H13 — Public map
The public map translates versioned, section-level evidence into a passenger-facing quieter-journey view. It must make coverage and uncertainty as visible as the acoustic result.
For: passengers, map designers, researchers, journalists, and tunes-web implementers.
Assumptions: map geometries use directed station-to-station sections; crowd data is survey-grade at best; “quieter” is comparative within the released dataset.
Goals
- Compare observed acoustic conditions between plausible journey sections.
- Show where evidence is dense, sparse, conflicting, or absent.
- Keep measured acoustics separate from reported passenger experience.
- Let every displayed result be traced to a dataset, schema, and pipeline version.
- Remain readable without implying engineering-grade or health-risk authority.
The first map is expected to use L2 section aggregates. Record-level observations and reprocessed metrics remain separate data layers; passenger-facing rankings and styles are interpretations.
flowchart LR
A[Consented derived observations] --> B[Validation and quality gates]
B --> C[Section aggregates]
C --> D[Coverage and confidence]
C --> E[Objective map layer]
F[Optional perception reports] --> G[Separate subjective layer]
D --> E
D --> G
Aggregation and confidence
Each section summary should expose:
| Element | Why it is required |
|---|---|
| Sample count | Distinguishes repeated evidence from isolated observations |
| Distribution or spread | Avoids presenting one value as universal |
| Quality-tier mix | Prevents weak and strong contributions appearing equivalent |
| Uncertainty dimensions | Separates acoustic, calibration, alignment, and metadata limits |
| Scope and recency | States what network and release the map represents |
| Exclusions and caveats | Makes filtering decisions auditable |
Objective and subjective layers require independent sample counts. A session-level perception report must not be copied onto every section.
Interpolation
TUNES does not fill unobserved or low-sample sections with confident acoustic values. Empty sections remain empty or insufficient evidence.
Topology snapping is not acoustic interpolation: it assigns a recording interval to a legal network section when GPS is weak. Reprocessing may estimate model-adjusted values only under a documented, versioned method; it creates a new release and does not overwrite observations.
Confidence maps may visualise coverage, tier mix, or a named uncertainty dimension. They must not collapse all uncertainty into an unexplained “accuracy” score.
Future work
Minimum sample gates, aggregate estimators, interval methods, recency rules, and any model-based spatial inference require pilot evidence and an ADR before becoming public-map defaults.
Colour and accessibility
Colour may encode a named acoustic metric or a clearly labelled perception measure, never both at once. The legend must state the metric, weighting, duration/aggregation basis, release, and insufficient-data state.
- Do not use colour as the only carrier of meaning.
- Do not imply false precision through narrow bands or smooth gradients.
- Keep low-confidence and missing-data states visually distinct from “quiet”.
- Provide text or pattern equivalents for confidence and coverage.
Future work
The repository does not yet define a canonical palette, breakpoints, or accessibility-tested scale. These belong to
tunes-webdesign work informed by pilot distributions and user testing.
Privacy
The map consumes derived, minimised data. Raw recordings are never exposed because carriage recordings may contain incidental passenger, staff, and announcement speech. Publishing waveform audio would increase re-identification and surveillance risk without being necessary for the default map.
Public layers must also avoid contributor identity, stable cross-release personal identifiers, precise location trails, identifying free text, and combinations of rare route and exact time that enable re-identification. Pseudonymous data is still personal data.
Related Documents
How it works · Privacy · Quality tiers · Legal and governance · Public data model · Calibration and quality tiers · Open data and reproducibility · Subjective experience · Claim language