Velorin Intelligence

Projects / Decision record

What we are testing—and what we stopped.

Every project begins with a failure that matters. The useful question is not whether the work sounds promising. It is whether the evidence changes a design, supports a result, or tells us to stop.

3 current decisions2 completed records3 stopped, held, or archived

01 / Current work

Open decisions

PASBuilding

The failure

Memory that cannot be corrected is not reliable memory.

A persistent system may retain useful context, but it also needs visible sources, explicit permission, correction, and revocation.

What we are doing
Private engineering of authorized, source-backed, correctable continuity.
Next proof
A remembered fact must be approved, recalled with its source, revoked, and no longer recalled.
Public boundary
No public product, personal-agent identity, or release commitment.
See the public direction
VCETesting

The failure

An evaluator can become more complicated without becoming more informative.

The custom layer is not worth keeping if an established framework produces the same decisions and the same evidence.

What we know
The frozen cases can run inside existing evaluation infrastructure.
Decision
Keep only a reproducibly distinct contribution—or retire the custom runner.
What is not proven
Comparative superiority or a general safety result.
Inspect the decision test
SDPSBlocked before trial

The failure

A run can look finished while its process is still running.

The search harness could certify terminal evidence without proving that its container had actually stopped.

What changed
No scored mathematical search proceeds on the affected harness.
Next proof
Repair termination evidence, independently reproduce it, and restore deterministic CI.
What is not claimed
No new matrix, theorem, nonexistence result, or discovery advantage.
Inspect the qualification record

02 / Completed work

Closed evidence records

VCTCompleted

The failure

A supported claim can conceal missing lineage.

A local release policy performed better on constructed cases, while a reduced external filter showed no improvement in unauthorized-action outcomes.

Decision
Freeze both the bounded positive result and the external negative result.
Limit
No general safety, portability, or superiority claim.
Read the evidence record
CRSFrozen · v1.0.0

The failure

A typed decision can still change when its context changes.

The frozen study tested how instructions, representation, source authority, and option order affected a constrained decision model.

Decision
Preserve the matched experiments, controls, and checksummed artifacts.
Limit
Broad transfer across models and real-world tasks remains unknown.
Read the frozen record

03 / Portfolio discipline

Stopped, held, and archived

CSStopped

The failure

Verification can prove that steps occurred without proving the task was completed.

Constraint Sentinel produced useful assurance signals, but did not establish enough additional value over simpler controls.

Decision
Stop product development. Preserve the negative result and implementation.
Read what failed
7PEELOn hold

The risk

A new validation program can duplicate work that already exists.

The question remains legitimate, but no active experiment is justified without a distinct decision purpose and baseline.

Reopen only if
The experiment does not duplicate the frozen context study or current VCE work.
Read the preserved concept
ASDArchived

The decision

Do not run a broad roadmap after the research value has narrowed.

RH-00 instrumentation was completed. The behavioral and multi-agent sequence was retired before experimentation.

What remains
A bounded instrumentation record—not behavioral or multi-agent findings.
Read the archived record

Decision standard

Activity is not progress. A test earns its place by changing a decision.

We continue work when the next result can change a design or claim. We narrow it when only one component remains distinctive. We stop when a simpler method is sufficient or the expected benefit no longer justifies the cost.