R8 — Passenger and carriage metadata
Status: Wave 1 research recommendation
Schema / required-field policy: tunes
Capture UX (minimise form burden): tunes-ios
Validation / campaign profiles: tunes-web + campaign configs in tunes
Problem
Placement and context change what a phone hears. Professional surveys fix height, position, occupancy, and exclusions and disclose deviations (acoustic-survey-methodology.md). Crowdsourced TUNES cannot demand a laboratory form on every commute — but omitting carriage position, phone placement, and crowding makes aggregates harder to interpret and easier to misuse.
Open question #9: How much route and carriage metadata should users enter manually? Answer with a tiered metadata policy: infer what is safe, ask little before record, allow rich optional after, reserve long forms for campaigns.
Principles
- Minimise pre-recording friction — starting a valid session must stay commute-friendly.
- Prefer derived / automatic fields over typed fields when accuracy is adequate.
- Never block contribution on optional human factors metadata for informal public tiers.
- Campaigns may require more — research protocols can demand fields that casual mode does not.
- Metadata confidence is its own axis — incomplete metadata lowers metadata confidence, not a fake blended score.
- No unsafe behaviour — never instruct contributors to obstruct doors, lean into tracks, or stand in forbidden positions for “better” data (see code of conduct).
Classification model
Every candidate field falls into one (primary) collection mode:
| Mode | Meaning | When shown |
|---|---|---|
| Auto-infer | Device / clock / network / sensors | Always attempted; user may override if wrong |
| Pre-select | Needed to constrain the session | Before or at record start; keep short |
| Post-add | Useful, recallable after the ride | After stop / during review |
| Optional | Improves interpretation; skip forever OK | Progressive disclosure |
| Campaign-required | Mandatory only under named protocol | Campaign profile gate |
A field may be auto-infer with optional confirm (best of both).
Field catalogue (decision-oriented)
A. Journey context (mostly R7; listed for completeness)
| Field | Mode | Notes |
|---|---|---|
| System / line / direction | Pre-select | Required for comparable section IDs |
| Origin / destination | Pre-select | Required prior path |
| Branch choice | Pre-select when topology needs it | Else post-correct |
| Local wall time / timezone | Auto-infer | Provenance |
| Planned vs actual timing flags | Auto + post | Disruption noted if user marks |
| Service disruption (subjective) | Optional / post | Free-text carefully — privacy (R5) |
B. Vehicle and carriage
| Field | Mode | Notes |
|---|---|---|
| Train / unit identity | Optional / campaign | Rarely known to passengers; do not require |
| Rolling stock / train type | Auto-infer if network service implies; else optional | Prefer service→stock maps in tunes data over guessing |
| Front / middle / rear of train | Post-add (recommended optional) | High interpretive value; coarse 3-way control beats precise carriage numbers |
| Approximate carriage number | Optional / campaign | Hard on many fleets; campaign-only unless user knows |
| Position in carriage (end / centre / near doors / near bogie) | Post-add optional | Affects door chimes, bogie/rail noise; coarse chips not free text |
| Sitting vs standing | Post-add optional | Maps to ISO-ish height intent (1.2 m vs 1.6 m) without claiming compliance |
Recommendation lean: Default informal UI asks front / middle / rear after recording (one tap row). Deeper carriage geometry is optional expand or campaign.
C. Phone placement and measurement hygiene
| Field | Mode | Notes |
|---|---|---|
| Phone in hand / pocket / bag | Post-add strongly recommended; campaign-required for higher tiers | Pocket/bag is a major quality flag (NoiseCapture-style protocol) |
| Orientation / which mic facing | Auto where OS allows; else optional | Do not quiz every user |
| Microphone obstructed / case concern | Auto hints (level floor) + optional confirm | Quiet-room check ≠ calibration (assumption #6) |
| Headphones worn (playback) | Optional post | Affects perception survey more than mic; still useful |
| Bluetooth / voice-processing / route changes | Auto-infer from audio session | Quality flags (R3) |
Protocol guidance (contributor education, not a claim of Class 1): prefer hand-held, unobstructed mic, minimise handling noise — aligned with participatory mapping practice notes, labelled as guidance.
D. Crowding and environment
| Field | Mode | Notes |
|---|---|---|
| Crowding (coarse) | Post-add optional | e.g. empty / light / busy / packed — ordinal chips |
| Estimated load % | Avoid in casual UX | Too precise; campaign only if trained |
| Windows / doors open | Optional where mode relevant | More for surface/overground; N/A many Tube cars |
| Weather | Auto (if on-surface + permission) or skip | Weak underground; do not require |
| Child / buggy / wheelchair space proximity | Optional | Accessibility research value; never required |
Crowding is context, not a substitute for acoustic metrics. Store separately.
E. Subjective discomfort (pointer only)
Overall discomfort and perception items belong in R9, not buried in metadata forms. Metadata may include a single optional “add perception survey” entry point after journey alignment confirm.
F. Device and processing (almost never asked)
| Field | Mode |
|---|---|
| Device model, OS, app version | Auto-infer |
| Calibration / profile state | Auto from device profile store |
| Clipping / dropout / AGC suspicions | Auto quality flags |
| Contributor pseudonymous ID | System |
Burden budget (UX policy for tunes-ios)
Pre-recording (casual mode) — keep to journey essentials:
- Line + direction
- From station
- To station (or “until I stop”)
- Branch if required
Optional one-liner: “You can add carriage details after.”
During recording: no forms. Status only (recording healthy / clipped / interrupted).
Post-recording — progressive, skippable:
- Confirm / edit section alignment (R7) — required for public comparison eligibility
- Phone placement chip (hand / pocket / bag / other) — recommended, not blocking
- Train third (front / middle / rear) — recommended
- Crowding coarse — optional
- Sitting/standing + in-carriage position — optional expand
- Perception survey (R9) — optional
Campaign mode: a versioned campaign profile can mark any optional field required, add trained codes (stock, carriage number), and require reference-protocol placement. Casual uploads must not inherit campaign required-sets by accident.
What not to ask casually
- Exact dB guesses from the user
- Medical diagnoses or hearing conditions (separate ethics path if ever researched)
- Free-text venting as a substitute for structured fields (privacy + moderation cost)
- Fine GPS share “to help metadata”
- Anything that encourages unsafe positioning
Interaction with quality tiers
Illustrative coupling (final tiers in calibration / quality ADR):
| Metadata completeness | Effect |
|---|---|
| Journey confirmed + device auto fields only | Eligible for lower public tiers if acoustics OK |
| + phone placement + train third | Raises metadata confidence; helps stratified analysis |
| Campaign-complete placement protocol | May be required for designated research subsets |
| Pocket/bag without disclosure | Auto flag; may cap tier regardless of other fields |
Missing optional metadata must not look like acoustic failure. Map and exports should expose “n with placement known” separately from “n measured”.
Schema implications (tunes)
- All human metadata fields nullable unless campaign profile says otherwise.
- Store source:
auto|user_pre|user_post|campaign. - Prefer enums / ordinals over strings.
- Version the enum sets; additive changes preferred.
- Do not put subjective scale items in the objective measurement object (R9).
Anti-patterns
- Long pre-flight questionnaires that kill recording starts
- Requiring carriage numbers on fleets where numbers are invisible
- Treating crowding self-report as occupancy telemetry
- Silent defaults like “hand-held” when unknown
- Blending metadata completeness into a single “trust score” with loudness
Current disposition (ADR-009)
ADR-009 accepts the minimise-manual-metadata direction. These details remain the research rationale:
- Casual pre-select = journey path only (plus branch when needed).
- Post-add recommended chips: phone placement + front/middle/rear.
- Everything else optional unless campaign-required.
- Auto-infer device, audio-session, and clock fields; never re-ask.
- Campaign profiles are data, not hard-coded app modes — owned as config in
tunes, enforced intunes-ios/tunes-web. - Open Q #9 → resolve as: minimise manual entry; prefer post-recording coarse chips over pre-forms.
Open questions for user tests
- Will contributors actually answer two post chips if skip is easy?
- Does asking placement before start improve behaviour (hand-held) more than asking after? (Hypothesis: short pre-nudge + post-confirm may win — test, don’t assume.)
- Ordinal crowding labels: wording that travels across cultures/cities.
- Whether “until I stop” destination reduces destination pre-select errors.
Recommendation
Adopt a burden budget: pre-recording asks only for the route prior needed for alignment; after recording, offer two recommended chips (phone placement; front/middle/rear) and keep carriage geometry, crowding, and posture optional. Auto-infer device and session quality fields. Use campaign-required profiles for research density instead of burdening every commuter. Record metadata source and keep metadata confidence separate from acoustic and alignment confidence.
Confidence: High for the mode taxonomy and burden-budget shape; Medium for the exact “recommended” chip set until usability testing; Low for any predicted completion rates (not invented).
Depends on experiment/legal/user-test? Yes — user tests for form abandonment vs metadata completeness; privacy review of any free-text; campaign ethics if required fields expand; no legal blocker for coarse carriage chips themselves.
Links to related docs: 07-journey-alignment.md; 09-subjective-experience.md; 04-calibration-uncertainty.md; 03-consumer-device-limits.md; 05-privacy-ethics.md; acoustic-survey-methodology.md; ADR-009; H14 recorder; ../governance/project-charter.md; tunes-ios product-scope.md.