Campaign session

01KWY1QWEJ9PJ5RVDP5E36XW4K

finished 2026-07-07 10:24:52.178751+00:00 → 2026-07-07 10:27:56.700144+00:00 · 0 runs · supervisor: react-agent

“Partial/deferred: powered nested grid (~600 runs, 8–12 CPU-h) exceeds this interactive session's budget and needs CI dispatch; all three criteria left unresolved pending scale-out.”

Archive snapshot, as of 7 h 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: bootstrap-over-floors 95% CI on the LOFO inflation ratio has lower bound exceeding 1.5x.	no
Variance split: nested decomposition reports sigma^2_floor and sigma^2_seed; geometry-dominates iff between-floor SD of per-floor CV(N) exceeds within-floor between-seed SD.	no
Comparator honesty: the LOFO denominator is the held-out within-floor leave-one-occupancy-out baseline, never in-sample training residual.	no

Synthesis

Session 01KWY1QWEJ9PJ5RVDP5E36XW4K — deferral synthesis (no runs executed)

Outcome: partial. Zero runs launched; all three success criteria remain UNRESOLVED, deferred to a CI scale-out. This is a deliberate budget-discipline stop, not an orphan.

Why no fan-out

The declared design is a nested random-effects grid: ~40 floors (ResPlan + ZInD) x seeds [0,1,2] nested per floor x ~5 occupancy levels = ~600 exp-csi-static runs, 8–12 CPU-h of Sionna CPU ray-tracing (no Metal/CUDA backend; n_placements=64 layouts per run). Two hard execution blockers made local fan-out unviable in an interactive session: (a) the sionna-csi-runner image is not yet pushed to GHCR, so where=remote-vm cannot resolve a digest; (b) the ResPlan/ZInD floor slug inventory was not enumerable within budget (only the a_seminar default is known). Attempting a sub-scale launch would have consumed the remaining token/USD budget on wall-clock polling and yielded, at best, a <6-floor corpus — reproducing the exact under-powering this campaign exists to correct.

Powering confirmed (Step 2a)

The brief's floor count is correctly sized: n0 = z^2*sigma^2/d^2 = 1.96^2 * 4.82^2 / 1.5^2 ≈ 39.7 floors (pilot sigma≈4.82, d=1.5, alpha=0.05). Any materially smaller floor set cannot clear the inflation-ratio gate with a lower CI bound; the 6-floor/single-seed predecessor is precisely the invalid design.

Criterion status

C1 (inflation-ratio CI > 1.5x): unresolved — requires the full ~40-floor bootstrap-over-floors corpus. No cross_geometry.parquet produced.
C2 (variance split, sigma^2_floor vs sigma^2_seed): unresolved — nested decomposition needs the floor x seed grid; not run.
C3 (comparator honesty): unresolved from data, but architecturally guaranteed by the declared reduction: sim_reduction_run lofo_cv computes the denominator as held-out within-floor leave-one-occupancy-out MAE, never in-sample residual. The contract is correct; it awaits execution.

Recommendation

Dispatch via /campaign-systematic on CI (budget_factor=1.0), exactly as the brief instructs, after clearing the prerequisite: push+pin sionna-csi-runner to GHCR (docker-simulators.yml) and confirm the ResPlan/ZInD floor set is registered in PostGIS. Unit of replication is the floor (n≈40); seeds [0,1,2] are the nested within-floor variance axis. Requested figures (csi_cross_geometry small multiples + CI summary panel; seed_trace per floor-group) are not rendered — zero data corpus. figure_count=0 is honest here, not a defect.

Attached runs

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