SpecLogician is live as of today: https://www.speclogician.dev 
The one-line version: a shared, machine-checkable model of complex software behavior, for coding agents and the humans reviewing them.
The problem it tackles is one this community knows well. In real systems, behavior is scattered across imperfect artifacts — requirements, code, tests, logs, policies, production traces — and a code diff only tells you what changed in the text, not what changed in the behavior. SpecLogician reconciles those artifacts into a formal model you can inspect, diff, test, and reason over, so edge cases surface before they become production failures.
In practice, that means every change — agent-made or human-made — can be reviewed against real questions:
- What behavior changed?
- Which scenarios still hold?
- Which paths became reachable?
- Which invariants were rechecked?
- What witnesses, counterexamples, tests, logs, or traces support the conclusion?
For those of you running ImandraX in regulated or high-stakes environments, that question list doubles as your evidence trail — machine-checkable answers, not review notes.
The end state we’re after: agent-driven development that moves from “the diff looks right” to “the behavior is understood.”
The Medium article goes deeper on how the model gets built and refined:
https://medium.com/imandra/coding-agents-need-a-shared-behavioral-model-of-the-software-they-change-062f2899ccf6
Would love this community to kick the tires first. Throw a real artifact at it — a trace, a test suite, a requirements doc — and post what it surfaces. Questions and feedback welcome below; we’re watching this thread today.