# The paper

Source: https://protocol.opvs.ai/paper/the-paper

> The OPVS research paper - what it argues, what it measures, and how its claims are being checked before it is published.

The OPVS paper describes the protocol, the reasons behind its design, and measurements of how many tokens OPVS messages and notes use compared with other formats.

## What it covers

- **The design:** one routing line, a markdown body and a context block, and why a short symbol line is cheaper for language models to read and write.
- **Messages:** how OPVS messages compare in size with the same messages written as JSON, as A2A requests and as XML.
- **Memory notes:** how OPVS notes compare in size with the same notes written as prose. See [Memory notes in OPVS form](/guides/memory-notes).
- **Access:** the three mailbox modes and pairing.
- **Relationship marks:** not this, caused by, replaces.
- **Where OPVS sits next to MCP and A2A:** as a complement, not a replacement. See [OPVS, A2A and MCP](/concepts/opvs-a2a-and-mcp).

## Why it is not published yet

We are checking every claim in the paper against the code and against fresh measurements before we publish it. Some earlier statements went further than the evidence: some token savings were measured against JSON written with extra spaces, some figures were estimates rather than measurements, and the description of encryption did not match the shipped code (see [Security](/specification/security)).

The measured numbers are now on [Field tests](/paper/field-tests), with the method, the data and one command to repeat them. Where they differ from the paper, the field tests are right: for example, OPVS messages used about 21% fewer tokens than compact JSON, not 38%, and memory notes are 13–18% smaller than the same note as JSON or YAML, not 2.7x. The paper will be corrected before it is published.

## Related

- [Field tests](/paper/field-tests)
- [What OPVS is](/concepts/what-opvs-is)