Skip to documentation content
TUNES — Transit User Noise Experience Survey

London pilot design

Must align with acoustic-survey-methodology.md: station-to-station units, layered metrics, documented exclusions, honest survey-grade labelling. R3291 is a methodology reference, not redistributable data or TfL affiliation.

Goals

  • Dense repeated observations on priority sections (not shallow network scrape)
  • Multi-device variance visible
  • Some co-measurements beside reference equipment
  • Privacy flow rehearsed end-to-end with known contributors
  • Produce a reproducible pilot release candidate in tunes

Priority lines (initial)

Central, Jubilee, Victoria, Northern — useful coverage and known acoustic variety. Exact section lists chosen for operational feasibility (both directions where practical).

Sample strategy

  • Prefer overlap: same sections, many passes, peak and off-peak
  • Vary carriage region and phone placement (logged as metadata)
  • Include known noisy and quieter comparison sections when identifiable from prior public methodology literature — without inventing dB claims
  • Exclude or flag non-representative holds (document like professional footnotes)

Devices

  • Mix of supported iPhone models
  • At least one reference Class 1 SLM campaign subset for calibration hierarchy experiments
  • External mic optional stretch goal — not required to start

Contributor protocol

  • Training: phone in hand, avoid handling noise, offline OK, complete alignment + privacy preview
  • No unsafe behaviour for “better data”
  • Derived-only upload default

Privacy procedure

Consent script; show upload fields; local raw retention choice; withdrawal IDs issued.

Analysis plan

  • Per-section distributions by tier and time window
  • Device-model bias exploration (hierarchical calibration hypothesis — do not claim success pre-evidence)
  • Alignment error audit on a labelled subset
  • Subjective vs objective kept separate in reports

Success criteria (qualitative gates)

  • Correction UX completable on real trips
  • Privacy tests pass
  • Enough repeated sections to demonstrate honesty chrome (sample size > 1 narrative)
  • Written limitations published before any public map from pilot data
  • No endorsement language in materials

Recommendation

Run pilot as Phase 6 only after Phases 2–4 spikes; publish pilot dataset as clearly pilot labelled release.

Confidence: Medium (ops details TBD).

Depends on experiment/legal/user-test? Yes — all three.

Links: 04-roadmap.md; 08-risk-register.md; ADR-001; H08 — London pilot.