SAMPLE DATA · The server sent no advice cards, so these are the built-in example cards and seal; their numbers, stations and hashes are written into the page. Hand-written for the interface, not a live run or a measured result.

Site Advisor & Pre-Registered Causal Experiments (P3-10 / P4-08) · Stations Only, Never People

Every number is recomputed by SQL before you see it.

Click any highlighted number in a card below to inspect the exact DuckDB/SQL query the Tier 1 model ran to verify it. Any number that does not reproduce is automatically redacted and flagged. Accepting a recommendation hashes a switchback measurement plan onto the SHA-256 chain before the change starts.

Tier 1 SQL Verification Rate

6/7

1 un-reproduced number redacted (`mr_hallucinated_delay`)

SEALED CAUSE MATCHED ✓

CRYPTOGRAPHIC BLIND EVALUATION SEAL · SEAL_9A4C8E20F1B2

Revealed Root Cause: “13:00 High-Bay Aisle 4 pallet replenishment occupies 2 of 3 counterbalance forklifts during peak carrier arrivals at Dock 3”

Committed `sha256(cause + salt)` before blind Site Advisor run · Post-run reveal verified against chain

✓ SEALED CAUSE REVEALED: SHA-256 MATCH

Click any yellow metric badge to view its SQL recomputation proof

Recommendation ID: rec-dock3-replenishment · Class: TEST · Provenance: RECORDED · Role: dock_lead

CAM-01 · North Dock Bay 3 + CAM-02 · Aisle B2 Intersection

3/3 SQL NUMBERS REPRODUCEDSTATUS: MEASURING

1. SAW (Observed Station Telemetry — Click Any Number for SQL Proof)

Outbound freight trucks wait for a counterbalance forklift at Dock 3 between 13:00–14:30 vs turnaround at Dock 1, while Aisle B2 records pedestrian-forklift <2.0m encounters per shift.

2. WHY (Causal Hypotheses Tested)

[H1] 13:00 pallet-racking replenishment on CAM-10 occupies 2 of 3 counterbalance forklifts during peak carrier arrivalsSUPPORTED · 74% GAP

[H2] Carriers arriving off-appointment window at North Loading DockRULED_OUT · 8% GAP

3. DO (Recommended Operational Action & Safety Veto Check)

RESEQUENCE_TASK: task="High-Bay Aisle 4 pallet replenishment" · from="13:00–14:30" · to="10:30–12:00 (off-peak dock window)"

Owner Role: dock_lead · Revert Condition: Restore High-Bay Aisle 4 replenishment schedule to 13:00 if carrier dwell > 18 min

✗ Vetoed Alternative: Add a 4th counterbalance forklift to North Loading Dock during 13:00 peak — increases Aisle B2 pedestrian-forklift <2.0m encounter density by +38% (vetoed by Safety Rule IR)

4. EXPECTED GAIN (ERLANG_C_SWITCHBACK)

-14 to -4 min/truck

95% Credible Interval · Primary metric: dock3_wait_for_forklift_min

5. PRE-REGISTERED SWITCHBACK CHECK (P4-08)

✓ PRE-REGISTERED AT CHAIN SEQ #18

Design: SWITCHBACK · 6 pairs of 30 min blocks · Min detectable effect: 3.5 · Smallest p: 0.0313

SHA-256 Method Hash: 8f4b2c91e04d7a129b3e6011…

6. SWITCHBACK EXPERIMENT VERDICT READ-OUT

Worked: −9.2 min/truck (CI −14 to −4.4), 6 pairs

Exact sign-flip p = 0.0313 (α = 0.05) · Entire CI < 0

Recommendation ID: rec-corridor-ppe-dispenser · Class: DO · Provenance: RECORDED · Role: supervisor

CAM-03 · East Visitor Corridor

2/2 SQL NUMBERS REPRODUCEDSTATUS: PROPOSED

1. SAW (Observed Station Telemetry — Click Any Number for SQL Proof)

East Visitor Corridor recorded unhelmeted visitor entries over 14 shifts, with of violations occurring within the first 4.5 metres of Door E2.

2. WHY (Causal Hypotheses Tested)

[H1] Visitor hardhat rack is mounted 11 metres inside the corridor past the CAM-03 threshold boundarySUPPORTED · 89% GAP

3. DO (Recommended Operational Action & Safety Veto Check)

RELOCATE_FIXTURE: fixture="Visitor Hardhat Dispenser Rack" · from="Pillar E4 (+11m inside)" · to="Door E2 Vestibule (-1.5m before boundary)"

Owner Role: supervisor · Revert Condition: Return dispenser rack to Pillar E4 if vestibule egress width drops below 1.5m

✗ Vetoed Alternative: Add a 3.0s grace delay before triggering CAM-03 hardhat alert — leaves visitors unprotected for the first 4.2 metres under overhead crane rail (vetoed)

4. EXPECTED GAIN (BEFORE_AFTER_SWITCHBACK)

-2.8 to -1.6 alerts/shift

95% Credible Interval · Primary metric: visitor_corridor_unhelmeted_entries_per_shift

5. PRE-REGISTERED SWITCHBACK CHECK (P4-08)

Recommendation ID: rec-kitchen-hood-glare-audit · Class: HOLD · Provenance: RECORDED · Role: kitchen_lead

CAM-05 · Commercial Kitchen Prep Line A

1/2 SQL NUMBERS REPRODUCEDSTATUS: PROPOSED

⚠ TIER 1 SQL AUDIT FLAG: One or more numbers in this card did not match independent DuckDB recomputation and have been redacted below. Click the redacted badge to inspect the discrepancy.

1. SAW (Observed Station Telemetry — Click Any Number for SQL Proof)

Station 3 on Prep Line A triggered head-covering false alarms suppressed by the Tier 1 check, while draft advisor claimed prep cycle delay.

2. WHY (Causal Hypotheses Tested)

[H1] Overhead stainless exhaust hood casts a high-contrast specular shadow across Station 3 at 45° camera pitchSUPPORTED · 82% GAP

3. DO (Recommended Operational Action & Safety Veto Check)

ADJUST_DIFFUSER: station="Prep Line A · Station 3" · action="Install matte LED diffuser strip under hood lip"

Owner Role: kitchen_lead · Revert Condition: Remove diffuser strip

4. EXPECTED GAIN (PAIRED_STATION_COMPARISON)

-4 to -1.5 candidates/day

90% Credible Interval · Primary metric: cam05_tier0_false_candidates_per_day

5. PRE-REGISTERED SWITCHBACK CHECK (P4-08)