Campaign session

01KX328XQESKCBSP7HQW28P3EJ

finished 2026-07-09 09:10:22.702075+00:00 → 2026-07-09 09:12:54.911510+00:00 · 1 run · supervisor: react-agent

“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.”

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.

Success criteria

Criterion	Resolved
Primary: bootstrap-over-floors 95% CI on LOFO inflation ratio, lower bound > 1.5x	no
Variance split: nested sigma^2_floor vs sigma^2_seed; geometry-dominates iff between-floor SD of CV(N) > within-floor between-seed SD	no
Comparator honesty: LOFO denominator is held-out within-floor leave-one-occupancy-out baseline	no

Synthesis

Session synthesis — c-csi-cross-geometry-scaleout (Phase 1 pilot)

Outcome: BLOCKED at the artefact-contract gate. Zero science runs launched. One preflight probe.

What the probe established (Step 2.5 dependency-chain gate)

A single cheap probe — exp-csi-static, floor resplan-1374-floor-0, n_agents=4, n_placements=8, seed=0, 2.4 GHz, where=local (run 01KX32AZAQTJ7G97JCXCA0NDCA) — cleanly separated infra health from the artefact contract:

Infra gate CLEARED. PostGIS is reachable (gis_stats: 2431 rows, 13 floors). Scene-staging via gis export-floor succeeded, the floor exists, and the run passed the gate (gate_passed=true, exit 0, 6.3 s). The :5435 connection-refused blocker that stopped sibling session 01KWWJ0REM… on 2026-07-06 is resolved — Prerequisite #1 is met.
Artefact-contract gate FAILS. Stock exp-csi-static emits raw CSI only: csi.hdf5, links.parquet, fig_csi_vs_occupancy.png, a schema, and a 694-byte runner-telemetry metrics.json. The run-record metrics field is empty {}. There is no occupancy-estimation MAE, no cross_geometry.parquet (per floor×seed×N), no per-floor CV(N) table, no LOFO within/held-out breakdown, and no σ²_floor/σ²_seed.

Why this blocks all three criteria

Each criterion is a downstream occupancy-inference + cross-floor reduction (LOFO inflation ratio; nested variance decomposition; honest held-out baseline). sim_reduction_run's lofo_cv/multi_seed_ci aggregate per-run metrics, which are empty — and no occupancy-MAE column exists in any emitted artefact. Fanning out the 150-run sweep (10 floors × 5 occupancy × 3 seeds) would produce 150 un-reducible piles of raw CSI. No parameter change to exp-csi-static recovers this; it is structural.

Prerequisite to unblock (design-loop, not a re-dispatch)

The brief must declare a notebook: reduction harness (IP-087) that trains an occupancy estimator on the emitted CSI, computes per-run occupancy MAE, and writes cross_geometry.parquet + per-floor CV(N) + the LOFO within/held-out breakdown + σ²_floor/σ²_seed into metrics.json. Once that harness exists, Phase 1 (the pre-declared under-powered pilot, n=10 floors) can run to resolve the variance split, and Phase 2 (≥40 floors) can power the primary criterion. Criteria remain unresolved — blocked pre-corpus, not refuted.

Attached runs

Run	Gate	Purpose	Replay
XCA0NDCA	✓	chain+artefact-contract validation	—