The results below are historical owner- or repository-reported
snapshots, not independently rerun for this page. Numerical-verification and AI evidence is
pending; A World documents are publicly available through the repository links below.
01
What we do
Three projects, one discipline: keep the claim close to its evidence.
PROJECT 01 / NUMERICAL
VERIFICATION
Follow the computation that actually ran.
We investigate how to trace executed numerical operations,
reconstruct their order, and account for rounding and error bounds in a specified
computation.
01Execution trace
02Operation order
03Error bounds
Research focus: numerical execution, not proof
that a physical model describes reality.
A Korean conversational AI project under development. The
supplied snapshot describes a small GRU prototype for preliminary relational
classification in controlled experiments.
30,450parameters in the reported GRU prototype
Owner-reported snapshot; evidence pending. This
is not a freely conversing model or a ready-to-use chat service.
A world grounded in physics. A researcher inside it.
The goal is to build A from laws grounded in real
physics, then let researcher B observe and experiment within A without real-world
data or privileged answers. People outside the world would test whether findings
from A hold in reality.
Life, intelligence, and civilization
are not prescribed outcomes. A finding inside A is not a real-world discovery until
it is tested in reality.
CURRENT SCOPE
A / active
Physical and numerical foundations are the current
priority.
L + B / later stages
Life-observation work (L) and the full internal researcher
(B) remain later-stage goals, not cancelled projects.
Design boundary
The general atomic/electronic world contract is approved
design, not implemented or certified. This status applies to that contract, not
to all existing A code.
Read every result with its conditions and its limits.
REPORTED RECORDS
These are historical project reports, not a new independent audit.
R.01–03 await evidence links; the A World record cites the public repository snapshot
below.
R.01EVIDENCE PENDING
A traceable Hénon–Heiles run
The supplied record reports a
100-step run with a 100/100 outcome and a separate checker returning
CHECKER_PASS.
97,762machine-instruction records
2,214numeric operations
reconstructed
What was tested
Tracing and reconstruction of numerical execution.
Conditions
Gala 1.12.0; specified Hénon–Heiles configuration; 100
steps. Full inputs and runtime details are not published here.
Reported result
100/100; 97,762 records; 2,214 reconstructed numeric
operations; CHECKER_PASS from an independent checker implementation.
Does not prove
An external independent audit, universal error detection, or
validity of the underlying physical model.
R.02EVIDENCE PENDING
The same formula can take a different path.
What was compared
Source-level and binary-level operation order for the same
formula.
Conditions
Specific source/binary comparison in the supplied project
record; reproducible configuration pending.
Reported result
A rounding difference associated with a difference in
operation order.
Does not prove
That every compiler, binary, or execution environment has the
same behavior.
R.03EVIDENCE PENDING
A bound also defines when to stop.
What was tested
Longer-horizon regular-motion and chaotic-case runs.
Conditions
Fixed test configurations and a specified criterion; the
full criterion and artifacts are not yet available here.
Reported result
Regular motion met the criterion at 100,000 steps. The chaotic
case hit the bound and returned refusal.
Does not prove
Unlimited valid prediction, general stability, or successful
verification of the refused chaotic case.
R.04 / AREPOSITORY-REPORTED
Keep the disagreement in the record.
The repository reports a two-atom
scattering comparison with eight agreements and one discrepancy. The original
outcome remains DISAGREE / STOP, not an overall pass.
A specified two-atom scattering calculation and an independent
numerical reference.
Conditions
Repository snapshot 15ae151 (6 October 2026); fixed O-1
protocol and sealed inputs. Not independently rerun for this site.
Reported result
8 agreements, 1 discrepancy, then STOP. A later diagnostic
supports time discretization as a contributor; it does not rewrite the original
failed outcome.
Does not prove
General physical validity, new life, replication, evolution, or
a discovery confirmed in reality.
03
The evidence record
A result should be inspectable and reproducible within its stated
scope.
Evidence has a status, too.
The checklist tracks evidence not yet linked to the
numerical-verification and Korean AI reports. A World has commit-pinned repository
documents below. Those records are publicly accessible but have not been independently
rerun for this site.
These boundaries are part of the result, not a footnote.
01
Numerical verification ≠ physical truth
Checking an executed computation is separate from establishing that
its physical model, assumptions, or inputs describe the real world.
02
Results stay within their conditions
The reported tests concern particular inputs, software versions,
and execution configurations. They do not establish universal correctness or 100% error
detection.
03
Cloud verification remains unfinished
In the supplied snapshot, GCP reconnaissance was implemented, while
the formal profile and N1 work were unfinished. This page has not verified a newer
status.
04
Conversational AI is still a prototype
Preliminary relational classification in a small GRU is not
evidence of open-ended conversational ability. No freely conversing model or public chat
service is available through this page.
05
A is a foundation, not a finished world
The current atomic experiments do not establish new life,
replication, evolution, or discoveries in reality. L and the full B researcher remain
later stages. Deterministic execution alone does not establish realistic physics.
06
Quantum outcome definiteness is unresolved
The general atomic/electronic contract does not establish how to
produce one definite quantum outcome while faithfully representing real physics. Its
approved design is not implemented or certified, and a virtual-laboratory result does
not certify an evolving world.