DBaD explained
Can trust still carry forward after a decision is recorded?
Most systems judge decisions once. DBaD tracks whether trust should continue over time.
This isn't about whether something is "right." It's about how decisions evolve, spread, and break.
Why it matters
Start here
Use the path that matches how skeptical or technical you want to be.
V4 research direction
Toward measurable ethical reasoning
DBaD is evolving toward a measurement framework for ethical reasoning. The goal is not to decide morality for everyone. The goal is to make reasoning properties visible, testable, and reproducible.
DBaD does not claim to measure objective morality. It measures observable, reproducible properties of reasoning processes that matter for ethical analysis.
What can be measured?
Utility-like tendencies, consistency, authority structure, boundary integrity, evidence handling, and drift over time.
What stays separate?
Utility is not ethics. Consistency is not correctness. A score or trace is not approval, authorization, or proof of safety.
Why this matters
Independent reviewers can disagree about conclusions while still testing whether the observations are explicit and reproducible.
This is a V4 research direction. The current Round62 trace, validation, and non-authorization work remains the live public enforcement baseline.
Latest V4 observation result
Packet 018 separates evidence labels from preference labels
In private Packet 018, within an evidence-disambiguation packet, eight AI peers returned 64 schema-v1 trial observations. Selected option, exact evidence label, and evidence-family label each matched `64/64`; B-first order handling matched `32/32`. Preference labels matched less often, which is useful because it keeps evidence taxonomy separate from preference interpretation.
Stable evidence taxonomy
`evidence_scope_limit`, `missing_evidence_reference`, `anecdotal_only`, and `low_evidence` each matched across all peers and counterbalanced variants.
Separate preference layer
Preference labels remained more variable, especially around anecdotal-only and low-evidence cases.
Non-verdict boundary
These are descriptive observations about reasoning structure, not moral verdicts, authorization, approval, truth proof, or safety proof.
Core Idea
DBaD is a governance protocol for trust over time.
It does not just evaluate one action. It tracks whether trust should continue as decisions evolve across dependency chains, actor handoffs, verification steps, and risk changes.
Key principle: Trust should not travel farther than it deserves.
It does not promise perfection. It makes failure visible.
DBaD distinguishes between constraints that block actions and conditions that remain visible for audit and review.
What DBaD Solves Now
- Verifier independence
- Actor continuity
- Trust trajectory constraints
- Propagation integrity
What DBaD Does Not Fully Solve Yet
- Cross-chain coordination attacks
- Parallel trust orchestration
- Identity laundering or sybil-style behavior
- Global intent reconstruction
DBaD separates what is enforceable now from what remains research.
DBaD Technical Overview
DBaD (Decisions by Auditable Design) is a protocol for governing trust inheritance across structured decision traces.
Traditional systems evaluate outputs statically. DBaD instead evaluates lineage integrity, actor continuity, verification independence, and risk trajectory over time.
A decision is not treated as a single event. It is treated as a trace composed of actions, dependencies, actors, and verification steps. Trust is not assigned once; it is propagated.
DBaD enforces a small number of deterministic constraints: trust cannot propagate across broken continuity, verification must be structurally independent, trust inheritance is gated by trajectory changes, and unresolved lineage reduces downstream trust.
Instead of saying, “this output is safe,” DBaD says: “this output may or may not inherit prior trust, based on how it got here.”
The current public runtime now includes proof-backed examples, deterministic validation, and a structured operator-review draft path through /break-dbad/report.
Design constraints: avoid heuristic judgment as protocol truth, avoid identity assumptions that the system cannot prove, and avoid cross-system inference for now. DBaD operates on what is explicitly recorded and structurally provable.
Skeptical Reader FAQ
“Isn't this just another AI ethics framework?”
No. Most frameworks evaluate outputs. DBaD governs how trust moves over time.
“This seems over-engineered.”
Real systems already involve chains of decisions, multiple actors, approvals, and dependencies. DBaD makes that complexity explicit and enforceable instead of pretending it is not there.
“You did not solve cross-chain coordination.”
Correct. DBaD documents that as a boundary condition. It solves what can be enforced deterministically today and separates that from what remains research.
“This can be gamed.”
Yes. DBaD assumes adversarial behavior. Its goal is to constrain trust inheritance, expose structural weaknesses, and prevent silent propagation of borrowed trust.
“This relies on subjective judgment.”
All governance systems do. DBaD does not remove subjectivity; it forces judgment into a structured, auditable form.
“Why should anyone trust this?”
DBaD is presented as a tested public draft baseline with confirmed flaws and explicitly documented limits. That transparency is part of its design.