AN INVENTOR'S FIELD NOTES OCTOBER 2026

THE QUESTION THAT STARTED UAEP

The machine said “done.”
I said: show me.

When AI began doing real work beyond the chat window, a successful command was no longer enough. The destination itself had to tell the truth.

Execution success
≠
verified destination reality.

THE UAEP DISTINCTION
Conceptual illustration of UAEP and the wider Infinity ecosystem; not a technical architecture diagram or test evidence
Conceptual visual · each symbol moves within its own boundary
THE CORE QUESTION · 01 / 17UAEP

Assurance of an authorised result at the external destination.

A successful command is not enough; the intended destination still has to be checked.

Public conceptual explanation—not a live product demonstration.

NOT ANOTHER CHATBOT STORY

A model can say “complete.”
The world may disagree.

Mangan was building AI intended to operate complex professional software and carry work across time. A workflow could report 100% success while the wrong target had changed, the result was incomplete, or the command never reached the intended destination.

That gap reaches beyond AI. Any consequential system can confuse a command, a receipt, an indicated state and the actual result.

THE PUBLIC ASSURANCE QUESTION

Did the authorised objective actually become true at the destination?

01

What was authorised?

Capability and a valid tool call do not create authority.

02

What happened in reality?

A system's success report is not, by itself, independent destination evidence.

03

Can “VERIFIED” be justified?

If the evidence cannot establish the objective, uncertainty must remain visible.

FEATURED STORY

One false “done” became a much larger map.

Mechanical thinking, software execution, unexpected failure cases, public falsification challenges and a platform that asks reality for the final answer.

Read Adam's story
THE ENGINEERING RULE

Activity
≠ progress
≠ verified completion.

THE IDEAS BEHIND THE ENGINEERING

Strange sayings.
Useful questions.

Mangan used simple images from machinery and movement to pressure-test systems that were becoming difficult to reason about.

Illustration of a steel bridge carrying force around a broken support
01 / THE LOAD PATH

Capability needs a path to a justified outcome.

Something doing work is not the same as proving the intended result arrived safely.

Illustration of an alternate illuminated route through a blocked mechanical maze
02 / NO DIRECT PATH

No direct path does not mean no path.

The point is to find a legitimate route—not to pretend a blocked route succeeded.

Illustration of a machine repeatedly circling without reaching a distant goal
03 / THE CIRCLE OF LIFE

Activity is not progress.

A loop can be busy forever while the authorised destination remains unchanged.

Read where these ideas came from

EVIDENCE CEILING

Built. Challenged.
Still open to being tested.

UAEP is implemented. Mangan reports bounded internal adversarial qualification, including a final 29-case synthetic execution-graph campaign with zero false VERIFIED in that declared test environment.

Those are internal results, not independent certification, production security proof or a universal guarantee. The implementation remains confidential while commercial options are explored.

Read the public challenge

FOLLOW THE PUBLIC RECORD

The questions and answers are visible.

These links document public discussion and coverage. Forum participants proposed challenges; they did not operate or independently certify UAEP. The qualification figures are my bounded internal results.

01 / THE ORIGINAL QUESTION

Execution success is not verified reality.

I published the assurance question and invited reproducible false-VERIFIED cases.

Read the original post ↗
02 / THE FIRST BOUNDARY

Historical truth is not current authority.

Penguink raised objective and authority supersession. My public reply acknowledges a real boundary gap and reports the bounded successor result.

Read the exchange ↗
03 / THE NEXT HANDOFF

Planning expansion is not authority expansion.

MoodND asked what happens when a later worker consumes an earlier receipt. I reported a broader 29-case successor campaign and its limits.

Read the exchange ↗
04 / EDITORIAL COVERAGE

KI Spotlight covered UAEP.

The German-language report names Adam Mangan and explains the public distinction. It is coverage, not independent product validation.

Read the article ↗

INTERACTIVE SCENARIO LAB

Press replay.
Watch “done” diverge from reality.

Three synthetic, side-by-side examples show why execution claims and assured outcomes are different. These are explanatory animations, not a live UAEP evaluator.

Open Scenario Lab
THE UAEP GUIDE

A character you can turn around.

Loading the interactive 3D model…

Public visual character only. It does not depict UAEP's implementation.