Hardware Research

Hardware Research

This page is the project gateway to the public hardware-oracle method. The research site remains the canonical findings and experiments registry.

Evidence path

flowchart LR
    A["Narrow public question"] --> B["Source-owned probe<br/>exact revision + identity"]
    B --> C["Transport qualification<br/>PSPLink / capture intact"]
    C --> D["Physical PSP run<br/>model + firmware recorded"]
    D --> E["Raw bounded result<br/>repeat count preserved"]
    E --> F["Scoped interpretation<br/>unknowns retained"]

Loading

Important

PPSSPP agreement, software-vs-Vulkan parity, or two host implementations agreeing can validate a probe/comparator chain. They are corroboration—not PSP silicon truth.

A recompiler and its reference model can share the same wrong assumption. Qualified physical PSP measurements are therefore stronger evidence for selected return values, timing, wake behavior, state transitions, and bounded media/display contracts.

Transport is not semantics

Observation	What it establishes
Transport succeeds	The connection and capture path worked for that attempt.
Transport fails	No semantic measurement can be accepted from that attempt.
Probe returns a value	A candidate result exists; source/binary identity, environment, repeatability, and interpretation still need review.
Qualified repeated result	Bounded hardware evidence for the exact measured contract—not a general platform rule.

The current planned physical test device is a Japanese-region PSP-3000 configured and presenting as U.S., firmware 6.61-ARK. Historical records may retain their recorded PSP-3001 identity; those records must not be rewritten to match the current device.

Public-safe replication path

1 · Choose	2 · Run	3 · Report	4 · Review
Start with a narrow question or replication request.	Use a source-owned synthetic probe and qualify transport first.	Share small scalar outputs, environment scope, and repeat count—never private title material.	Separate observation from inference and preserve remaining unknowns.

Research portal

Method	Work queue	Canonical records
Hardware Oracle Methodology How to Run an Experiment	Hardware Replication Replication Wanted	Findings Registry Experiments Registry Contradictions Registry

Caution

Publish only redacted, reproducible, public-safe artifacts. Keep commercial inputs, derived captures, private traces, identifiers, and private paths local. Use Submitting Results Safely before sharing.

The repository-side protocol is maintained in docs/HARDWARE_ORACLE.md.