Ajinomatrix Opens the JRF Initiative to Make Recipes Portable as Computable Processes

The JRF - JSON Recipe File project targets open, machine-readable exchange of ingredients, transformations, intermediate states, provenance and sensory-linked food information - with a public video-to-recipe translation tool already live at jrf.recipes

Rebecq, Belgium - 21 August 2026 - Ajinomatrix today announced the development of JRF - JSON Recipe File, an open recipe-file initiative designed to represent recipes not only as lists of ingredients and written instructions, but as structured transformation processes that software can inspect, validate, exchange and reason over.

The project is being developed within the Ajinomatrix and Ajinoverse research and software ecosystem and is intended to become an openly documented recipe interchange format. Its central design proposition is simple: the process is the recipe.

Conventional digital recipes generally reproduce the structure of a recipe card: a title, an ingredient list and textual instructions. JRF instead aims to preserve the sequence through which ingredients are transformed into intermediate and final products.

A JRF object is therefore being designed to represent information including:

ingredients, quantities and units;
preparation and processing operations;
process conditions such as time and temperature;
intermediate products and physical states;
dependencies and parallel preparation branches;
final or serving states;
provenance and source information;
optional sensory and perceptual attachments;
extensions that do not corrupt the canonical recipe core.

The underlying file remains human-readable JSON and is intended to be usable independently of any single Ajinomatrix application.

Try it today: recipe videos become editable process files at jrf.recipes

The most direct way to understand the initiative is already public. At https://www.jrf.recipes, Ajinomatrix has released a translation tool that converts an online recipe video into a JRF file containing both an editable recipe flowchart - the process graph, which users can inspect and correct - and the complete written recipe extracted from the video. The same translation engine is available inside the broader creator environment at https://www.ajinoverse.studio.

One practical note, stated up front to avoid frustration: the extraction engine currently requires users to supply their own Anthropic Claude API key. The tools remain free to operate and usage stays under the user's own control, but without a key the translation will not run.

Extraction output follows the programme's central integrity rule, described below: missing information is shown as missing rather than silently invented.

Built on existing process science rather than replacing it

Ajinomatrix explicitly positions JRF within an existing body of process and food-information research. Industrial standards such as ISA-88 have long formalised recipes and procedural process structures for batch manufacturing. Research frameworks such as PO2/TransformON (INRAE, 2023) have demonstrated process-, step-, observation- and result-oriented representations for food and biomass transformation. Earlier open recipe formats also exist and are acknowledged.

JRF does not seek to replace these approaches. The project instead investigates whether a lightweight, developer-accessible and portable recipe file can serve as an interchange layer between conventional recipe sources, process representations, food knowledge systems, sensory data and culinary applications.

A large empirical corpus is being built to test the format

Ajinomatrix is developing the specification alongside a multilingual recipe acquisition and analysis programme, rather than attempting to freeze a format before testing it against real recipe diversity.

A current August 2026 acquisition snapshot records more than 6 million raw recipe records collected across more than 8,000 domains and more than 100 detected languages. A separate fusion and deduplication stage has already identified approximately 2.7 million unique recipe records, while retaining 58,828 multi-source relationships useful for studying recipe equivalence, provenance and variation.

These corpus figures are not presented as proof that JRF itself is already validated. They provide the empirical substrate required to test questions such as:

Can ingredients and quantities be extracted reliably across languages?
Can intermediate states be reconstructed without invention?
Can parallel culinary processes be represented consistently?
Can the source of every structured assertion remain traceable?
When are two different recipe documents representations of the same underlying process?
How much human correction is required before a machine-extracted recipe can be treated as conforming JRF?
No Silent Invention

A central design rule of the programme is No Silent Invention. When information is missing from a source recipe, extraction and conversion systems should not fabricate plausible cooking facts simply to satisfy a schema. Unknown information should remain unknown; missing information should remain missing; and where transformation causes information loss, the converter should identify what was preserved, transformed, inferred, unsupported or dropped. Ajinomatrix intends this principle to become part of JRF's conformance and validation model.

From concept to executable specification

The JRF programme is being developed together with reference tools planned or already under development for recipe extraction, inspection, validation, conversion, process visualisation, structural comparison and compatibility reporting. The current research phase is focused on formalising the object model, vocabulary, schema, compatibility rules and benchmark corpus.

Ajinomatrix is deliberately reserving a later specification milestone for the first fully evidenced end-to-end reference implementation - source recipe → structured process → .jrf file → validation → sensory attachment → knowledge-graph projection → measured benchmark. Until those gates are crossed, JRF remains an open format initiative under development rather than a declared finished standard.

An open interoperability layer for food software

JRF is intended to complement, rather than absorb, other Ajinomatrix technologies. Within the developing architecture, JRF represents the portable recipe and transformation process; MP6 represents associated sensory and perceptual information; SRS/KG-SRS provides broader provenance-aware knowledge-graph infrastructure; TT-o supplies ingredient-origin and lot-level provenance; and the Ajinoverse provides tools that produce, consume, validate and inspect these objects.

The long-term objective is to allow recipes to move between systems through transparent files rather than requiring every tool to share a proprietary backend. The initial interoperability model is deliberately simple: export → validate → inspect → convert → import.

Next step: measurable conformance

Ajinomatrix is now preparing a stratified benchmark of real recipes across languages, sources and structural complexity. The evaluation programme will measure not only ingredient extraction but also process reconstruction, state identification, provenance retention, final-state detection, unsupported inference and human correction effort.

The goal is not to demonstrate that every recipe can be perfectly understood automatically. It is to determine, experimentally, what information a portable recipe-process representation can preserve reliably - and where the limits remain.

A detailed public white paper, Beyond Ingredients and Instructions: Toward a Computable Recipe File, is published on the Ajinomatrix blog (see about Ajinomatrix)

AJINOMATRIX SRL
Drève de la Chapelle, 34
1430 Rebecq
BELGIUM

https://www.ajinomatrix.org

Press contact: François Wayenberg - Founder and CEO
+32 470 596 768
francois.wayenberg@ajinomatrix.org

Ajinomatrix develops technologies for the digitisation, modelling and interoperability of sensory and food-related information. Its work includes sensory representation (the MP6 format and its public studio), food-process intelligence, knowledge graphs and tools connecting product formulation, perception and human validation.

Website: https://ajinomatrix.org Try the translation tool: https://www.jrf.recipes · https://www.ajinoverse.studio White paper: https://www.ajinomatrix.org/post/beyond-ingredients-and-instructions-why-ajinomatrix-is-building-jrf-for-recipes

This release was published on openPR.