Signed, attributed records of what humans and AI systems produce together — portable across tools, verifiable after the fact, defensible under audit.
GAAIM (Generally Accepted AI Metrics, pronounced "game") specifies a canonical event envelope, a first-class attribution model, a cryptographic signing protocol, and a compliance-level framework — together letting any tool in a knowledge-work stack emit audit-defensible evidence of what was produced, by whom, and with what AI assistance.
This is the event architecture underlying ForgeTrack. It is published here as a reference document for implementers, auditors, and anyone reasoning about the provenance of AI-amplified human labor.
Built as a profile of CloudEvents 1.0, extended with the fields that audit-defensible knowledge-work evidence requires but which generic event formats do not carry.
RFC 8785), a public key registry format, and chained-hash construction linking events into tamper-evident sequences."GAAIM carries the same linguistic fossil as horsepower — the horses have been gone from the measurement for over a century, but the unit persists because it was the right name at the moment the category crystallized."— §note on the name
This specification defines the event model used internally by ForgeTrack. It is published for transparency, implementer reference, and ecosystem optionality.
Edson Technologies does not currently operate a formal standards working group. Issues and pull requests on GitHub are welcome, but response times are not guaranteed and there is no RFC process in flight.
If ecosystem interest emerges — independent implementers, interop requests, or demand from regulators and auditors — the specification will be contributed to an appropriate neutral foundation. Until then, treat it as a published reference artifact.
The full specification, reference schemas, test fixtures, and example code are published openly under Apache 2.0 (implementations) and CC-BY-SA 4.0 (specification text).
GAAIM began as an internal question: how do we quantify human labor amplified by AI? The question mattered because the answers have concrete stakes — R&D tax credit substantiation under IRC §174, venture diligence on engineering productivity, audit trails for regulated work, contribution reporting under frameworks that have not yet been written.
The event model was extracted from the architecture of ForgeTrack, a product built to answer that question for our own engineering work first. Publishing the specification openly does two things: it lets ForgeTrack's provenance claims be independently verified, and it gives implementers a reference if they want to emit compatible events from their own tools.