Import/LetMeGetIn

From OpenStreetMap Wiki
Jump to navigation Jump to search

LetMeGetIn — accessibility contributions to OpenStreetMap

LetMeGetIn (https://letmegetin.org) is a public, open-source accessibility guide for places in France. It is the modern rewrite of Jaccede: one normalised accessibility referential, exposed as a public API and contributed to by people who know a place.

This page documents what we import from OpenStreetMap and what we propose to write back. It follows the Import/Plan Outline and the FR:Import/Guidelines.

Status: proposal. Nothing has been uploaded. The repository generates an .osmChange file for a human to review; there is no automated uploader and no edit is made without the discussion below being closed.

Goals

OSM is our community base, and its accessibility gaps are exactly what LetMeGetIn exists to fill.

  • Import: hold the places that are not yet documented for accessibility. A bakery with no wheelchair tag is a place a contributor can complete — but only if it is visible on our map.
  • Write back: return the accessibility answers our contributors record for those places, as OSM tags, so the work benefits OSM too.

We do not duplicate OSM, we do not compete with it, and we do not overwrite what mappers have surveyed by hand — a human answer in OSM always wins over ours.

Geographic scope

LetMeGetIn covers France today. The long-term ambition is to open the referential to the rest of Europe, and then beyond it, if and only if the French project works: enough places documented, enough contributors, and a working relationship with the local OSM community. We would rather say this now than let it be discovered later.

This proposal covers France only. It reads the French extract and writes to French objects. It does not ask for, and does not imply, any agreement for an import elsewhere.

A wider scope would not be an extension of this import. Each country has its own tagging conventions, its own community and its own import guidelines, so each one would come with its own proposal and its own discussion in that community. The reason to mention it here is honesty about where the project is going, not a request for a continental mandate.

Schedule

Nothing is scheduled until this proposal is accepted. The first upload, if accepted, will be a single reviewed changeset of a few dozen objects, announced in the discussion thread before it is made.

Import Data

Background

OSM Data Files

The national extract, read in streaming. Nothing is fetched from the OSM API and no Overpass query is issued.

Import Type

One-time manual upload, then recurring manual uploads of small changesets. Not a bot run, not a continuous sync. An operator reviews a generated .osmChange document and sends it from a dedicated account (see #Changeset Tags).

Licences

Three licences meet in this round trip, and they do not all travel in the same direction.

  • OSM data is ODbL 1.0. Anything derived from it stays ODbL, with the share-alike obligation that implies.
  • Our contributors' answers are offered under a dual licence (Licence Ouverte 2.0 or ODbL, the recipient chooses), so they can go to OSM under ODbL and to a Licence Ouverte partner under LO 2.0.
  • AccesLibre, our main data source, publishes under Licence Ouverte 2.0. That licence allows redistribution and combination with other information on a single condition: attribution, meaning the source and the date of last update. It also states that it was designed to be compatible with any free licence requiring at least attribution. AccesLibre data can therefore enter an ODbL database and reach OSM, and this page plus our changeset tags are where the attribution lives.

The other direction is refused. An ODbL row carries share-alike, and sending it to a database that publishes under Licence Ouverte would impose an obligation that licence cannot carry. The redistribution rule is reviewed data (data/referential/licence_destinations.csv), not a branch in the code, and a row whose licence may not leave for a destination is counted rather than written.

Data Preparation

Data Reduction & Simplification

No geometry is created, simplified or removed: we only ever modify existing objects, and we keep their identifiers and versions.

Tagging Plans

Two reviewed files drive everything, and both are data, never code:

  • category_mapping.csv — which OSM tag maps to which of our categories (currently 124 of our 158 categories resolved);
  • osm_accessibility_tags.csv — which OSM accessibility tag pair maps to which of our criteria values.

A POI whose tags match no reviewed row is skipped and counted, never guessed. The import report publishes the most frequent unmatched tag combinations, so the mapping grows from evidence.

On the write-back side, we add only tags a reviewed row can express. The mapping is deliberately conservative: values we cannot map legally are left alone rather than approximated — for instance wheelchair=no, ramp, handrail and hearing_loop are not mapped today, because our referential has no criterion that can carry them without inventing meaning.

Two OSM keys carry our answers today, wheelchair and toilets:wheelchair, across four values. Everything else we hold stays on our side until a reviewed row can express it.

What we do not propose

Our referential contains 182 criteria and four handicap families. We are not proposing it as an OSM tagging scheme, and we would ask you to hold us to that.

OSM's accessibility vocabulary is about thirty keys — wheelchair, toilets:wheelchair, ramp, handrail, tactile_paving, traffic_signals:sound, capacity:disabled, kerb, plus wheelchair:description for what no key covers. It took years of discussion to settle, and it works. Our criteria are finer-grained than any key set: a card-reader height, a door-handle shape, a contrast against the wall. Proposing them would mean asking OSM to adopt a taxonomy designed for one product, backed today by two mapped keys. That would be a hostile act dressed as a contribution, and it would be rejected.

If something in our model ever looks worth proposing, it will arrive through the tagging proposal process on its own: a Proposed features page, the tagging mailing list, and a discussion that starts from your vocabulary rather than ours. Not folded into an import.

The one thing we do have that OSM has no place for is which questions matter for which kind of place — a bakery and a swimming pool are not surveyed the same way. That belongs in wiki prose per tag, not in a schema, and we would rather write it with you than at you.


Changeset Tags

  • comment = LetMeGetIn accessibility contribution — see <this page>
  • created_by = LetMeGetIn <version>
  • source = survey for answers recorded in the field by a contributor; the generated document records the provenance per answer so the operator can state it accurately.

Changeset size is capped by the review workflow, not by the tool: an operator sends one reviewed batch at a time. No changeset will exceed a few hundred objects.

Data Transformation

The generator re-emits the object's full stored tag set with our accessibility tags added, because an .osmChange <modify> replaces the whole tag set — omitting a tag would delete it. An object whose version or raw tag set we do not hold is not emitted and is counted: we would rather skip a place than damage it.

Nothing is ever created and nothing is ever deleted.

Data Transformation Results

The generated document is a plain .osmChange file plus a one-line changeset comment, both left beside each other for review. It is not published anywhere automatically.

Data Merge Workflow

Team Approach

One operator, reviewed by hand. Sébastien Billion (contact below) is the maintainer.

Workflow

  1. A contributor records an accessibility answer on LetMeGetIn for a place we imported from OSM.
  2. osm publish refuses to run unless every policy artefact and an explicit, human-granted contributor-terms consent are present, and it only ever emits objects we hold a version and a full tag set for.
  3. The operator opens the generated .osmChange in JOSM, checks the objects, and uploads.
  4. Every changeset links this page.

Conflation

Deduplication is internal, against our own database — we never query an external API to find a match, and we never merge in OSM.

When we import, a candidate within 100 m whose normalised name scores ≥ 0.90 is linked to our canonical row (identifier kept, name kept as an alias, merge logged); between 0.75 and 0.90 it goes to a review queue. Nothing is ever deleted or hard-merged on either side. A linked OSM object is still emitted under its own node id: we modify the OSM object that exists, never our canonical placeholder.

We are not the first to work on this, and we would rather say so than be told.

  • FR:Complète Tes Commerces (https://complete-tes-commerces.fr, by MerylStreet) already helps French contributors complete the establishments of a city in OSM, wheelchair included, with a per-city dashboard and completion wizards. It answers what is missing; we answer *what a place is like for each kind of disability*. We import from OSM and write back, so the two should reinforce each other rather than compete, and we would welcome being pointed at the right way to align.
  • OSM Mon Commerce pursues the same completion goal from the shop side.
  • The forum thread Acceslibre et utilisation détournée des données OSM asked in 2023 whether a public accessibility platform built on OSM had to publish under ODbL, and in 2025 whether its data could be extracted to enrich OSM. Neither received a code answer. This proposal is an attempt at one.

QA

  • Every answer written back is validated before it leaves: a null or out-of-type value is refused, never coerced — a null would delete the OSM tag.
  • Freshness is compared first: if OSM's object is newer than our answer, we skip and journal the decision. A re-run of the same batch patches nothing.
  • Every attempt is journalled with the losing value kept, so a mistaken write can be traced and reverted object by object.
  • The generator refuses to write at all if a policy artefact is missing or invalid.

Discussion

Discussion is open and this page will record its outcome before any upload happens.

Four things we would like challenged, beyond the mapping itself:

  1. Is the tag mapping reasonable, or are there values we should not write at all?
  2. Should a declared accessibility answer be tagged as such, or do you expect source=survey and nothing less? We can carry our provenance into the changeset and stop there.
  3. How should we coordinate with Complète Tes Commerces and OSM Mon Commerce? If there is a mapping or a workflow we should align with, we would rather adopt it than invent a third one.
  4. Scope. This proposal is French. If the project later covers other countries, would you rather see one round trip per country, coordinated with each local community, or a different shape?

Contact

Sébastien Billion — contact@letmegetin.org — https://letmegetin.org