SHAKE ATLAS — DAY REPORT 2026-10-02 (UTC) local 2026-10-02 09:00 EDT -> 2026-10-02 13:00 EDT (schema shake-atlas/answers/v3-day) generated 2026-10-05T16:57:50+00:00 script 3.0.0 alignment: first full UTC hour at or after each recording start; hour blocks [H, H+1h); no clock-model shift HEADLINE upstairs: multi_1790959520.160099 — 4.21 h at 100 Hz, background 1 s RMS 0.00518 m/s^2, 0/0 high-confidence/sensitive events, 2 sub-threshold candidates downstairs: multi_1790959615.928701 — 3.75 h at 100 Hz, background 1 s RMS 0.00491 m/s^2, 1/1 high-confidence/sensitive events, 2 sub-threshold candidates overlap 2.76 h; common rate 100 Hz; safe spectral limit 45 Hz median coherence 0.001 vs significance 0.0045; corrected correlation r=0.0137; matched event pairs 0 verdict: NO SHARED SIGNAL ABOVE THE NOISE FLOOR coherence (robust view, not cherry-picked bins): median 0.001, p90 0.0037, max 0.0135 over 1351 bins; no bin survives FDR positive control: the two microphones correlate at r=0.623 (lag -0.199 s) over the same window, so the phones really did share an environment even though the accelerometers show no shared vibration data integrity: PASS (0 failure(s), 9 warning(s)) answer trust: insufficient_data=11, supported=4, tentative=28, unsupported=7 HOUR BLOCKS (UTC; local America/New_York) 1300Z 2026-10-02 09:00 EDT up 1.00 dn n/a unpaired full up RMS 0.01407 dn RMS n/a ev n/a/n/a pairs n/a coh n/a r n/a -> hours/1300Z.json 1400Z 2026-10-02 10:00 EDT up 1.00 dn 1.00 paired full up RMS 0.01405 dn RMS 0.01290 ev 0/0 pairs 0 coh 0.0031 r 0.016 -> hours/1400Z.json 1500Z 2026-10-02 11:00 EDT up 1.00 dn 1.00 paired full up RMS 0.01383 dn RMS 0.01226 ev 0/0 pairs 0 coh 0.0031 r -0.007 -> hours/1500Z.json 1600Z 2026-10-02 12:00 EDT up 0.76 dn 0.78 paired partial up RMS 0.01380 dn RMS 0.01222 ev 0/0 pairs 0 coh 0.0041 r 0.009 -> hours/1600Z.json day window: 14:00Z -> 17:00Z, 3 paired hours (1 partial); per-site statistics below describe each whole recording, cross-phone statistics are restricted to this window DATA INTEGRITY / INSTRUMENT REPORT multi_1790959520.160099 | upstairs | 1514228 rows | fs 100.0001 Hz | duration 15141.9 s | jitter 7.8512% | gaps 0 (max 0.023 s) | missing ax/ay/az none | GPS:ok | valid 99.93% | PASS multi_1790959615.928701 | downstairs | 1350572 rows | fs 100.0001 Hz | duration 13505.6 s | jitter 8.289% | gaps 0 (max 0.016 s) | missing ax/ay/az none | GPS:fix-static/speed-moving(277.0 m vs 27.0 m fix span) | valid 99.93% | WARN multi_1791026911.329183 2 | upstairs | 4390299 rows | fs 100.0001 Hz | duration 43903.0 s | jitter 7.4735% | gaps 1 (max 0.435 s) | missing ax/ay/az none | GPS:fix-static/speed-moving(530.0 m vs 32.0 m fix span) | valid 99.98% | WARN multi_1791027086.65238 | downstairs | 4399380 rows | fs 100.0001 Hz | duration 43995.5 s | jitter 7.3658% | gaps 1 (max 2.744 s) | missing ax/ay/az none | GPS:ok | valid 99.98% | WARN DATASET INTEGRITY PASS: 4 recording(s), 3 pass-with-warnings, 0 fail RECORDING upstairs csv/multi_1790959520.160099.csv rows=1514228 fs=100 Hz span=15142 s clock=utc nyquist=50 Hz safe=45 Hz 2026-10-02T12:32:58.200Z -> 2026-10-02T16:45:20.143Z sha256=ff7df0b8b67bfa6771cef205cdbfabc2d2dcef5ab9871616afb59a4a1aa695a6 RECORDING downstairs csv/multi_1790959615.928701.csv rows=1350572 fs=100 Hz span=13506 s clock=utc nyquist=50 Hz safe=45 Hz 2026-10-02T13:01:50.293Z -> 2026-10-02T16:46:55.899Z sha256=76a80d4ea5b4091222ca63c68c25d133dd01b3b74f31c665e8d20142dd1792c9 SESSION D2026-10-02 analysed [day] upstairs csv/multi_1790959520.160099.csv downstairs csv/multi_1790959615.928701.csv hours 1400Z-1600Z (3, 1 partial) overlap 2026-10-02T14:00:00.000Z -> 2026-10-02T16:45:20.143Z (2.76 h, 9920.0 s) alignment: UTC hour grid; no clock-model shift applied BEST NEXT EXPERIMENT no reproducible cross-floor vibration was found in the passive recording; next perform a labelled impulse test and the phone-swap test why (from this run): 0 matched event pairs; median coherence 0.0010 against a 0.0045 significance level; |r| = 0.014; no bin survives FDR control; 1 single-site detection(s) (0 upstairs / 1 downstairs) with no counterpart inside the matching tolerance; plain RMS is spike-dominated on: downstairs [primary] known impulse test (labelled tap/drop at known locations) (impulse) why: the passive recording holds no excitation above the sensor noise floor, so there is nothing for the structure to transmit; a known, repeated impulse is the cheapest way to create a signal both phones must record if they are mechanically coupled steps: lay both phones flat on their floors, screens still, not on carpet | note the UTC time to the second immediately before each tap | tap firmly on the floor next to the upstairs phone: 10 taps, 5 s apart | repeat the 10 taps next to the downstairs phone | repeat both series with a heavier impulse (dropped book) duration: 10-15 min labels: tap_up=impulse@:-,tap_down=impulse@:- success: every impulse becomes a matched pair with a stable onset offset and SNR > 10 dB on both phones; upstairs taps that never reach the downstairs phone are themselves a strong negative result [alternate] phone-swap test (same sequence, phones exchanged) (phone_swap_a) why: with a real excitation present, swapping the phones separates a floor response from a unit/coupling sensitivity difference - the single most valuable control for any amplification claim steps: run A: phone 1 upstairs, phone 2 downstairs, repeat the tap series | run B: swap the phones and repeat the identical tap series | keep cases, orientation and surface identical between legs duration: 2 legs x 10-15 min labels: swapA=phone_swap_a@+,swapB=phone_swap_b@+ success: the upstairs/downstairs ratio follows the FLOOR rather than the PHONE across the two legs [alternate] quiet baseline, then HVAC on/off (hvac_on) why: establishes the noise floor of each unit and whether the largest ambient contribution in this band is equipment-driven steps: 30 min quiet baseline with the HVAC off | then 30 min with the HVAC forced on | note the UTC switch times duration: 2 x 30 min labels: base=quiet_baseline@,hvac=hvac_on@:- success: a reproducible difference in band RMS or event rate between the two labelled windows [alternate] repeat a passive run without touching the phones (quiet_baseline) why: some recordings are dominated by rare samples (handling, start/stop artefacts), so their large plain-RMS values do not describe continuous vibration; a hands-off repeat confirms that steps: start the recording, lay the phone down flat, do not touch it again | stop it only at the very end from the app duration: 1 h labels: baseline=quiet_baseline@ success: plain RMS and 99.9%-trimmed RMS agree within 1.5x, i.e. no handling spikes protocol: answers/experiment-protocol.md RMS VERSUS ROBUST BACKGROUND (why the two numbers diverge) downstairs: plain RMS is dominated by rare samples, not by continuous vibration: for aT the top 0.01% of samples carry 97.71% of the total energy, so the plain RMS 0.046668 m/s^2 collapses to 0.005429 m/s^2 once samples above the 99.9th percentile are removed (8.6x), while the robust sigma is 0.003149 m/s^2 and the 1 s background is the right measure of continuous motion EVENT CATALOG (1 entries) one row per matched event pair when pairs exist; otherwise one row per single-site detection with the cross-phone fields explicitly empty, because a one-sided detection is not a shared event Event 00001 — 2026-10-02T16:46:50.889Z (downstairs) / Upstairs peak: n/a g / Downstairs peak: 0.00212 g / Amplification: n/a / Dominant frequency: 6.25 Hz / Cross-phone correlation: n/a / Delay: n/a / Classification: single-site detection (SNR 15.2 dB, 1.04 s): the other phone has no independent detection within the matching tolerance, so it is not evidence of a shared event ANSWER TRUST COUNTS: insufficient_data=11, supported=4, tentative=28, unsupported=7 Q1. What vibration events occurred at essentially the same time on both phones? [unsupported / confidence high for the onset pairing; see caveats] limiting factor: no matched event pairs within the tolerance MEASURED: No matched event. The two phones overlapped for 2.756 h; 0 upstairs and 1 downstairs detections were found, and no upstairs detection had an independent downstairs detection within 1.5 s. Detections that do not pair up are reported as unmatched, not as shared events. INTERPRETATION: There is no coincident set to interpret: no detection on one phone has a counterpart on the other inside the 1.5 s tolerance, so this run contains no evidence of a shared excitation. The correct reading is "nothing matched", not "nothing was looked at": 1 single-site detection(s) were examined and rejected as pairs. NOT ESTABLISHED: Coincidence within 1.5 s does not identify a shared source, a room of origin or the equipment involved, and it does not by itself show that the two phones recorded the same physical event; that needs a labelled or triggered test. SENSITIVITY: {"match_tolerance_s": 1.5, "loose_tolerance_s": 5.0} ! UTC clocks come from the two phones; the match tolerance absorbs any clock offset. ! A match means "within 1.5 s", not proof of a shared source. method: per-phone envelope detection, then matching of real detection index sets across the common UTC overlap with a 1.5 s tolerance Q2. Which events appeared only upstairs and which appeared only downstairs? [unsupported / confidence medium] limiting factor: no matched event pairs within the tolerance MEASURED: Upstairs: 0 detections, 0 of them unpaired upstairs-only. Downstairs: 1 detections, 0 unpaired downstairs-only. 0 pairs matched within 1.5 s. INTERPRETATION: The unmatched sets are the real set difference of the two match index sets, so an "only" event is one that rose above one phone's detector threshold and not above the other's at that moment. NOT ESTABLISHED: An unmatched detection does not establish that the source was local to that floor; it can be a genuinely local source, a site-specific sensitivity difference, or a shared event that stayed below the other phone's detector threshold. SENSITIVITY: {"match_tolerance_s": 1.5, "loose_tolerance_s": 5.0} ! Detection thresholds are per-site, so a weak shared event can look unpaired. Compare the per-pair ratios (Q4) before calling an event local. method: same detector configuration on both sites; unmatched detections are the complement of the matched index sets Q3. For simultaneous events, which phone recorded the larger acceleration amplitude? [unsupported / confidence low] limiting factor: no matched event pairs MEASURED: No matched event pair carries a usable peak amplitude on both floors, so there is no simultaneous event to compare. INTERPRETATION: Not applicable: there is no simultaneous pair, so no statement about the relative peak amplitude across the two floors can be made from this run. Reporting a "direction" here would require inventing a coincidence that the detector did not find. NOT ESTABLISHED: A peak amplitude is set by a single sample, by each unit's calibration and mounting, and by the event window; the direction therefore does not establish that the floor itself amplifies, and a magnitude claim needs a labelled on/off test with the phones swapped. SENSITIVITY: {"match_tolerance_s": 1.5, "loose_tolerance_s": 5.0} ! Peak amplitude is sensitive to a single sample; the RMS ratio in the same table is the more stable statistic. ! the phones are the same model but separate physical units; individual accelerometer calibration, mounting, cases, surface coupling and orientation may still differ method: peak |a| of the resultant acceleration inside each matched event window Q4. What is the upstairs/downstairs amplitude ratio for every significant event? [unsupported / confidence low] limiting factor: no matched event pairs MEASURED: Per-event ratios are tabulated for 0 matched pairs (peak and RMS upstairs/downstairs). Summary: median peak ratio not available, median RMS ratio not available, fraction above 1 not available. INTERPRETATION: Each matched pair has its own ratio, so the answer is a distribution rather than a single number; the spread shows how consistent the floor-to-floor relationship is. NOT ESTABLISHED: Ratios exist only for events that matched within 1.5 s and cleared both detectors, so events below either noise floor or outside the overlap are not represented; the ratios also carry each unit's calibration and mounting. SENSITIVITY: {"match_tolerance_s": 1.5, "loose_tolerance_s": 5.0} ! the phones are the same model but separate physical units; individual accelerometer calibration, mounting, cases, surface coupling and orientation may still differ method: peak_aT and rms_aT inside each matched event window, ratio taken as upstairs / downstairs Q5. Does the upper floor consistently amplify vibration relative to the first floor? [insufficient_data / confidence medium] limiting factor: fewer than 4 matched pairs MEASURED: Verdict: INSUFFICIENT MATCHED EVENTS. Only 0 matched event pairs carry an amplitude ratio on both floors; the decision rule needs at least 4. INTERPRETATION: The conclusion is taken from statistics only: the median peak and RMS ratios, the fraction of pairs above 1, the bootstrap confidence interval of the median and the two-sided sign test. "INSUFFICIENT MATCHED EVENTS" means the evidence is not strong enough to claim a direction, not that the floors are identical. NOT ESTABLISHED: Even a supported direction does not establish the magnitude of structural amplification, because the two phones are separate physical units; a magnitude claim needs a labelled on/off test (and ideally a phone swap) so unit sensitivity can be separated from the floor. SENSITIVITY: {"match_tolerance_s": 1.5, "loose_tolerance_s": 5.0} ! the phones are the same model but separate physical units; individual accelerometer calibration, mounting, cases, surface coupling and orientation may still differ ! Without a shared event there is no excitation to amplify, so all-time ratios would measure the phones rather than the building. method: matched-pair ratios only: median peak and RMS ratios, fraction above 1, bootstrap 95% CI of the median (util.bootstrap_ci) and two-sided sign test (util.sign_test_p); fewer than 4 pairs returns INSUFFICIENT MATCHED EVENTS Q6. Are there particular vibration frequencies that are substantially stronger upstairs? [tentative / confidence medium-high] limiting factor: frequency resolution MEASURED: Median upstairs/downstairs PSD ratio 0.548 (-2.62 dB). Strongest upstairs bins (gated on both sites being above their own floor): 0.2 Hz (x1.04); 4.27 Hz (x1.03); 3.2 Hz (x1.01); 1.1 Hz (x1.01); 2.83 Hz (x1). Strongest downstairs bins for contrast: 22.7 Hz (x0.5); 22.3 Hz (x0.52); 22.83 Hz (x0.54); 21.73 Hz (x0.54); 20.73 Hz (x0.54). Gated regions with |median gain| > 3 dB: not available. INTERPRETATION: The continuous comparison is the useful part: the upstairs channel carries about 0.55x the downstairs power across the whole usable band (-2.62 dB), and the strongest upstairs-dominant bin reaches only about 1.04x. A bin a few percent above equality is a candidate frequency-dependent difference; with 0 gated region(s) above 3 dB it is not a structural amplification signature. NOT ESTABLISHED: A bin ratio above 1 does not establish that the upper floor responds more strongly at that frequency: a flat ratio offset is a unit sensitivity and mounting difference, and a genuine structural peak cannot be separated from a phone/surface coupling resonance without a swap test or a labelled source. SENSITIVITY: {"noise_floor_gating": null, "n_segments": 660.0} ! the phones are the same model but separate physical units; individual accelerometer calibration, mounting, cases, surface coupling and orientation may still differ ! A ratio computed at a bin where either site is at its noise floor is not a measurement. method: Welch PSD of the signed/derived channels on the common grid over the overlap, ratio of medians per frequency bin where both sites are above their noise floor Q7. What are the dominant resonant frequencies measured at each location? [tentative / confidence medium] limiting factor: no analysed upstairs/downstairs session MEASURED: upstairs: dominant not available (source csv/multi_1790959520.160099.csv; top peaks not available) downstairs: dominant not available (source csv/multi_1790959615.928701.csv; top peaks not available) INTERPRETATION: The dominant frequency is the strongest PSD peak of each site's own recording; it is the frequency that carries most energy at that location. NOT ESTABLISHED: The dominant peak is often a phone/surface contact resonance rather than a building mode, so the measurement does not establish a structural resonant frequency at either location. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! With a phone lying on a surface, the dominant peak is often a phone/surface contact resonance rather than a building mode. ! the phones are the same model but separate physical units; individual accelerometer calibration, mounting, cases, surface coupling and orientation may still differ method: Welch PSD of the resultant acceleration with the recorded sample rate; peaks ranked by power, restricted to the usable band Q8. Are the dominant frequencies the same upstairs and downstairs? [tentative / confidence medium] limiting factor: no analysed upstairs/downstairs session MEASURED: Dominant frequency unavailable at one or both locations (upstairs not available, downstairs not available). INTERPRETATION: Comparing the ranked peak lists says whether the two floors peak at the same frequency bin or at different frequencies; agreement in one bin is weak evidence for a shared mode. NOT ESTABLISHED: A shared physical mode can appear at slightly different bins because the two units have separate calibration and sensor bandwidth, and a single dominant bin cannot establish that a common structural mode exists. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! the phones are the same model but separate physical units; individual accelerometer calibration, mounting, cases, surface coupling and orientation may still differ method: comparison of the ranked PSD peak lists (Q7) at the common frequency resolution Q9. Are there frequencies that appear downstairs but disappear upstairs? [tentative / confidence medium] limiting factor: sensor noise floor MEASURED: Downstairs-dominant bins (upstairs/downstairs < 1): 22.7 Hz (x0.5); 22.3 Hz (x0.52); 22.83 Hz (x0.54); 21.73 Hz (x0.54); 20.73 Hz (x0.54); 19.73 Hz (x0.55); 22.6 Hz (x0.55); 22.27 Hz (x0.55); 22 Hz (x0.55); 21.57 Hz (x0.55). Median ratio 0.548; 5th-95th percentile of the ratio 0.05 to 0.93. INTERPRETATION: Bins where the downstairs PSD exceeds the upstairs one are candidate downstairs-dominant frequencies; the median ratio (0.55) and the percentile spread are the robust part of this comparison, while individual bins scatter by a factor of several on a noise floor and are not "frequencies that disappear upstairs". NOT ESTABLISHED: "Disappear" here only means "weaker at the same bin", which is within the sensitivity spread of two separate units; a shared band can fall below one unit's noise floor, so a downstairs-only source is not established. SENSITIVITY: {"noise_floor_gated": null} ! the phones are the same model but separate physical units; individual accelerometer calibration, mounting, cases, surface coupling and orientation may still differ ! A ratio at the noise floor is not a measurement. method: ratio of the two Welch PSDs on the common grid; bins where downstairs exceeds upstairs Q10. Are there frequencies that are weak downstairs but become strong upstairs? [tentative / confidence medium] limiting factor: frequency resolution MEASURED: Upstairs-dominant bins (upstairs/downstairs > 1, both sites above their floor): 0.2 Hz (x1.04); 4.27 Hz (x1.03); 3.2 Hz (x1.01); 1.1 Hz (x1.01); 2.83 Hz (x1). Median ratio 0.548 (-2.62 dB). Gated regions above 3 dB: not available. INTERPRETATION: A frequency that is weak downstairs and strong upstairs would be the pattern expected from transmission up through the structure plus a floor-level response, so these bins are candidate frequency-dependent differences. Individual-bin verdict for this run: no bin survives FDR control. Even so, none of these bins is established as a transmission path until a controlled experiment (labelled source or phone swap) reproduces it. NOT ESTABLISHED: The same bin pattern can be produced by phone coupling, mounting or unit calibration rather than the structure, so a structural transmission path is not established from the ratio alone. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! the phones are the same model but separate physical units; individual accelerometer calibration, mounting, cases, surface coupling and orientation may still differ method: same ratio analysis as Q6/Q9, ranked from the upstairs-dominant side Q11. Can the time delay between corresponding vibration peaks at the two phones be measured? [insufficient_data / confidence medium-high for "measurable"; low for "propagation"] limiting factor: no matched events and no coherent phase MEASURED: Not available: no matched event pair and no resolvable cross-spectrum phase slope, so no delay estimate exists. INTERPRETATION: A time difference between corresponding peaks is measurable whenever matched events exist and the shared timing resolution is finer than the offset; it is the difference of the two phone clocks plus any propagation delay. NOT ESTABLISHED: The measured delay cannot be attributed to vibration propagation through the structure because the two phone clocks are not synchronised: clock offset and propagation delay cannot be separated without a common clock or trigger. SENSITIVITY: {"resolution_s": 0.01, "coh_min": 0.5} ! The resolution is set by the common sample rate and the segment length, not assumed. method: matched-event waveform cross-correlation and the cross-spectrum phase slope inside the coherent band, both on the common grid Q12. Is that delay consistent enough to suggest vibration propagation through the structure? [insufficient_data / confidence low] limiting factor: no matched events and no coherent band MEASURED: Not available: no matched events and no coherent band, so the consistency of the delay cannot be assessed. INTERPRETATION: A tight delay distribution across independent events is what a single physical transmission path would produce, whereas a wide or unresolvable spread argues against reading the delay as propagation. NOT ESTABLISHED: Consistency of the delay does not establish propagation through the structure: a common clock offset alone would produce a consistent delay, so propagation needs a synchronised or triggered recording. SENSITIVITY: {"resolution_s": 0.01} ! A single-house delay is far below one sample at the common rate; the claim is bounded by the timing resolution, not measured below it. method: dispersion of the per-event delay estimates plus the cross-spectrum phase-slope confidence interval Q13. Does the apparent delay change depending on vibration frequency? [insufficient_data / confidence low] limiting factor: too few events with both delay and dominant frequency MEASURED: Not available: too few events carry both a delay estimate and a dominant frequency. Phase-slope result: {"available": false, "clock_correction_applied": false, "coh_min": 0.5, "delay_meaning": "delay_s is the cross-spectrum phase delay measured on the UNCORRECTED grid, so it includes any clock offset and is only a propagation time if the clocks are independently known; delay_after_clock_fit_s is the residual left by the clock model", "measured_before_clock_correction": false, "method": "linear fit of unwrapped cross-spectrum phase inside one contiguous coherent band", "note": "no contiguous band reached the coherence threshold", "raw_uncorrected": {"available": false, "band_coherence_median": null, "delay_ms": null, "delay_s": null, "f_hi_hz": null, "f_lo_hz": null, "n_bins": null, "note": "no contiguous band reached the coherence threshold", "r2": null}, "resolvable": false} INTERPRETATION: Splitting the per-event delay by the event dominant frequency tests whether the delay is frequency dependent, which is what dispersive propagation would look like. NOT ESTABLISHED: With only a handful of usable events, any trend is weakly constrained and no dispersion curve can be established; the phase slope is fitted only inside the coherent band. SENSITIVITY: {"match_tolerance_s": 1.5, "loose_tolerance_s": 5.0} ! Frequency dependence needs many more events than this dataset provides. method: per-event delay split at the median dominant frequency, plus the cross-spectrum phase slope Q14. Does one axis—X, Y, or Z—consistently respond more strongly than the others? [tentative / confidence medium-high] limiting factor: single recording per site (first recording used) MEASURED: upstairs: body-frame RMS ax 0.00698, ay 0.00688, az 0.00712 (shares ax 0.332, ay 0.323, az 0.345; largest az, spread 0.00024); gravity-aligned vertical_signed 0.00985 m/s^2, horizontal 0.00992 m/s^2, resultant 0.01398 m/s^2 downstairs: body-frame RMS ax 0.02474, ay 0.02611, az 0.03204 (shares ax 0.264, ay 0.294, az 0.442; largest az, spread 0.0073); gravity-aligned vertical_signed 0.03237 m/s^2, horizontal 0.03599 m/s^2, resultant 0.0484 m/s^2 - Only the gravity-aligned vertical projection, the horizontal energy and the resultant are physically comparable between the two phones; the body-frame x/y/z axes point in different directions on the two units and are reported per phone only. Near-equal body-frame shares are what isotropic sensor noise looks like, while a real vibration source would usually make the gravity-aligned vertical stand out. INTERPRETATION: Comparing the gravity-aligned vertical projection, the horizontal energy and the resultant says how the vibration is distributed; the body-frame x/y/z shares are reported per phone because those axes point in different physical directions on the two units. NOT ESTABLISHED: Axis names follow the phone body, not the building, and the two units were not mounted in a known common frame, so the axis split cannot be attributed to structural directions. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! Axis naming follows the phone body, not the building; the two phones were not mounted in the same orientation, so a cross-phone body-axis comparison is not physically meaningful. ! Near-equal shares (about 0.33 each) mean no body axis stands out; that is the expected result for isotropic sensor noise. ! the phones are the same model but separate physical units; individual accelerometer calibration, mounting, cases, surface coupling and orientation may still differ method: per-axis AC RMS and its share of total AC energy with gravity removed, plus the gravity-aligned products (aVert, aHoriz, aT) that are comparable across the two phones Q15. Are upstairs/downstairs differences predominantly vertical vibration or horizontal vibration? [tentative / confidence medium] limiting factor: unknown relative orientation of the two units MEASURED: Upstairs/downstairs RMS ratio by direction: {"ax": 0.2823, "ay": 0.2636, "az": 0.2222, "horizontal": 0.2756, "resultant": 0.2887, "vertical_signed": 0.3042}. Comparable directions: ["vertical_signed", "horizontal", "resultant"]. INTERPRETATION: The per-direction ratios show whether the floor-to-floor difference is carried mainly by the gravity-aligned vertical projection or by the horizontal/resultant energy. NOT ESTABLISHED: The vertical/horizontal attribution is only physically meaningful for the gravity-aligned products; the body-frame axes point in different directions on the two units, so a vertical-versus-horizontal split beyond those products is not established. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! Check the orientation comparison (Q49) before trusting the axis split. ! the phones are the same model but separate physical units; individual accelerometer calibration, mounting, cases, surface coupling and orientation may still differ method: AC RMS ratio per axis on the common grid, plus the same ratio restricted to the loudest windows Q16. What is the resultant 3-axis acceleration magnitude during each event rather than looking at individual axes? [supported / confidence high] limiting factor: event detector threshold MEASURED: The resultant |a| = sqrt(ax^2+ay^2+az^2) is the primary detection channel on both phones: 0 upstairs and 1 downstairs events carry peak_aT, rms_aT and duration. Upstairs median event duration not available, downstairs 1.04 s; upstairs median SNR not available, downstairs 15.17 dB. INTERPRETATION: Using the resultant magnitude removes the orientation problem for detection and for event amplitude, because it is invariant to how the phone is lying on the surface. NOT ESTABLISHED: The resultant carries no direction, so it cannot establish which axis or which direction the vibration came from, nor whether an event was translational. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! The resultant mixes axes, so it is not a substitute for the signed channels in spectral work. method: derived resultant channel used throughout detection and event metrics Q17. What is the normal background vibration level at each phone? [supported / confidence high] limiting factor: single recording per site (first recording used) MEASURED: upstairs: envelope median 0.00416 m/s^2, quiet 1 s RMS p10 0.00467 m/s^2, active p90 0.00572, AC sigma 0.00529 m/s^2, axis shares {"ax": {"p99_abs_ms2": 0.017945436, "rms_ms2": 0.006984805, "share_of_energy": 0.33225, "std_ms2": 0.006984805}, "ay": {"p99_abs_ms2": 0.017494885, "rms_ms2": 0.006882686, "share_of_energy": 0.32261, "std_ms2": 0.006882686}, "az": {"p99_abs_ms2": 0.018278595, "rms_ms2": 0.00711906, "share_of_energy": 0.34514, "std_ms2": 0.00711906}, "horizontal": {"rms_ms2": 0.009918647, "std_ms2": 0.004658011}, "resultant": {"rms_ms2": 0.013976471, "std_ms2": 0.005422275}, "vertical_signed": {"rms_ms2": 0.009846937, "std_ms2": 0.007119233}} downstairs: envelope median 0.00399 m/s^2, quiet 1 s RMS p10 0.00432 m/s^2, active p90 0.00599, AC sigma 0.02794 m/s^2, axis shares {"ax": {"p99_abs_ms2": 0.017715754, "rms_ms2": 0.024739499, "share_of_energy": 0.26378, "std_ms2": 0.024739499}, "ay": {"p99_abs_ms2": 0.017279646, "rms_ms2": 0.026106014, "share_of_energy": 0.29373, "std_ms2": 0.026106014}, "az": {"p99_abs_ms2": 0.021309284, "rms_ms2": 0.032042004, "share_of_energy": 0.44249, "std_ms2": 0.032042004}, "horizontal": {"rms_ms2": 0.035988394, "std_ms2": 0.034872826}, "resultant": {"rms_ms2": 0.048403851, "std_ms2": 0.046657515}, "vertical_signed": {"rms_ms2": 0.03236925, "std_ms2": 0.032025367}} Watch the difference between the plain RMS and the robust background - downstairs: plain RMS is dominated by rare samples, not by continuous vibration: for aT the top 0.01% of samples carry 97.71% of the total energy, so the plain RMS 0.046668 m/s^2 collapses to 0.005429 m/s^2 once samples above the 99.9th percentile are removed (8.6x), while the robust sigma is 0.003149 m/s^2 and the 1 s background is the right measure of continuous motion INTERPRETATION: These are single-site measurements: each number is the quiet floor of that one phone in that one recording, which is the baseline every event at that phone is measured against. Where a plain full-record RMS is much larger than this background it is set by a few rare samples (handling, start/stop artefacts), not by continuous vibration: the trimmed RMS and the robust sigma in the evidence are the right comparison. NOT ESTABLISHED: A single-site background does not establish how quiet the other floor is, nor how much of the floor is unit self-noise; the two separate physical units have their own noise floors and calibration. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! the phones are the same model but separate physical units; individual accelerometer calibration, mounting, cases, surface coupling and orientation may still differ method: moving-average envelope of the high-passed resultant; the p10 of the 1 s RMS is the quiet floor Q18. How far above background noise is each detected event? [supported / confidence high] limiting factor: sensor noise floor MEASURED: Each catalog entry carries snr_db: upstairs median not available (p10 not available, p90 not available, n=0); downstairs median 15.17 dB (p10 not available, p90 not available, n=1). INTERPRETATION: SNR is the event peak envelope divided by that same site's quiet floor, so it says how far each event stands above its own measurement noise. NOT ESTABLISHED: SNR is relative to each site's own floor, so a higher SNR on one phone does not mean a physically stronger event there; it can reflect a lower unit noise floor. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! the phones are the same model but separate physical units; individual accelerometer calibration, mounting, cases, surface coupling and orientation may still differ ! The SNR floor is set by the envelope smoothing and the quiet-window definition. method: event peak envelope divided by the median quiet RMS at the same site Q19. How does background vibration change by hour of the day? [tentative / confidence medium] limiting factor: elapsed-hour indexing, no shared wall clock MEASURED: Median 1 s RMS per elapsed hour: {"downstairs": {"h00": {"n_s": 3601, "rms_max_ms2": 0.006303041, "rms_median_ms2": 0.00473078}, "h01": {"n_s": 3600, "rms_max_ms2": 0.005330183, "rms_median_ms2": 0.003982258}, "h02": {"n_s": 3600, "rms_max_ms2": 0.004925724, "rms_median_ms2": 0.003776333}, "h03": {"n_s": 2704, "rms_max_ms2": 1.820514833, "rms_median_ms2": 0.003777069}}, "upstairs": {"h00": {"n_s": 3601, "rms_max_ms2": 0.007968656, "rms_median_ms2": 0.004136495}, "h01": {"n_s": 3600, "rms_max_ms2": 0.006001336, "rms_median_ms2": 0.004300848}, "h02": {"n_s": 3600, "rms_max_ms2": 0.005820915, "rms_median_ms2": 0.004216883}, "h03": {"n_s": 3600, "rms_max_ms2": 0.005954715, "rms_median_ms2": 0.004230024}, "h04": {"n_s": 741, "rms_max_ms2": 0.015537746, "rms_median_ms2": 0.004118665}}}. dB-channel hourly medians: not available. INTERPRETATION: The hourly profile shows how the quiet/standing level moves during each recording, which is the background against which events stand out. NOT ESTABLISHED: The two recordings do not cover the same absolute hours and the hour indices are elapsed from each recording's own start, so a shared occupancy or equipment schedule is not established. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! Elapsed hours, not wall-clock hours; the profiles are not directly comparable between the two recordings. method: median 1 s RMS per elapsed hour at each site (site analysis when present, otherwise computed from the recording) Q20. Are there repeating vibration patterns at regular intervals such as every few seconds, minutes, or hours? [tentative / confidence medium] limiting factor: autocorrelation of a finite envelope MEASURED: Candidate repetition intervals from the envelope autocorrelation: {"downstairs": [], "upstairs": [{"autocorrelation": 0.0266, "lag_minutes": 0.12, "lag_s": 7.25}, {"autocorrelation": 0.0255, "lag_minutes": 0.15, "lag_s": 9.0}, {"autocorrelation": 0.0252, "lag_minutes": 0.17, "lag_s": 10.5}, {"autocorrelation": 0.025, "lag_minutes": 0.09, "lag_s": 5.25}, {"autocorrelation": 0.0247, "lag_minutes": 0.31, "lag_s": 18.5}, {"autocorrelation": 0.0245, "lag_minutes": 0.26, "lag_s": 15.5}, {"autocorrelation": 0.0244, "lag_minutes": 2.11, "lag_s": 126.75}, {"autocorrelation": 0.0243, "lag_minutes": 0.57, "lag_s": 34.25}]} INTERPRETATION: Peaks in the envelope autocorrelation mark lags at which the vibration level tends to repeat, which is the signature of a cyclic machine or a repeating activity. NOT ESTABLISHED: An autocorrelation peak is a candidate period, not proof of a machine cycle, and it cannot establish which equipment (if any) is running on that period. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! The candidate list is capped at the strongest few lags; a weak but real period can be missed. method: autocorrelation of the RMS envelope with the mean removed; peaks at lag L indicate repetition every L seconds Q21. Are any periodic signatures consistent with HVAC equipment, compressors, pumps, fans, appliances, traffic, or other machinery? [tentative / confidence low (hypothesis)] limiting factor: sensor noise floor MEASURED: Persistent narrowband lines above the local noise floor: upstairs: {"available": true, "detection_threshold_psd": 1.860412e-06, "fmax_safe_hz": 45.0, "interpretation_caveat": "these are band-based hypotheses for equipment classes, not identifications of specific devices; a definitive call needs a labelled on/off test", "lines": [], "noise_floor_median_psd": 1.064679e-06, "noise_floor_sigma_psd": 9.9467e-08} | downstairs: {"available": true, "detection_threshold_psd": 2.004711e-06, "fmax_safe_hz": 45.0, "interpretation_caveat": "these are band-based hypotheses for equipment classes, not identifications of specific devices; a definitive call needs a labelled on/off test", "lines": [], "noise_floor_median_psd": 1.190222e-06, "noise_floor_sigma_psd": 1.01811e-07} INTERPRETATION: Mapping persistent lines onto equipment frequency bands gives hypotheses about the class of machine (fan, compressor, pump, appliance, traffic) that could produce them. NOT ESTABLISHED: These are band-based hypotheses for equipment classes, not identifications of specific devices, and a definitive call needs a labelled on/off test of each candidate. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! This is a band-based hypothesis, not an identification. A labelled on/off test of each appliance is the only way to confirm. method: persistent narrowband PSD lines above the local noise floor, mapped to equipment frequency bands Q22. Can individual vibration events be clustered automatically into recurring "event types" based on their waveform and frequency spectrum? [insufficient_data / confidence medium where silhouette > 0.25, otherwise weak] limiting factor: few events per site MEASURED: upstairs: k=not available silhouette=not available over 0 events downstairs: k=not available silhouette=not available over 0 events INTERPRETATION: Unsupervised clustering of the per-event band profile groups events with similar spectral shape, which is the practical way to define recurring event types without hand labelling. NOT ESTABLISHED: Cluster labels are descriptive, not physical: the analysis does not establish that a cluster corresponds to one piece of equipment, and overlapping clusters leave individual events ambiguous. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! Cluster labels are descriptive, not physical. Overlapping distributions mean some events are genuinely ambiguous. method: k-means on the band profile of every detected event; k chosen by silhouette Q23. How many distinct recurring event signatures exist in the dataset? [tentative / confidence medium] limiting factor: few events per site MEASURED: upstairs: not attempted (0 scored events; at least 10 are needed for a meaningful k-means split) downstairs: not attempted (0 scored events; at least 10 are needed for a meaningful k-means split) INTERPRETATION: The count is the number of silhouette-selected groups in the per-event band-profile clustering, which is the dataset's best estimate of how many recurring signatures exist. NOT ESTABLISHED: The number is a model choice constrained by the k search, not a physical count of sources, and it cannot establish what any signature is. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! The k-search table shows how flat the objective is; a flat objective means the count is not sharp. ! With only a handful of detections no clustering is attempted. method: silhouette-selected k per site over the searched range Q24. At what times do each of those event types normally occur? [insufficient_data / confidence low] limiting factor: no event types derived MEASURED: No event types were derived, so there are no type times to report. Individual detection times are in Q2/Q16. INTERPRETATION: The per-cluster hour histogram shows when each recurring signature tends to occur, which is the first step towards tying a signature to a schedule. NOT ESTABLISHED: Elapsed hours are not wall-clock hours, and with few events per cluster the time-of-day pattern of any type is not established. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! Elapsed hours, not wall-clock hours. method: hour-of-elapsed-recording histogram per cluster Q25. Which recurring event types are detected by both phones? [insufficient_data / confidence low] limiting factor: no matched event pairs MEASURED: None: no event pairs matched across the phones (Q1), so by construction no event type is detected by both. Upstairs produced 0 detections and downstairs 1, none of them sharing an onset within 1.5 s. INTERPRETATION: Event types are compared only through events that matched across the phones, so a type seen on both floors means matched physical detections, not merely similar-looking spectra. NOT ESTABLISHED: Cluster identities are derived independently at each site, so a shared type is not established unless matched events carry the same signatures, and it still would not identify the source. SENSITIVITY: {"match_tolerance_s": 1.5, "loose_tolerance_s": 5.0} ! Cluster ids are site-local; only matched onsets justify comparing them. method: matched events attached to their cluster ids at both sites Q26. Does the waveform of the same event change between downstairs and upstairs? [insufficient_data / confidence low] limiting factor: no matched event pairs MEASURED: Not answerable: zero event pairs matched across the phones (Q1), so there is no shared event whose waveform could be compared. INTERPRETATION: For a matched event the peak, RMS and dominant frequency can be compared between the two floors, which is how the observed waveform changes when the same excitation is seen upstairs and downstairs. NOT ESTABLISHED: A change in the observed waveform does not establish a change in the structure's response: the two units have separate calibration, sensor bandwidth and mounting, any of which changes the recorded waveform for an identical input. SENSITIVITY: {"match_tolerance_s": 1.5, "loose_tolerance_s": 5.0} ! the phones are the same model but separate physical units; individual accelerometer calibration, mounting, cases, surface coupling and orientation may still differ method: both sites on the common grid inside each matched window, comparing peak, RMS and dominant frequency Q27. How correlated are the two phones' acceleration signals during quiet periods? [tentative / confidence high for the number; the window definition is a caveat] limiting factor: sensor noise floor MEASURED: Median cross-phone correlation in the quiet windows: -0.0025. Window classification: "windows sorted by combined acceleration RMS; quiet = lowest 20 %, active = highest 20 %, normal = the rest"; quiet window count 7. INTERPRETATION: Quiet windows are selected by vibration energy of the common grid (lowest band of the RMS distribution), never by correlation, so the reported correlation is an independent measurement of whether the two floors share a signal when nothing is happening. NOT ESTABLISHED: A correlation near zero in quiet windows does not prove the absence of a shared source; it only shows no shared signal above the two noise floors, and the result depends on the chosen window length. SENSITIVITY: {"window_classification": "windows sorted by combined acceleration RMS; quiet = lowest 20 %, active = highest 20 %, normal = the rest"} ! The quiet class is defined by an energy threshold, so a very quiet recording could have too few quiet windows to be representative. method: Pearson correlation of the high-passed channels on the common grid in windows whose class comes from the vibration-energy distribution Q28. How correlated are they during major vibration events? [tentative / confidence medium-high] limiting factor: sensor noise floor MEASURED: Median cross-phone correlation in the active windows: 0; event window correlation: not available; active window count 7. INTERPRETATION: The active class is the highest-energy band of the common-grid RMS distribution; rising correlation there, relative to the quiet windows, is the positive control for a shared signal. NOT ESTABLISHED: Correlation during loud windows is bounded by clock alignment, the noise floor and the resampling, so it does not establish that the loud energy came from one shared source. SENSITIVITY: {"window_classification": "windows sorted by combined acceleration RMS; quiet = lowest 20 %, active = highest 20 %, normal = the rest"} ! The active class mixes genuine vibration with non-stationary transients. method: same windowed correlation as Q27; the active class is the highest-energy band, plus peak cross-correlation per matched event Q29. At what time offset is cross-correlation between the phones maximized? [unsupported / confidence high for the estimate; low for attributing it to propagation] limiting factor: timing resolution MEASURED: Raw cross-correlation peak at lag 3.29843 s with r=0.01367 (variance explained 0.019%). No clock correction was applied: the offset between the two phones could not be estimated from this data, so the raw lag stands as-is and is not a propagation time. The peak is not physically meaningful: |r| < 0.05 means the alignment explains less than 0.25% of the variance, and any lag can win on a noise floor. INTERPRETATION: The lag at which the whole-record cross-correlation peaks is the best global alignment of the two signals; removing the clock offset estimated from the shared timeline shows how much of that lag was clock rather than signal. NOT ESTABLISHED: The peak lag is dominated by the phone clock offset and by the non-stationary nature of the signal, so it is not a propagation time and the residual is not established as a source delay. SENSITIVITY: {"match_tolerance_s": 1.5, "loose_tolerance_s": 5.0} ! The peak lag is dominated by the phone clock offset (tens of ms) and by the non-stationary nature of the signal. ! A correlation peak on a noise floor has no physical meaning unless the coefficient is large enough to explain a useful share of the variance. method: whole-record cross-correlation before and after applying the timeline-derived clock offset Q30. Does that optimal time offset remain stable throughout the recording? [tentative / confidence medium] limiting factor: timing resolution MEASURED: Clock model: offset 3.29843 s, slope 0 s/s, drift 0 s (0 ppm), method not available, anchors 0, CI not available, applied=False. Window correlation by third: "not available". INTERPRETATION: Estimating both an offset and a slope from the shared timeline shows whether a single constant offset is enough or whether the two clocks also drift apart during the recording. NOT ESTABLISHED: Residual variation between the halves of the recording could come from clock drift or from event clustering, and without a common reference clock the drift model itself is not established. SENSITIVITY: {"n_anchors": 0, "method": "cross-correlation peak lag (no matched events)"} ! The correction uses shift and trim, never a wraparound shift, so no samples from the far end of a recording reappear at the other end. method: shift-and-trim clock correction with a robust offset/slope fit over the shared timeline; stability checked by the per-third correlation Q31. Can spectral coherence show frequencies where both phones are responding to the same physical source? [tentative / confidence medium-high] limiting factor: sensor noise floor MEASURED: Over 660 Welch segments the robust result is the central tendency: median coherence 0.001 and 90th percentile 0.0037, with only 5.85% of bins above the single-bin 0.0045 significance threshold. Those 76 bins are a multiple-comparison artefact: with 1351 bins tested, about 67.5 exceed the threshold by chance alone, and after Benjamini-Hochberg FDR control 0 bin(s) pass (q_min 0.1766). The maximum coherence 0.0135 (42.1 Hz) is then checked family-wise: the segment-permutation surrogate p for the maximum is 0.204. Combined verdict on individual bins: no bin survives FDR control. INTERPRETATION: Magnitude-squared coherence is the standard test for a linear shared source between two channels. The defensible reading is the one that does not cherry-pick a bin: with 1351 bins tested, individual crossings of the per-bin threshold are expected by chance, so the headline is median coherence 0.001 against a 0.0045 significance level - and, where that holds, individual bins may only be highlighted after FDR control or a family-wise surrogate test on the maximum. NOT ESTABLISHED: Coherence at the significance level does not establish that no shared source exists: a shared source below either noise floor is invisible to this test, and high coherence with low PSD is not physically meaningful. SENSITIVITY: {"significance_threshold": 0.00454, "n_segs": 660.0} ! Coherence needs both signals to carry energy at that frequency. method: magnitude-squared coherence from Welch segments with 50% overlap, restricted to the usable band Q32. At which frequencies is coherence between upstairs and downstairs highest? [tentative / confidence medium] limiting factor: frequency resolution MEASURED: Robust view first: median coherence 0.001, p90 0.0037, and per-band mean coherence {"0.25-1 Hz": {"fraction_above_significance": 0.0, "median_coherence": 0.00077, "n_bins": 22}, "1-4 Hz": {"fraction_above_significance": 0.0667, "median_coherence": 0.00108, "n_bins": 90}, "12-32 Hz": {"fraction_above_significance": 0.05, "median_coherence": 0.001, "n_bins": 600}, "32-45 Hz": {"fraction_above_significance": 0.0692, "median_coherence": 0.00108, "n_bins": 390}, "4-12 Hz": {"fraction_above_significance": 0.0667, "median_coherence": 0.00114, "n_bins": 240}}. The individually highest bins are 42.1 Hz; 42.8 Hz; 9.37 Hz; 9.2 Hz; 14.9 Hz; 15.83 Hz; 34.5 Hz; 43.13 Hz - but none of them is established: no bin survives FDR control INTERPRETATION: Ranking the coherence curve shows where the two phones are most tightly related, but ranked bins are adjacent, correlated and drawn from a thousand-bin search, so the band means and the median are the defensible view. A bin may only be called a shared frequency if it survives multiple-comparison control or the family-wise surrogate test on the maximum coherence. NOT ESTABLISHED: A high coherence value alone does not establish that a physical source exists there, and a bin that does not survive FDR control is not evidence at all: the band means are the robust view. SENSITIVITY: {"significance_threshold": 0.00454, "n_segs": 660.0} ! Read the band means for a robust view rather than single bins. method: rank coherence by frequency bin inside the usable band Q33. Are there frequencies with strong energy but very low cross-phone coherence, suggesting highly localized vibration? [tentative / confidence medium] limiting factor: sensor noise floor MEASURED: 12 low-coherence bins above 1 Hz (coherence < 0.2) with energy at both sites. Upstairs/downstairs PSD ratio median 0.548 (5th-95th percentile 0.05 to 0.93), i.e. not flat. INTERPRETATION: A frequency with strong local energy but very low cross-phone coherence would be the signature of vibration that does not reach the other phone - a localized source. NOT ESTABLISHED: Low coherence can also mean the signal is below the coherent detection floor, so a single low-coherence bin does not establish that the vibration is localized. SENSITIVITY: {"significance_threshold": 0.00454, "n_segs": 660.0} ! Pair this with the SNR check (Q18): low coherence at low energy is not evidence of localization. method: combine each site PSD with the cross coherence on the same frequency grid Q34. Can wavelet analysis reveal short-lived vibration bursts that a normal FFT would hide? [tentative / confidence medium-high] limiting factor: envelope smoothing window MEASURED: 0 upstairs and 1 downstairs events were found from the time-domain envelope, with a median duration of 1.04 s; the shortest events are 1.04 s. Duration quantiles: upstairs "not available", downstairs {"p0": 1.04, "p1": 1.04, "p100": 1.04, "p25": 1.04, "p5": 1.04, "p50": 1.04, "p75": 1.04, "p95": 1.04, "p99": 1.04}. INTERPRETATION: Short-lived bursts are recovered from the time-domain envelope and given their own spectrum per event, so a whole-recording FFT is not the only available view of transients. NOT ESTABLISHED: This pipeline does not run a continuous wavelet transform, so frequency drift inside a longer event and structure below the envelope smoothing scale are not established. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! A CWT would localize the frequency drift inside longer events, which this pipeline does not attempt. method: time-domain envelope detection rather than a CWT; per-event FFT for the spectral content of each burst Q35. How long does each vibration event last? [supported / confidence high] limiting factor: event detector threshold MEASURED: Median duration upstairs not available, downstairs 1.04 s. Upstairs quantiles "not available"; downstairs quantiles {"p0": 1.04, "p1": 1.04, "p100": 1.04, "p25": 1.04, "p5": 1.04, "p50": 1.04, "p75": 1.04, "p95": 1.04, "p99": 1.04}. INTERPRETATION: Duration is the time between the envelope crossing at 20% of the event peak either side of the peak, so it measures how long each event stands above its own detector scale. NOT ESTABLISHED: Duration depends on the detector threshold and the envelope smoothing, so an absolute physical duration of an event, and any difference between floors that is smaller than that dependence, is not established. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! Duration depends on the threshold definition. method: envelope crossing at 20% of the event peak Q36. How rapidly does vibration decay after a strong impulse at each location? [insufficient_data / confidence low] limiting factor: no resolvable decays MEASURED: Decay time constant tau (envelope falling to 1/e): upstairs not available, downstairs not available. Upstairs quantiles "not available"; downstairs quantiles "not available". INTERPRETATION: tau measures how quickly the envelope of each site falls after an impulse, which is the raw quantity behind ringing and damping comparisons. NOT ESTABLISHED: The envelope smoothing sets a resolution floor, so decay constants shorter than that floor cannot be resolved, and the measured tau mixes the surface, the phone mounting and the unit, so a structural decay rate is not established. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! The envelope smoothing sets the resolution floor; decay constants below it cannot be resolved. method: exponential fit to the smoothed envelope after the event peak Q37. Does the upstairs signal "ring" longer after an impulse than the downstairs signal? [insufficient_data / confidence low] limiting factor: no resolvable decays at one site MEASURED: not enough events with a resolvable decay at both sites INTERPRETATION: Comparing the per-event decay constants says which floor keeps ringing after an impulse; a longer tau upstairs would be consistent with a floor-level mode with less damping. NOT ESTABLISHED: The comparison is not paired and the two floors have different numbers of usable events, so a site difference in ringing is not established from this sample, and ring time still mixes the surface and the phone coupling. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! Downstairs may have far fewer usable events, so its tau distribution is weakly constrained. method: compare the per-event decay constants; the paired comparison uses only matched events Q38. Can the decay curve be used to estimate damping characteristics at the two locations? [insufficient_data / confidence low] limiting factor: no events with both tau and dominant frequency MEASURED: Damping ratio zeta from tau and the event dominant period (zeta = 1/sqrt(1+(2*pi*f*tau)^2)): "not available" INTERPRETATION: The decay curve gives tau directly, and combining tau with the dominant frequency gives a damping ratio and a Q for the mode that is ringing. NOT ESTABLISHED: The estimate assumes a single lightly damped mode and a pure impulse; events are not pure impulses, the surface is not a single-degree-of-freedom system and the phone case adds damping, so the structural damping of either floor is not established. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! Treat as an order of magnitude only. method: envelope decay constant plus dominant frequency, with the rounding rule zeta = 1/sqrt(1+(2*pi*f*tau)^2) Q39. Do strong acceleration events coincide with changes in the gyroscope readings? [tentative / confidence medium-high] limiting factor: gyro/accelerometer coupling MEASURED: upstairs: |w| max 0.2241 rad/s (median 0.004919), handling 0 s over 0 periods; not available detected events carry gyro activity above 0.5 rad/s; per-event peak |w|/|a| median not available over 0 events (not available rotation-dominated) downstairs: |w| max 4.8647 rad/s (median 0.003064), handling 1.45 s over 4 periods; 0 detected events carry gyro activity above 0.5 rad/s; per-event peak |w|/|a| median 0.194 over 1 events (0% rotation-dominated) INTERPRETATION: The per-event ratio of peak |w| to peak |a|, and the count of detected events whose peak rotation rate exceeds the handling threshold, show how often an acceleration event also involved rotation of the phone. NOT ESTABLISHED: Gyro and accelerometer are physically coupled on a rocking phone, so a rotation-dominated event does not establish that the vibration source was rotational rather than translation with the phone rocking on its surface. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! A phone free on a surface does rock slightly. method: gyro activity statistics from the sensor checks plus the per-event peak |w| / peak |a| computed from the event catalog Q40. Are apparent vibration events actually tiny rotations/rocking of the phone rather than translational acceleration? [tentative / confidence medium] limiting factor: gyro/accelerometer coupling MEASURED: upstairs: median |w|/|a| not available (p25 not available, p75 not available) over 0 events, not available rotation-dominated; gyro handling time 0 s downstairs: median |w|/|a| 0.194 (p25 0.194, p75 0.194) over 1 events, 0% rotation-dominated; gyro handling time 1.45 s INTERPRETATION: A ratio below 1 means translation dominates the recorded motion; events above the ratio are rotation-dominated and are flagged as such in the per-event evidence. NOT ESTABLISHED: The ratio is indicative only and depends on the event window and on the physical coupling of the two sensors, so whether a specific event was rocking rather than translating is not established. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! The ratio is indicative only because the sensors are physically coupled. method: per-event peak gyro magnitude divided by peak resultant acceleration Q41. Do magnetic-field changes occur simultaneously on both phones? [tentative / confidence medium] limiting factor: different static fields at the two sites MEASURED: The two |B| series correlate at r = 0.0935 with the upstairs series offset by -3.1344 s, and their median coherence over the shared window is 0.0011 against a significance level of 0.0045 (significant=False). Two phones in two rooms sit in different static fields, so equality is not expected; a weak shared low-frequency magnetic variation (mains wiring, large appliances) is a plausible but unproven reading. INTERPRETATION: Cross-correlating the two |B| series over the overlap, and comparing the coherence of that pair with the significance level for the segment count, tests whether any magnetic variation is seen by both phones, which would point at a shared electrical or moving-metal source. NOT ESTABLISHED: Two phones sitting in different static magnetic fields cannot establish a shared magnetic source: a coincident deviation is evidence only, and phone magnetometer filtering and drift limit what can be concluded. SENSITIVITY: {"coherence_significance": 0.00454, "n_samples": 992016} ! A house has a static magnetic field gradient; the two phones sit in different fields, so a shared source must show as a coincident deviation, not an equal value. method: resample |B| onto the common grid over the overlap, then normalised cross-correlation and Welch coherence of the two |B| series Q42. Are magnetic-field disturbances correlated with vibration events, potentially indicating nearby electrical equipment or moving metal? [tentative / confidence medium-low] limiting factor: magnetometer filtering and drift MEASURED: Per-site magnetic versus vibration: {"downstairs": {"anomaly_seconds": 0.08, "fraction_events_above_3sigma": 0.0, "log_corr_with_vibration": null, "median_event_delta_uT": 0.5868, "median_uT": 69.7776, "sigma_uT": 0.31839}, "upstairs": {"anomaly_seconds": 0.0, "fraction_events_above_3sigma": null, "log_corr_with_vibration": null, "median_event_delta_uT": null, "median_uT": 57.0973, "sigma_uT": 0.42175}} INTERPRETATION: The magnetic change measured inside each vibration-event window, and the fraction of events whose change exceeds three robust sigmas of the magnetic channel, show whether magnetic and mechanical disturbances happen together. NOT ESTABLISHED: Phone magnetometer output is aggressively filtered and drifts, so a coincidence does not establish nearby electrical equipment or moving metal as the source of either signal. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! Treat these as coincidences to investigate, not measurements. method: per-event magnetic excursion against the site magnetic robust sigma, plus the anomalous-time total from the sensor checks Q43. Do particular magnetic disturbances repeat at regular intervals? [tentative / confidence low] limiting factor: few flagged excursions MEASURED: Interval quantiles of magnetic excursions beyond median + 4 robust sigma: {"downstairs": {"p0": 0.501, "p1": 0.5098, "p100": 5654.31, "p25": 0.721, "p5": 0.531, "p50": 1.29, "p75": 2.951, "p95": 70.1, "p99": 2326.8548}, "upstairs": null}. Flagged samples: {"downstairs": 85137, "upstairs": null}. INTERPRETATION: A regular interval distribution or a clear autocorrelation peak in the excursion train would indicate a repeating magnetic disturbance rather than random drift. NOT ESTABLISHED: Single intervals are weak evidence: the number of flagged samples is small, the phone filtering varies with firmware, and a machine cycle is not established from the interval distribution alone. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! Use the interval distribution as a candidate pattern, not as proof of a cycle. method: interval distribution of flagged magnetic excursions (median + 4 robust sigma), grouped in questions.py when the analysis layer does not score it Q44. Do sound-level spikes in the dB channel coincide with accelerometer events? [tentative / confidence medium] limiting factor: dB channel weighting/saturation; uncalibrated microphone MEASURED: upstairs: dB-versus-acceleration correlation 0.02, lag -0.0719 s; loud 4.2 s, vibration 0.22 s, both 0 s; microphone median 33.414, p99 44.348 (uncalibrated) downstairs: dB-versus-acceleration correlation 0.025, lag -0.0247 s; loud 107.57 s, vibration 1.53 s, both 0.4 s; microphone median 33.812, p99 48.472 (uncalibrated) INTERPRETATION: The correlation between the dB channel and the acceleration channel, together with the number of seconds that are loud, vibrating or both (excursions beyond median + 6 robust sigma in each channel), shows how often sound and vibration move together. NOT ESTABLISHED: The dB channel is uncalibrated, weighted and filtered differently per device and saturates, so a coincidence does not establish that the sound and the vibration came from the same physical source. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! The microphone band and weighting are not calibrated. method: excursion counts beyond median + 6 robust sigma in the dB and acceleration channels, plus their normalised cross-correlation Q45. Are there strong vibration events with little or no corresponding acoustic event? [tentative / confidence medium] limiting factor: dB channel band/saturation MEASURED: upstairs: 0.22 s of vibration excursions with no loud sound out of 0.22 s of vibration time (0 s had both) downstairs: 1.13 s of vibration excursions with no loud sound out of 1.53 s of vibration time (0.4 s had both) INTERPRETATION: Seconds in which the acceleration channel exceeds its median + 6 robust sigma while the dB channel does not are the candidates for strong vibration with little or no corresponding acoustic event. NOT ESTABLISHED: "Weak sound" can also mean sound outside the microphone band or below its noise floor, so the absence of a corresponding acoustic event is not established from this threshold count. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! The dB channel saturates and is weighted/filtered differently per device. method: seconds flagged as vibration excursions without a simultaneous loud excursion (median + 6 robust sigma in each channel) Q46. Conversely, are there loud acoustic events that cause essentially no measurable structural vibration? [tentative / confidence medium] limiting factor: threshold-based scoring MEASURED: upstairs: 4.2 s of loud sound with no vibration excursion out of 4.2 s of loud time downstairs: 107.17 s of loud sound with no vibration excursion out of 107.57 s of loud time INTERPRETATION: The mirror image of Q45: seconds in which the dB channel is loud while the acceleration channel stays below its excursion threshold are the candidates for sound that did not couple into the structure. NOT ESTABLISHED: The count is threshold-based and the acceleration channel is only compared against the same phone's floor, so "essentially no measurable structural vibration" is not established for any individual acoustic event. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! Pure acoustic events below the vibration threshold are counted, but an inaudible vibration outside the band is not. method: seconds flagged as loud-sound excursions without a simultaneous vibration excursion (median + 6 robust sigma in each channel) Q47. For events appearing in both sound and acceleration, does sound arrive at the phones at a measurably different relationship than the structural vibration? [unsupported / confidence low-medium] limiting factor: timing resolution MEASURED: Per-site sound-versus-vibration lag: {"downstairs": -0.0247, "upstairs": -0.0719}. Cross-phone sound lag -0.199 s (r=0.623); cross-phone vibration delay not available. INTERPRETATION: Comparing the sound-versus-vibration lag at each site with the cross-phone vibration delay tests whether the acoustic path behaves differently from the structural path. NOT ESTABLISHED: Airborne sound crossing a house takes a few milliseconds, which is below the timing resolution of the common grid, so any measured difference is dominated by clock and filter delay and the relationship is not established. SENSITIVITY: {"match_tolerance_s": 1.5, "loose_tolerance_s": 5.0} ! Airborne sound would cross a house in a few ms, far below one sample at the common rate; any measured difference is dominated by clock and filter delay. method: lag between the dB channel and the acceleration channel at each site, then the same comparison across the two phones Q48. Can anomalous GPS/location changes while the phones are stationary be identified and removed rather than interpreted as physical movement? The sample already illustrates why this matters: the reported GPS speed changes despite the device apparently being stationary. [tentative / confidence high for the artefacts; the removal rule is conservative] limiting factor: GPS sentinels / stale fixes MEASURED: GPS/position artefacts per site: {"downstairs": {"available": true, "fix_span_m": 26.81, "integrated_speed_m": 277.14, "lat_median": 27.548981, "lon_median": -80.362236, "n_fixes": 1350572, "n_speed_missing": 22232, "speed_max_mps": 2.312}, "upstairs": {"available": true, "fix_span_m": 0.0, "integrated_speed_m": 0.0, "lat_median": 27.548847, "lon_median": -80.362099, "n_fixes": 1514228, "n_speed_missing": 0, "speed_max_mps": 0.0}} INTERPRETATION: Comparing the stability of the fix with the reported speed, and integrating the reported speed against the actual fix displacement, identifies GPS rows whose movement cannot be real while the phones are stationary. NOT ESTABLISHED: The artefacts are identified and excluded from movement interpretation, but with no independent ground-truth position a true displacement of a stationary phone is not established either way. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! Rows flagged as GPS artefacts are excluded from movement interpretation but are still present in json/; nothing is deleted. method: compare the stability of the fix to the reported speed; integrate the reported speed and compare with the fix displacement; flag sentinels and invalid fixes Q49. Are there long-term sensor biases, drift, temperature-like trends, orientation changes, dropped samples, or sampling-rate changes that need to be corrected before comparing the phones? [tentative / confidence high for dropouts and rate; medium for drift as a temperature proxy] limiting factor: no temperature channel MEASURED: upstairs (csv/multi_1790959520.160099.csv): sampling 100 Hz (rate changes 0, gaps beyond 5x median 0, largest gap 0.023 s, median interval 0.01 s); gravity tilt change 0.014 deg; orientation drift 1.07 deg max; largest accelerometer DC change 0.0009 m/s^2; exclusions "not available" | downstairs (csv/multi_1790959615.928701.csv): sampling 100 Hz (rate changes 0, gaps beyond 5x median 0, largest gap 0.016 s, median interval 0.01 s); gravity tilt change 0.067 deg; orientation drift 0.94 deg max; largest accelerometer DC change 0.0108 m/s^2; exclusions "not available". Concrete correction rules used here: per-recording axis medians are removed before every RMS/PSD (so static bias cannot masquerade as vibration), both recordings are resampled onto the common 100 Hz grid (usable band up to 45 Hz), and GPS rows flagged in Q48 are excluded from movement interpretation. Temperature was not recorded, so "temperature-like" can only mean a slow monotonic block-median trend (see per-site longterm.bias). INTERPRETATION: Sampling health (gaps, duplicate timestamps, rate spread), gravity-vector orientation drift and slow block-median DC trends determine which corrections must be applied before the two phones can be compared. NOT ESTABLISHED: Temperature was not recorded, so a temperature-like trend cannot be established; only a slow monotonic block-median trend is visible, and an absolute bias correction for either unit is not established. SENSITIVITY: {"upstairs_threshold_sweep": [{"percentile": 99.0, "threshold_ms2": 0.005842415, "events": 2, "threshold_over_median": 1.404}, {"percentile": 99.5, "threshold_ms2": 0.006052113, "events": 2, "threshold_over_median": 1.455}, {"percentile": 99.9, "threshold_ms2": 0.006548966, "events": 2, "threshold_over_median": 1.574}, {"percentile": 99.95, "threshold_ms2": 0.006781433, "events": 2, "threshold_ove ! Temperature was not recorded, so "temperature-like" can only mean a slow monotonic block-median trend. method: block medians for drift, gravity-vector angle for orientation, timestamp gaps and duplicates for dropouts and rate changes Q50. After correcting clock offset, sensor bias, orientation and sampling differences, which physical events can be demonstrated statistically to be common to both locations rather than coincidence or sensor noise? [unsupported / confidence medium] limiting factor: no matched event pairs within the tolerance MEASURED: 0 matched event pairs out of 0 upstairs / 1 downstairs detections (expected random matches not available). Median per-event amplification not available over 0 pairs; sign test unavailable Median cross-phone window correlation -0.002 (quiet) / 0 (active). Clock offset estimate 3.2984 s estimated from the shared timeline. Cross-phone coherence median 0.001, p90 0.004. INTERPRETATION: After the clock correction and the bias/orientation/sampling handling recorded per site, the supported statement is which detections are coincident beyond the random-match expectation with a consistent amplitude relationship. Verdict: NO SHARED SIGNAL ABOVE THE NOISE FLOOR NOT ESTABLISHED: Demonstrating a shared physical source needs a synchronised or triggered recording: coincidence statistics do not identify the equipment, the room of origin, or a structural defect. SENSITIVITY: {"match_tolerance_s": 1.5, "loose_tolerance_s": 5.0} method: event matching + sign test + random-match expectation + coherence END — 50 answers (status: insufficient_data=11, supported=4, tentative=28, unsupported=7)