Detect motion in synthetic Wi-Fi CSI

Steps

Screens captured against Python 3 in your browser (JupyterLite) Python 3.12.7 / Pyodide 0.27.6 in JupyterLite 0.6.4; Chromium 148.0.7778.96; captureSource playwright:jupyterlite on 2026-09-10; the tool may have moved on — the text steps are the contract.

1. 1 Step 1

   Do

   In the JupyterLite notebook, run the Step 1 cell to freeze the CSI simulation contract.

   You will see

   Input: execute notebook step 1 with its stated deterministic fixture. Observed: subcarriers=24; samples/window=18; rooms=A,B Observed: activities=empty,still,walk,turn; seed=437 Observed: No packets or RF hardware are measured.

   Why it matters

   Expected: the reproducible CSI fixture contains 24 subcarriers and two rooms. The run reports 18 samples per window and four activity labels. This freezes the comparison before deriving activity features. Limit: there is no packet capture or RF hardware measurement.

2. 2 Step 2

   Do

   In the JupyterLite notebook, run the Step 2 cell (run_step(2)) to generate calibrated CSI windows.

   You will see

   Input: execute notebook step 2 with its stated deterministic fixture. Observed: A empty 0.036 Observed: A walk 0.308 Observed: B walk 0.308

   Why it matters

   Expected: empty-room magnitude change is below the walk fixture in both rooms. The executed means are 0.036 for A-empty and 0.308 for A-walk and B-walk. The fixture produces a clear synthetic activity contrast across rooms. Limit: these values are not calibrated CSI measurements.

3. 3 Step 3

   Do

   In the JupyterLite notebook, run the Step 3 cell (run_step(3)) to calculate window features.

   You will see

   Input: execute notebook step 3 with its stated deterministic fixture. Observed: empty 0.037 0.0005 0.0019 Observed: still 0.104 0.0029 0.0136 Observed: STEP 3 complete

   Why it matters

   Expected: window summaries report change, variance, and energy per activity window. The run reports the displayed empty and still feature rows. The multiple features preserve more than a single threshold signal. Limit: rows describe synthetic windows, not people or rooms.

4. 4 Step 4

   Do

   In the JupyterLite notebook, run the Step 4 cell (run_step(4)) to fit two thresholds on room A.

   You will see

   Input: execute notebook step 4 with its stated deterministic fixture. Observed: empty if mean change < 0.10 Observed: uncertain from 0.10 through 0.20 Observed: occupied if mean change > 0.20

   Why it matters

   Expected: thresholds split low, uncertain, and high magnitude-change regions. The notebook trains these thresholds using room A. An explicit uncertain band avoids forcing every window into two classes. Limit: thresholds are fixture choices, not deployment constants.

5. 5 Step 5

   Do

   In the JupyterLite notebook, run the Step 5 cell (run_step(5)) to score the held-out room B.

   You will see

   Input: execute notebook step 5 with its stated deterministic fixture. Observed: 7 0 0 1.000 1.000 0.458 Observed: Room B was not used to choose thresholds. Observed: STEP 5 complete

   Why it matters

   Expected: held-out room B reports TP, FP, FN, precision, recall, and abstention. The run reports 7 TP, no FP/FN, precision/recall 1.000, and 0.458 abstention. Room B is held out from threshold selection in this exercise. Limit: held-out synthetic performance is not room validation.

6. 6 Step 6

   Do

   In the JupyterLite notebook, run the Step 6 cell (run_step(6)) to stress a shifted baseline.

   You will see

   Input: execute notebook step 6 with its stated deterministic fixture. Observed: 1.000 1.000 0.500 Observed: Baseline shift changes the decision distribution. Observed: Action: recalibrate or abstain.

   Why it matters

   Expected: a shifted baseline can change the confidence distribution. The shifted fixture reports precision/recall 1.000 and abstention 0.500. The notebook recommends recalibration or abstention after the shift. Limit: a generated shift does not cover measured environmental variation.

7. 7 Step 7

   Do

   In the JupyterLite notebook, run the Step 7 cell (run_step(7)) to write a bounded result card.

   You will see

   Input: execute notebook step 7 with its stated deterministic fixture. Observed: held-out precision=1.000; recall=1.000; abstention=0.458 Observed: boundary=synthetic CSI only; no RF, room, or person validation Observed: decision=repeat with consented measured rooms before use

   Why it matters

   Expected: the card states fixture performance, validation boundary, and next evidence. The result card reports 0.458 abstention on held-out room B. The next step is a consented measurement in real rooms. Limit: no RF, room, or person validation has occurred.