Show HN: A hash-chained ledger for AI reasoning you can verify yourself
SDI Protocol launches a hash-chained reasoning ledger for AI agents, with a public agent called Chromite running uninterrupted on one chain since spring 2026.
SDI Protocol, a solo-built open protocol for verifiable AI reasoning, launched publicly on August 25, 2026, via a Show HN post by its creator. The system records each step of an LLM's reasoning as a hash-chained ledger that anyone can pull and independently verify without an account or SDK.
The protocol works by running a language model one turn at a time through what the project describes as a "reasoning grammar" — a structure that is both natural language and algebraic, per the project site. Before any reasoning act is committed to the ledger, it passes through a compile gate that checks safety floors, a workfloor constraint, and grammar conformance. The result of each committed act is a sealed, immutable record; sealed records cannot be rewritten, and each permanently carries the formula from the schema era in which it was created. An argumentation graph sits on top of the chain, serving as the agent's retrievable memory and allowing the agent to hold or defeat prior reasoning acts to keep context under control.
The project's public agent, Chromite, has been running on one continuous hash chain since spring 2026 and has used multiple models across three frontier providers, according to the Show HN post. Three ethical constraint floors are built into the gate: DECLARATION_INTEGRITY, which has fired at least once, and two conjunction-based safety guards — PRIMUM and Absolute Value Guard — which by design admit no human attestation and have not yet been triggered. The refusal path runs on a separate hash chain ledger. The creator states the verify script detects each entry's schema era and applies the correct math when recomputing hashes.
The builder, posting under the handle dj_johnsonMid, describes a background of 20-plus years as a Marine working in command and control, with the system drawing on decision doctrine and intelligence-gathering techniques for evidence handling rather than a computer science lineage. The build is described as a solo project roughly three years in the making. A GitHub repository at StructuredDecisionIntelligence/Structured_Decision_Intelligence contains verification scripts with checksums and a committed offline test record.
The protocol is currently free and open. The project site discloses that a commercial product, SDI Reckoner, is planned at sdireckoner.com, with private Reckoner commissions opening in late fall 2026. The creator is explicit that nothing in the open protocol layer requires the paid product to verify any reasoning act — Chromite exists as a public transparency exhibit for the protocol itself. Feedback is directed to support@sdi-protocol.org.
No comments yet — start the thread.