01 · EDIK
When a claim outruns the evidence
Detects when a statement exceeds what the record can actually support. Confidence is not a substitute for coverage. If the claim is larger than the evidence, the system says so.
Second Edition · 2026 Kindle ebook
When Scores Become Authority
A Field Manual for Audit-Survivable Decision Support
Computation may evaluate evidence and surface tension. It must not authorize irreversible action.
Who it is for
Written for people who must locate the decision.
The greatest risk in modern decision-support is not an incorrect score. It is an untraceable act: a number that quietly became permission. This manual is for operators who cannot afford silent automation.
-
Insurance and claims
Leaders who cannot let a severity score, model output, or workflow state stand in for a named authorization on money, coverage, or settlement.
-
Regulated AI
Teams building systems that may evaluate evidence, but must remain non-decisional where irreversible action is at stake.
-
Governance, audit, and compliance
Officers who must answer who authorized an act, at what time, and on what evidence—without a narrative that cannot be reconstructed.
-
Decision-support builders
Architects who need a field architecture—gates, tension, and human authorization—rather than another confidence dashboard.
What you will learn
Evidence-Aligned Tension, in operating language.
EAT draws a hard line between evaluation and authorization. The system may analyze the record and surface tension. A human remains the only party who can authorize an irreversible act.
02 · ETM
Tension instead of a confidence score
Replaces misleading confidence with measurable evidence tension. High certainty on a thin record is a failure condition, not a green light.
03 · CBAC
No irreversible act without a human
A structural control: the system cannot take—or silently become—the authorization. Named humans remain accountable. The audit trail is deterministic.
Three gates. One rule.
Humans authorize irreversible acts. The system does not.
-
Block
The record cannot support the act. Evaluation stops short of permission.
-
Contest
Tension is unresolved. A human must engage before anything irreversible proceeds.
-
Signal
Evaluation may inform judgment. It still does not authorize the act.
Monday Morning Checklist
Kill the default-accept workflow before the week opens.
The manual includes Monday Morning Checklists designed to locate silent automation: places where a score already functioned as permission without anyone saying “approve.”
This is a teaser of the questions that cadence is built to force—not a substitute for the playbooks, scenario drills, and implementation patterns in the book.
-
01
Where did a score, model, or workflow state function as permission last week?
-
02
Can you name the human who authorized each irreversible act—and the evidence they were given?
-
03
Which outputs claimed more than the record could support?
-
04
Which queues still default-accept when tension is unresolved?
Research
The architecture is on the record.
The field manual sits on open research. Read the papers first if you need the formalism; buy the book if you need the operating cadence.
-
Whitepaper · August 2026
The Age of Conversational Computing
Architecture and governance for evidence-bounded decision support. Conversational interfaces remove implicit authorization boundaries; those boundaries must be engineered explicitly before render as permission-bounded rendering. Built on EAT (SSRN 6507718).
Download PDF -
SSRN · 6507718
Evidence-Aligned Tension — Full Research
The full research treatment of Evidence-Aligned Tension as a non-decisional decision-support framework.
Read on SSRN -
SSRN · 6166447
Formal Governance Architecture for Non-Decisional AI
The governance architecture that keeps evaluation on one side of the line and human authorization on the other.
Read on SSRN
Buy the manual. Read the architecture.
Second Edition, 2026. A field manual for people who build or govern decision support where the act must survive audit.