CIs + effective n): unresolved — no floor/link data to correlate.
C3 (Welch t + permutation + BH-FDR): unresolved — no statistics computable on an empty corpus.

Mandatory figures (LOS/NLOS sensitivity distributions; seed_trace per-floor spread): not rendered — figure_count=0 is a consequence of the empty corpus, not a rendering defect.

Secondary design finding for the re-run. Beyond the infra blocker, the current exp-csi-static config surface (floor, placement∈{random,grid} for bodies, n_agents, n_placements, radio.fading_seed) exposes no first-class key for LOS/NLOS-balanced Tx/Rx placement stratified on wall-count-on-path. The brief's "≥10 LOS + ≥10 NLOS links per floor by construction" requires either a new placement-policy config key or a host-side scene-staging step that constructs balanced Tx/Rx pairs before the run. This must be resolved for the fair test to be expressible at all — it is the precise gap that made the sniffer-count precursor an unfair race.

Recommended before re-dispatch (via /campaign-systematic, budget_factor=1.0):

Bring PostGIS up and bound on :5435 in the runner environment; re-run this probe to confirm floor export.
Confirm the ≥20 ResPlan+ZInD floors are actually loaded into PostGIS (gis floorplans-inventory); the default catalogue is a_seminar.
Add/implement the LOS/NLOS-balanced placement-policy mechanism so the per-floor ≥10+≥10 stratified corpus is constructible.

The prior "area beats topology" state remains ambiguous and un-retired — this fair test did not run.

Attached runs

The attached runs are not in the current atlas snapshot — rebuild via web --build-sim-atlas.