Technical methods note
The model behind the journey.
Equations, objective weights, data boundaries, known limitations and the tests still required before external scientific claims.
Farleda calculates journey comparisons from distances, published emissions factors and disclosed model assumptions. This page explains the technical implementation and the evaluations still needed. It is not an independent scientific validation.
Abstract. Farleda composes public-transport routing, road geometry, activity-based emissions factors and contextual grid data into European travel-planning estimates. The current route ranking evaluates emissions, duration and transfers under four user priorities. Grid-aware electric-rail estimates can use recent provider history or an explicitly labelled modelled profile. This document describes the implementation and its limits; it does not claim peer review, novelty, ISO conformity, legal reporting compliance or proof of travel.
1. Scope and status
This is a technical methods note for the current Farleda implementation. It is not peer-reviewed, is not an ISO conformity report and makes no patentability or scientific-novelty claim. Its purpose is to make assumptions and testable behaviour explicit.
The engine produces planning estimates. It does not measure vehicle energy, prove that travel occurred, sell a fare, or certify an accommodation or activity as sustainable.
2. Emissions model
Each leg is assigned a mode, distance, activity unit and factor. Public transport and flights use passenger-kilometres. The average-car factor is per vehicle-kilometre; Farleda counts one vehicle for each group of up to four travellers, rounded up.
Eleg [kg CO₂e] = d [km] × f [g CO₂e / activity-km] × u [activity units] ÷ 1000
| Mode class | Activity units u | Current source boundary |
|---|---|---|
| Rail, coach, ferry, flight | number of travellers | UK DESNZ 2026 rounded passenger factors; flight includes radiative forcing |
| Average car | ceil(travellers / 4) | UK DESNZ 2026 average unknown-fuel car, direct vehicle operation |
| Electric rail with grid context | number of travellers | grid g/kWh × 0.09 kWh/pkm; 0.09 is a Farleda assumption |
| Walk and bicycle | number of travellers | zero direct operational emissions; lifecycle effects excluded |
The current factors are not a complete well-to-wheel or life-cycle assessment. Separate well-to-tank values are not added. UK factors are used outside the UK as disclosed proxies where specific European operator data is unavailable.
3. Distance and route evidence
- Public transport: Transitous/MOTIS schedule alternatives and feed-supplied geometry where available. Coverage and geometry quality vary by feed.
- Road, walking, cycling: The current OSRM integration uses a driving profile. Walking and cycling reuse that road geometry as an approximation; it is not a validated pedestrian or bicycle route.
- Flight: great-circle distance between selected OurAirports coordinates; not the flown track.
- Fallback: great-circle distance multiplied by a disclosed mode detour factor when routed geometry is unavailable.
Consequently, ‘real routed kilometres’ is valid only for a leg whose source metadata identifies a routed source. It is not a universal property of every itinerary.
4. Candidate ranking
For a returned set of transit candidates, Farleda removes duplicate route signatures, calculates whether each candidate is Pareto-dominated on emissions, duration and transfers, normalises each objective within the candidate pool and applies the selected priority weights.
cost(r) = wEN(Er) + wTN(Tr) + wXN(Xr); score = 100 − 100 × cost
| Priority | Emissions | Duration | Transfers |
|---|---|---|---|
| Balanced | 0.45 | 0.35 | 0.20 |
| Time | 0.15 | 0.70 | 0.15 |
| Carbon | 0.72 | 0.18 | 0.10 |
| Cost | 0.30 | 0.25 | 0.45 |
‘Cost’ is currently a proxy that favours fewer changes, combined with an upstream preference for coach in some mode-selection paths. Transitous does not provide live bookable fares, so this objective must not be interpreted as cheapest price. Reliability and accessibility are not yet numeric ranking objectives.
5. Grid-aware rail
When a transit leg has a known departure, Farleda uses its UTC hour to sample a 24-hour country profile. A configured electricityMaps history response is labelled hourly-live. Otherwise the system uses a hand-authored daily shape around an annual country value and labels the result hourly-modelled. The latter is a heuristic, not observed data and not a future grid forecast.
Cross-border values are the arithmetic mean of the two endpoint countries. Track-share weighting is not implemented. electricityMaps' consumption-based lifecycle values and the production-based annual snapshot have different accounting boundaries.
6. Context layers outside the optimiser
| Layer | Current behaviour | Not claimed |
|---|---|---|
| Klimaticket | Endpoint-based Austrian coverage approximation in the UI | Service-level fare validity or every operator's rules |
| Weather | Open-Meteo forecast ≤14 days; 2023 month average farther out | ERA5 normal or commercial entitlement to the current keyless endpoint |
| Hotels and activities | OpenStreetMap/Overpass candidate discovery | Availability, price, certification or verified sustainability |
| Language model | Optional narrative generation after deterministic validation | Source of distances, emissions or passport signatures |
7. Reproducibility and integrity
Public JSON exposes factor values, activity units, ranges and grid constants. A Trip Carbon Passport signs a canonical payload with HMAC-SHA256 and a methodology version. This detects payload changes under Farleda's server-held secret; it is not independent public-key verification and does not prove journey completion.
GET /api/public/v1/modes.json GET /api/public/v1/grid.json GET /api/public/v1/metrics.json POST /api/public/v1/passport/verify
The repository is proprietary under its root licence. Publicly readable documentation and CC BY 4.0 JSON responses do not grant an open-source licence to the product code.
8. Evaluation plan — not results
The following are acceptance tests to be run and published before external scientific claims are made. They are not completed-study results.
- Create a frozen corridor set with timestamps, travellers and expected upstream source availability.
- Compare returned schedules and geometry against operator or national journey planners at the same query time; report missing-route and timing error rates.
- Recompute every factor from the pinned DESNZ workbook and publish machine-checkable row references and rounding rules.
- Compare grid-derived rail estimates only against operator data with matched accounting boundary, geography, rolling stock and occupancy; do not mix lifecycle and production-only series.
- Run ranking invariants: dominated alternatives cannot outrank non-dominated ones solely because of labelling; identical input and candidate set must yield identical ordering.
- Commission legal/accounting review before describing outputs as ISO-, CSRD-, ESRS- or Taxonomy-compliant.
9. Reading the evidence
A result is meaningful only within its stated boundary. Published factor ranges represent different assumptions or classes; they are not statistical confidence intervals. The ranking score is relative to the returned candidate pool, not a probability or a guarantee of the best route across Europe.
| Record | What is available | What still requires evidence |
|---|---|---|
| Calculation | Distance, activity units and declared model factors | Matched operator energy, occupancy and accounting boundary |
| Passport | Signed totals, leg summaries, traveller count and declared versions | A complete frozen source/factor snapshot for independent replay |
| Journey | A saved plan is explicitly labelled planned | Actual travel completion and causal environmental impact |
The current passport contains rounded summaries, not every source response or applied leg factor. The group flight comparison in kg CO₂e uses distance in km × travellers × 128 ÷ 1,000; its comparison version is separate from the emissions-method version. Work through the units or discuss an evaluation partnership.
Appendix A — illustrative calculation
This example demonstrates arithmetic only. It is not a captured live route, price or operator claim. Assume 300 km, one traveller and the current point factors.
National rail: 300 km × 31 g/pkm × 1 ÷ 1000 = 9.3 kg CO₂e Coach: 300 km × 39 g/pkm × 1 ÷ 1000 = 11.7 kg CO₂e Short flight: 300 km × 128 g/pkm × 1 ÷ 1000 = 38.4 kg CO₂e Modelled difference from the same-distance flight scenario: max(0, 38.4 − 9.3) = 29.1 kg CO₂e
The comparison is conditional on equal 300 km activity. A real itinerary may use different routed distances by mode, access legs and unavailable schedules, so production output must calculate each leg from its own evidence.
References
- ISO 14083:2023 public abstract.
- UK DESNZ 2026 GHG conversion factors, revised flat file updated 31 July 2026.
- Transitous documentation and per-feed source catalogue.
- OurAirports data.
- electricityMaps methodology and code.
- Ember electricity-data methodology.
Documentation checked against the local calculation, ranking, routing and passport code on 10 October 2026. This was not a live-service validation or independent scientific review. Methodology v1.1 and factor values are unchanged.
Farleda methodology v1.1 · document v1.2 · 2026-10-10
