For AI-amplified engineering work. Audit-grade substantiation for R&D tax credits, investor reports, and provenance chains — produced from structured event logs that conform to an open public specification.
The 60× figure above is a conservative reference case. On AI-first projects tracked in production — measured against the scope of work a traditional team would have planned — observed compression ratios have exceeded 500×. The reference case is what to model against. The field is producing materially higher numbers.
The tools for measuring and substantiating AI-amplified engineering output haven't caught up with the compression. Founders, CPAs, and auditors all face versions of the same problem.
IRS Form 6765 substantiation requires narratives for permitted purpose, technological nature, uncertainty, and experimentation — per work item, with supporting evidence. AI-augmented work doesn't fit time cards or commit counts. Without structured capture, the credit gets left on the table.
Your team is one human and a language model. Your velocity looks impossible on a spreadsheet. Without defensible data on compression ratios and capacity-equivalent output, your story doesn't survive technical due diligence.
A transcript and a commit log are not the same thing as a signed, time-stamped, independently verifiable provenance chain. Auditors need the chain, not the artifacts that fed it.
From session to substantiation. Every number on every generated report is traceable back to the events that produced it — not reconstructed, recorded.
AI sessions, commits, pull requests, manual work logs, explicit architectural decisions. Each event is timestamped, signed, and written to an append-only store conforming to the GAAIM envelope.
The effort engine compares each session against a benchmark database, applies expertise multipliers, and produces confidence-scored equivalent engineering-weeks. Human corrections feed back into calibration.
IRS four-part test narratives, investor engineering summaries, CPA work papers. PDF, XLSX, CSV, JSON. Any of 22 languages. SHA-256 integrity hash on every file.
The same underlying event log produces the artifact each reader needs, in the format they read, in the language they work in.
Evidence for the credit, the raise, and the board meeting. Stop hand-waving about AI productivity. Start showing the numbers.
Structured IRS four-part test narratives per work item. Work paper exports in XLSX and CSV. Tamper-detecting hash on every report.
Engineering velocity, compression ratios, decision logs, provenance chains. The data room you wish every portfolio company had.
GAAIM — Generally Accepted AI Metrics — is the open event model underneath ForgeTrack, published as a reference specification under Creative Commons at gaaim.org. No vendor lock-in. No black boxes. CPAs and auditors can verify every claim against a document that does not belong to us.
If ForgeTrack disappeared tomorrow, the specification and its reference implementations in TypeScript and C# would remain — stable, deterministic, and ready for another implementer to pick up. That is the kind of commitment a CFO needs to hear before trusting a platform with R&D tax filings.
Read the specification →ForgeTrack generates the four-part test narratives, qualified expense breakdowns, and evidence indices that IRS Form 6765 substantiation requires. Every report carries a SHA-256 data version hash so post-generation tampering is detectable. We recommend CPA review on every filing; the product is designed to make that review fast.
The Effort Engine compares each session against a calibrated benchmark database of common engineering tasks. You see the benchmark, the AI's estimate, the confidence score, and any human corrections. Every number is traceable to its inputs.
ForgeTrack is multi-provider by design. Claude, OpenAI (GPT-4o and GPT-4.5), Gemini, Grok, and Ollama are supported today, along with a custom-endpoint option for self-hosted and enterprise models. If your provider isn't on that list, tell us — every request routes directly to engineering, and new providers are prioritized by customer demand. Email [email protected] with the provider name and your use case, and we'll reply with a timeline.
Yes. PDF, XLSX, CSV, and JSON on every report. A programmatic API for integration with external tools and data rooms. Your data is yours; the export formats are not proprietary.
Both. Contributors can be humans or AI agents, each with expertise profiles. Role-based access covers owner, admin, member, and viewer. Multi-tenant isolation is architectural, not a setting.
Free tier. No credit card. Fifteen minutes to your first report.