Release of 'DEE (Decision Evidence Engine)', a Protocol that Makes Decision-Making 'Provable'

Logo

'VOICE PROTOCOL' (operated by Japan Realtor Co., Ltd.), which deploys a web service that structures decision-making processes and records them in a retrospectively verifiable format, announces that it has released the verification model of its core technology 'DEE (Decision Evidence Engine)' and open-sourced it (on GitHub) as a reference implementation that allows third parties to independently confirm the concepts and procedures of verification.

■ Background: The Limits of 'Accountability' in the AI Era
Currently, while decision-making by AI is becoming widespread in corporate activities and social infrastructure, the problem of the "black box"—where "why that conclusion was reached" is obscured—is becoming increasingly serious.
Because traditional systems only save "the results," companies are asked for an "explanation" when problems arise later. However, in decision-making involving complex algorithms and human intervention, it has become difficult to gain sufficient trust with unevidenced explanations and post-event reports.
What is demanded is not an explanation, but an "objectively verifiable mechanism."

■ Solution: DEE is a 'Mechanism to Convert Accountability into Provability'
DEE (Decision Evidence Engine) is a protocol designed to record the decision-making process itself so that third parties can verify it later.
It records who saw what, how they judged it, and how they reached a conclusion, in a format that allows third parties to verify whether any tampering has occurred.
"Proof" in DEE refers to "the ability to recalculate the exact same result from the exact same input (event history, snapshot, validation rules)."

DEE is a "verifiable decision logging engine" that operates on the following pipeline:

1. Event (Utterance/Vote/Operation)
2. Ledger (Append-only log)
3. Snapshot (Fixation of state at a point in time)
4. Freeze (Immutability)
5. Bundle (Generation of an evidence trail package)
6. Verify (Hash verification)
7. Evidence (Third-party verifiable output)

Through this process, it ensures recalculability where "the same hash is obtained under the same conditions (deterministic)," building a Truth layer independent of the UI.
*Non-deterministic elements (time, display order, environmental differences, etc.) are excluded from verification or normalized.

Image diagram

■ Decisive Difference from Conventional Systems
- Log → Only records
- Report → Only shows
- DEE → Can be verified

By breaking away from the black box where only results remain, and treating the decision-making process as a recalculable history, it becomes possible to use it as an auditable trail.

■ What DEE Guarantees
DEE does not guarantee the "correctness of the result."
What it guarantees is "that the process leading up to that result can be reproduced and verified later under the same conditions."
Because of this, decision-making is treated not as an "assertion" but as a "verifiable history."

■ Verification Pathway on the Web (/verify) and Open Sourcing
The defining feature of DEE is "that third parties can verify it."
On the published verification portal (/verify), by specifying a target identifier (such as topicId), users can check the verification results (consistency of snapshotHash and reportHash) executed on the server side in their browser.
Furthermore, to ensure the transparency of this verification model, it has been published on GitHub as a reference implementation that allows third parties to independently check the concepts and procedures of verification.
This published implementation is a reference implementation of the verification logic and operates independently of the internal implementation of the production service.

■ Future Outlook: Toward a Society Premised on Verifiability
DEE is not closed off within a single service, but is part of an information infrastructure premised on "verifiability."
Going forward, the verifiability of decision-making is expected to become increasingly important in the following areas:

- Corporate Governance (Board of Directors, request for decision, audit)
- Finance / Credit Judgment (Verification of screening processes)
- AI Decision Making (Verification of the rationale behind model outputs)
- Public Policy / Consensus Building (Ensuring transparency)

In these areas, there will be a demand for third parties to be able to verify not only "what was decided" but also "how it was decided."