Local schemas and semantic data: Protobuf, Avro, JSON-LD, RDF and CRS

Choose the interpretation your receiving task needs

When you have bytes and their .proto, inspect the selected Protobuf message. When you have writer and reader schemas, compare Avro resolution directions. Byte inspection reads a specific message; schema compatibility asks whether two declarations can resolve in the chosen direction.

Use offline JSON-LD processing to expose term IRIs or compact expanded data with a supplied context. Use canonical RDF comparison to test whether complete N-Quads datasets are equivalent despite temporary blank-node labels. Use CRS reprojection when geometry coordinates must move between the explicitly supported longitude/latitude and Web Mercator pair. These are separate jobs; formatting JSON does not perform them.

Inspect Protobuf with the original schema and retain presence

Supply a binary file or strict hex/base64 text, paste the standalone proto2/proto3 schema in the secondary field and specify its full message path. In demo.Event below,087b120161 carries seq123 and payload byte a. Decoded output represents seq as decimal string "123" and payload as base64 "YQ==". No type is guessed, import fetched or service executed.

Read final decoded values and presence separately from defaults. A proto3 implicit zero can be omitted; optional and oneof zero values retain explicit presence. Singular submessage fragments merge before recursive final required checks. A later oneof member wins, while repeated fragments of the same message member merge. Unknown fields, wrong wire types and closed-enum unknown values retain original fragments; the complete binary stays available. Inspection JSON is not lossless re-encoding.

Declared groups, extensions, editions, services, external imports and features overrides are unsupported. Protected identifiers reject, as do explicit proto2 byte defaults and unsafe/nonfinite numeric schema defaults. Schema stays data, without user script or service execution; controlled internal decoder generation runs in a same-origin independent HTTP Worker without changing page CSP.

syntax = "proto3";
package demo;
message Event { int64 seq = 1; bytes payload = 2; }

Label the Avro writer and reader before reviewing both directions

The primary input is the writer schema; the secondary input is the intended reader. An int writer can promote into a long reader; the reverse cannot guarantee every long fits int. Valid incompatible schemas produce a complete false report. Invalid JSON or schema rejects. Forward/backward labels still require clearly identified participants.

A new reader field needs an appropriate default. Supported reader record/field aliases can resolve renames, and fullnames include namespace. Underlying resolutionError remains text rather than a fabricated exact field path. Named recursive schemas are bounded; datum, IDL and Schema Registry history/policy are not evaluated.

Pinned avsc5.7.9 requires union field defaults to match the first branch. Logical types and protected identifiers are unsupported. Removing a logicalType changes the question, so do not present the altered result as compatibility of the original schema. Resolution does not validate every business value constraint or registry policy.

Provide every JSON-LD context locally

Paste an object or array, choose expand or compact, and supply the secondary envelope with contexts and optional compactContext. Each mapped document must contain @context and its URL key must match the requested absolute HTTP(S) URL exactly. A missing mapping fails; the tool never requests the URL. Compaction requires a non-null compactContext.

For example, a context mapping name to https://example.com/name expands name:"Alice" to that IRI with @value:"Alice". An unbound property can disappear, and an empty expansion is valid. Check the term mapping before deciding that an empty result is a processing error. Set a document base explicitly for relative IRIs; the base of an imported document is not automatically inferred from its local filename.

{"contexts":{"https://example.com/context":{"@context":{"name":"https://example.com/name"}}},"compactContext":"https://example.com/context"}

Compare RDF datasets, then interpret the diff

Paste the right dataset and paste or open the left export. Both must be strict N-Quads, with absolute IRIs and supported RDF 1.1 terms. If your source is RDF/XML or Turtle, export it first. The parser retains named graphs, datatype and language meaning; repeated quads follow set semantics and language tags are normalized to lowercase.

Two lines that differ only in _:alice versus _:person can describe the same dataset. RDFC-1.0 with SHA-256 assigns a common canonical form. For unequal datasets, the displayed additions/removals compare their separately canonicalized line sets. A graph edit can change many blank-node labels, so the difference is not a minimal edit plan or a persistent node correspondence. Download both canonical files for review.

Make CRS and axis order explicit

The supported pair is EPSG:4326 longitude,latitude in degrees and EPSG:3857 easting,northing in meters. Select different source and target values. A six-digit coordinate does not establish its CRS; obtain that information from the source. Unknown CRS, 3D/M positions and coordinates outside the supported area reject.

At longitude120 and latitude30, Web Mercator is approximately [13358338.895192828,3503549.843504374] meters. Confirm one known point and reverse it within the stated numerical tolerance. GeoJSON geometry/ring/hole order and Feature IDs/properties are retained; metadata numbers keep their source tokens. Existing bbox/crs members reject so stale bounds cannot survive. Preserve those separately before deliberate removal. No UTM, datum/grid conversion or topology repair is supplied.

Use the stated budgets and recover from errors

These budgets are input and computation contracts, not evidence that a receiver accepts the download. Each preview shows at most200 rows/nodes while downloads contain the complete successful result. A failed run supplies no partial result.

• Retain originals and the exact context/CRS choices; reopen the downloaded result in the receiving application.
• Fix a missing JSON-LD mapping under its exact URL, correct N-Quads syntax, or correct source CRS/axis order before rerunning.
• When a work budget fails, partition only if the meaning of the dataset survives that partition; renaming an extension does not change support.

Tools used in this article