Asking the fleet what it is doing…
monad-knowledge Wi-Fi sensing lab · FIIT STU
Campaign

Cross-geometry CSI drift at real floor count — 40 floors, nested seeds, honest LOFO baseline

c-csi-cross-geometry-scaleout · exp-csi-static

Archive snapshot, as of 19 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.

Sessions

state
Session State Runs Synthesis Criticism Figures Verdict
C0YFVS84 2026-07-09T17:20 finished 1 ✓ — — Preflight PASSED and launch-recipe proven; powered 1650-run corpus deferred to a dedicated long-running CI dispatch (insufficient LLM budget this instance to supervise it to completion). No science criterion resolved this session.
JR88VHFP 2026-07-09T14:13 finished 1 ✓ — — Prereq #2 CLEARED (PostGIS 13→113 floors; Phase-2 ResPlan import landed since 12:49). Probe run on newly-imported floor resplan-10156 passed the gate end-to-end (where=local, gis export-floor, 7 artefacts, 5s). C1 now UNBLOCKED but unrun — the powered ~600–1650-run sweep exceeds this session's remaining LLM budget; dispatch a fresh /campaign-systematic run. C2/C3 stand resolved from 01KWYBE.
RZTRETB3 2026-07-09T12:49 finished 0 ✓ — — Blocked no-op re-verification (0 runs): PostGIS reachable (prereq #1 cleared, 13 floors) but Phase-2's ≥40-floor prerequisite unmet — only the same 6 of 10 Phase-1 floors staged, so re-running the pilot is pure redundancy vs session 01KWYBE… (2026-07-07). C2/C3 already settled there; C1 stays blocked pending the floor import. Correct outcome per Step 2.5: re-confirm known block cheaply, do not burn budget on science that cannot move the gate.
QW28P3EJ 2026-07-09T09:10 finished 1 ✓ — — BLOCKED (partial): infra gate cleared, but stock exp-csi-static emits no occupancy-MAE/LOFO artefacts to satisfy the criteria; brief needs a declared notebook: reduction harness before Phase 1 can run.
0HCK9J4S 2026-07-09T08:31 in progress 1 — — — —
TP9QVAVH 2026-07-08T20:31 in progress 0 — — — —
QFTF7DNT 2026-07-08T20:16 finished 0 ✓ — — BLOCKED at preflight: PostGIS auth failure (user `monad`, port :5434) — 0 runs launched; all three criteria unresolved pending infra fix.
B7CY16Z4 2026-07-07T23:01 failed 0 ✓ — — Blocked on prerequisite: PostGIS auth failed (user 'monad', :5434) — 0/3 validation floors could stage a scene; all criteria unresolved. Fix DB creds + fold occupancy grid into the sweep layer, then re-dispatch.
XDNJQCT5 2026-07-07T22:48 finished 0 ✓ — — All 3 criteria UNRESOLVED — Phase-1 blocked by a launch-path/machinery defect (exp-csi-static rejects the sweep-shaped params), compounded by budget exhaustion before a corrected sweep could complete. Follow-on: re-dispatch via sim_sweep_launch.
Y4D6KG6S 2026-07-07T21:05 finished 0 ✓ — — Prerequisite-blocked: PostGIS unreachable from runner (connection refused :5432); zero runs launchable since exp-csi-static stages scenes via gis export-floor. No new evidence; C2/C3 stand resolved from prior session 01KWYBEJQK, C1 remains under-powered/Phase-2-blocked.
PVWXP97R 2026-07-07T20:53 finished 0 ✓ — — Infra-blocked: PostGIS unreachable (127.0.0.1:5432 connection refused) — 0 runs staged, all 3 criteria unresolved; dispatch-layer schema fix recorded for next session.
KMRY0DEY 2026-07-07T13:14 finished 90 ✓ ✓ — Phase-1 pilot (6 of 10 floors staged): machinery validated + variance-split settled, powered primary not met as pre-declared. LOFO cross-floor inflation pooled ×3.37 (honest leave-one-occupancy-out within 0.026 vs held-out 0.088), floor-cluster 95% CI [1.04, 9.85] — at 6 clusters the interval is unreliable, its exclusion of 1.0 not leaned on. The ×3.37 hides extreme per-floor heterogeneity (0.70× resplan-7421 no-inflation → 49.8× resplan-1374; mean-of-ratios 10.1 vs ratio-of-means 3.37) — that spread IS why the CI is order-of-magnitude wide. C2 geometry-dominates the CV(N) signal DECISIVELY (σ²_floor 0.0109 vs σ²_seed 0.000165, 66×; between-floor SD 0.105 vs between-seed 0.0116). C3 honest held-out baseline used. Criteria 2/3 (C2, C3; C1 under-powered fail). geometry-dominates is a CV(N) property, distinct from the MAE-inflation drift. No hypothesis strength change; design input quantifying that Phase 2 needs ≥40 floors. 90 runs gate-passed.
5E36XW4K 2026-07-07T10:24 finished 0 ✓ — — 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.
D41PQTQG 2026-07-06T20:23 finished 0 ✓ — — All 3 criteria UNRESOLVED — blocked pre-launch: sim_params declares only seeds [0,1,2], no enumerated ~40-floor axis / occupancy grid / base geometry. Fix is a brief-design amendment (add floors + occupancy axes), not a mid-session invention (IP-085). Plan pre-registered; powered-n confirmed at ≈40 floors.

Brief

Why this campaign exists

c-csi-cross-geometry-resplan's inflation ratio is "the one number the thesis chain leans on", and the 2026-07-06 audit found (a) the comparator mislabeled in-sample training error as "within-floor", and (b) the honest bootstrap-over-floors CI touches the decision gate at only 6 hand-picked floors, single seed. The CI-width inversion (n₀ = z²σ²/d², σ≈4.82, d=1.5) demands ≈40 floors. Single-floor/single-seed is the binding external-validity gap across the whole sim corpus — this campaign is the scale-out that separates geometry variance from seed noise.

Design

Nested design, not two arms: floors (random effect) × seeds ([0,1,2] nested within floor) × occupancy grid (5 levels, 2.4 GHz). Estimand is a variance decomposition + the inflation-ratio CI (cluster treatment per thompson2012_e41f ch. 12). Dispatch via /campaign-systematic on CI, budget_factor=1.0.

Two-phase reality (2026-07-07 amendment after session 01KWWHJCTHAD… correctly launch-blocked on an under-specified frontmatter):

  • Phase 1 — pilot (superseded): the original 10 ResPlan floors × 3 seeds × 5 levels = 150 runs, declared under-powered for the primary criterion (at n=10 the inflation-ratio CI half-width ≈ z·σ/√n ≈ 3.0, double the 1.5 target); it resolved only the variance-split (σ²_floor vs σ²_seed) and machinery validation. Banked in session 01KWYBEJQK… (2026-07-07).
  • Phase 2 — powered n, now staged (2026-07-09): the ResPlan import batch (converters/01b_resplan_scaleout.ipynb → radio-coverage/01c_resplan_csi_links.ipynb) staged 110 CSI-linked floors — the current replications: block. n=110 ≫ 40 (n₀ = z²σ²/d² ≈ 40 at σ≈4.82, d=1.5), so the primary inflation-ratio CI is powered. Each floor carries a 1 Tx (living/corridor hub) + 3 Rx (spread rooms) link; the csi_cross_geometry_scaleout.py reduction pools per-link CV(N) across floors.

Prerequisites

  1. PostGIS reachable from the runner env on :5435 — session 01KWWJ0REM… (topology sibling) probed and found connection refused on 2026-07-06; every exp-csi-static run stages its scene via gis export-floor. Verify before dispatch.
  2. Phase 2 floor corpus: satisfied (2026-07-09) — 110 CSI-linked ResPlan floors staged in replications: (import + placement notebooks above).

Artefact contract (hard gate)

cross_geometry.parquet (per floor × seed × N), per-floor CV(N) reduction table, LOFO within/held-out MAE breakdown with bootstrap-over-floors CI, σ²_floor/σ²_seed in metrics.json.

Figure render request (mandatory)

  • csi_cross_geometry small multiples (per-floor CV(N)) + a CI-bearing summary panel showing cross-floor spread — the uncertainty visualization whose absence hid the original finding.
  • seed_trace per floor-group.

Honest scope

A design input for the thesis drift claim and the multi-site field capture — it quantifies how many distinct geometries the field study needs. Still in-silico.