Add a governed GTFS repair and enrichment stage before MOTIS import

[Medformatik]

Problem

Many useful GTFS feeds are structurally valid but incomplete for routing: they may publish only Fares v1, omit platform connectors or walking transfers, lack stop-time detail inside subway systems, or contain provider-specific gaps. Today those problems reach MOTIS without a controlled place to repair, enrich, audit, or roll back a feed.

Current state

• services/data-manager/src/jobs/transitous/ fetches, validates, imports, stages, probes, and promotes feeds.
• The MOTIS adapter already requests and surfaces Fares v2 when a feed publishes it.
• Platform, accessibility, park-and-ride, fare-zone, and transfer concepts exist in the contracts and UI.
• No deterministic pre-import repair stage or rule provenance was found.

Proposed solution

Insert a governed transformation stage between source validation and MOTIS import. Repairs must be opt-in, deterministic, feed-scoped, versioned, observable, and reversible. Preserve the original artifact and produce a transformed artifact plus a machine-readable report.

Initial rule families:

1. Lossless Fares v1 to Fares v2 conversion where semantics permit it.
2. Platform/station connector synthesis from defensible GTFS and OSM relationships.
3. Stop-time interpolation for explicitly configured subway segments.
4. Computed walking transfers.txt entries with accessibility and distance limits.
5. Declarative backfills for known stop areas, route patterns, station links, and missing references.

Acceptance criteria

• The pipeline has an explicit repair stage before MOTIS import and re-validates its output.
• Rules have stable IDs, versions, schemas, declared inputs, and documented safety conditions.
• Reapplying a rule set is idempotent.
• Original/transformed checksums, rule versions, row counts, warnings, and validation deltas are retained.
• A bad repair quarantines the candidate and cannot replace promoted data.
• Per-feed/region allowlists prevent speculative global heuristics.
• The initial rule families have fixtures, negative tests, and MOTIS probe coverage.
• Fares conversion refuses lossy cases and explains why manual mapping is needed.
• Synthesized rows carry provenance and never overwrite authoritative source rows.
• Admin job details expose the rule set and concise before/after report.
• Operator documentation explains adding, disabling, and rolling back a rule.

Alternatives considered

Upstream corrections remain preferred, but deployments need predictable behavior while fixes are pending. Ad-hoc scripts are difficult to audit and roll back; relaxed validation does not improve routing semantics.

Area

Transit data engineering, Transitous pipeline, MOTIS operations.

Non-goals

Do not invent fares, accessibility attributes, or geometry where no defensible source exists.