A promising result enters your research programme. Three decisions begin to rely on it. Then its assumptions change. Can your team identify exactly what needs another look?
That question gave us a practical role in the September Navier-Stokes developments. The mathematical work belongs to its authors. Our work was to preserve what had been checked, connect it to the exact claim, and make the consequences of a change inspectable.
8DB is an omnimodal database. Its research evidence twin brings together statements, verification events and explicit dependencies. The useful payoff is a more precise review queue: which conclusions use this result, what supports them, and which apparently related results stand on their own evidence?
Start with the exact claim
OpenAI's published repository describes formalizations of forced Navier-Stokes breakdown on all of space and on the periodic torus, alongside a separate unforced Euler result. Those conditions matter. A theorem involving a force does not by itself settle a question that requires the force to be absent.
For this article, the record is our historical review of the pinned September development. It is not a new mathematical assessment or a current prize-status announcement. The Clay Mathematics Institute's problem page is the appropriate starting point for its official description and status.
A useful evidence record keeps several questions distinct:
| Question | Evidence to retain |
|---|---|
| What does the author claim? | Exact statement, source and revision |
| Does the project build? | Toolchain, command, host and recorded outcome |
| What assumptions does the theorem use? | Axiom-audit output for the named theorem |
| Does the proof match the challenge? | Pinned challenge and comparison result |
| Does the challenge express the intended problem? | Clause audit and specialist interpretation |
One successful check should answer its own question. That makes it easier to use a strong result confidently within the scope it actually establishes.
What our September record contains
The preserved verification module records the OpenAI development at commit f9e8bc5b, Lean 4.34.0-rc2, on WSL Ubuntu 24.04. Its events are dated 11 September 2026 and carry an actor, outcome, time and supporting details.
The sequence includes a source audit, a comparison with the pinned Formal Conjectures statement, local builds, axiom audits, Comparator acceptance and fresh kernel replay of the solution modules' import closures. The axiom records name propext, Classical.choice and Quot.sound for the audited headline theorems.
The conditions travel with the results. The recorded Comparator runs used a reduced sandbox on that host, and the solutions had already been compiled locally. Separate kernel replays were subsequently recorded. Repeating a check can add useful evidence; it does not erase the conditions of an earlier check.
This revision reviews those preserved records and their implementation. It does not report a new execution of the proof project. The statement-fidelity article explains the separate work of matching mathematical meaning to the original requirement.
Give each connection a precise job
Our recorded dependency graph connects the Navier-Stokes theorem to the whole-space breakdown statement, then to the periodic consequence. The separate Euler result has its own support and is not placed downstream simply because both results concern fluids.
The distinction is useful well beyond mathematics. Two experiments can share a technique without one depending on the other's result. Two product decisions can cite the same report while using different claims inside it. A change-impact query needs the actual dependency, rather than a broad “related to” link.
In the inspected implementation, the retraction query walks the recorded dependencies and returns the affected descendants. Its source includes a regression case expecting the two Navier-Stokes consequences in that result and excluding the independently supported Euler statements. This is a source-reviewed query contract, not a fresh test-run report.
The query is read-only. It identifies a review set; it does not change the claims or establish that each conclusion is false. Before withdrawing a conclusion, a reviewer must consider whether another admitted derivation still supports it. That distinction preserves useful work while directing attention to what actually changed.
Recorded dependence, not shared subject matter
A change should trigger the right review
Simplified shape of the historical Navier-Stokes evidence graph.
Navier-Stokes theorem
Select this claim to inspect its recorded downstream dependencies.
Whole-space consequence
Included in the affected set because the graph records this derivation.
Periodic consequence
Included through the next recorded dependency.
Outside this dependency chain. A shared research topic does not make it a downstream consequence.
Keep unresolved questions attached to the work
An open obligation deserves its own record. What is missing? Which claim needs it? What evidence would discharge it? Who or what can perform the next check?
The evidence twin represents hypotheses and gaps separately from completed results. Its status rules also distinguish a registry entry, an author statement, a local build and a formal comparison. This gives a reader a route from a status label back to the recorded event that supports it. The event's source still needs to be trustworthy; a typed field alone does not establish that the work occurred.
Established provenance standards already address derivation and responsibility. The W3C PROV family is useful context for that foundation. Our application brings provenance into a research review workflow: retrieve the supporting check, inspect a dependency, identify an unresolved obligation and decide the next action.
Try it on one decision that matters
The first evaluation can be small. Choose one consequential claim, attach its supporting records and map the decisions that actually use it. Then revise one input and inspect the review set. Check that dependent work is found and independent support is preserved.
That is a practical test of whether connected evidence improves your team's work. A growing library is more valuable when its history helps you decide what to trust next.
Bring us one claim your team keeps rechecking. Share a non-confidential example with its sources and one downstream decision. We can scope an 8Braid Reasoning Fabric evaluation around evidence continuity and change impact.
Evidence scope Recorded verification records · Source-reviewed dependency query · Mathematical review remains distinct. Explore the evidence.