Planned Experiments: Open, Discriminating, and Safe
Registry record
Stable ID: EXP-PLANNED-REGISTRY-001
State: PLANNED. This registry contains open questions only; it has no results, acceptance claims, or inferred outcomes.
Registry: PLANNED EXPERIMENTS · Record family: source-owned, non-destructive proposals · Rule: no hardware run means no measured result.
These records turn mature unresolved questions into discriminating procedures. They are PLANNED, not PASS, and contain no fabricated outcomes.
EXP-THREAD-UID-REUSE-001 — ThreadMan UID reuse and object lifecycle
Question: When a thread is deleted, when can its UID be reused, and what do stale handles return before and after reuse?
Competing explanations: immediate reuse; generation-tagged delay; stale-handle error until an epoch boundary.
Observable: documented status/result scalars for create, exit, delete, query, wait, and the next allocation. Hardware needed: PSP-3000 plus PSP-1000/2000 replication. Fixture: minimal source-owned ThreadMan probe. Safety: owned user threads only; no kernel memory inspection.
EXP-GE-SIGNAL-ORDER-001 — GE SIGNAL/FINISH callback ordering
Question: Does a SIGNAL or FINISH callback become observable before list completion, at completion, or only after a display/scheduler boundary?
Competing explanations: command-list order; callback queue order; VBLANK/display coupling.
Observable: source-owned list markers, callback scalar log, list-sync result, and VBLANK counter under deterministic synchronization. Hardware needed: PSP-3000, then second model. Safety: small owned lists; no arbitrary GE/MMIO writes.
EXP-CACHE-DMA-SMC-001 — Cache/DMA/self-modifying-code visibility
Question: Which writeback, invalidate, and executable-write sequence makes a CPU or DMA observer see updated bytes?
Competing explanations: D-cache writeback requirement; I-cache invalidate requirement; uncached alias visibility; transfer completion timing.
Observable: source-owned RAM pattern, cache-operation sequence, first instruction/data read, and delayed readback. Hardware needed: PSP-1000/2000/3000 matrix. Safety: owned RAM and source-owned code pages only; no arbitrary executable writes or MMIO.
EXP-ATRAC-PRIMING-001 — ATRAC first-frame and reset priming
Question: How do GetNextSample, first DecodeData, refill, reset, loop, and non-frame-aligned positions affect output and sample position?
Competing explanations: codec priming; PSP buffer contract; loop-tail state; bridge reset error.
Observable: source-owned legal ATRAC fixture, decode counts, sample position before/after, and normalized PCM hash. Hardware needed: PSP-3000; cross-model if available. Safety: no retail media; use redistributable or source-owned fixture only.
EXP-VFPU-PREFIX-INVALID-001 — Invalid-form prefix consumption
Question: Does an illegal vector size/alignment or rejected memory form consume S/T/D prefixes?
Competing explanations: prefixes are consumed before validation; validation is failure-atomic; only selected prefixes are consumed.
Observable: unique register/prefix state before and after a safe invalid instruction, plus exception/result code. Hardware needed: PSP-1000 and PSP-3000. Safety: controlled user-mode invalid forms only; no intentional crash or privileged operation.
EXP-PSPREL2-TOOLCHAIN-001 — PSPREL2 cross-toolchain fixture
Question: Which relocation encodings and GP/load-base combinations are emitted by documented PSP toolchains, and how should malformed records fail?
Competing explanations: one format family; toolchain-era extensions; loader-specific interpretation.
Observable: tiny source-owned ELF/PRX fixtures, relocation table bytes, rebased pointer values, and fail-closed parser results. Hardware needed: none for the parser fixture; hardware only if a loader semantic remains ambiguous. Safety: synthetic/open-source binaries only.
EXP-ASYNC-IO-CANCEL-001 — Async-I/O cancellation and close race
Question: What result, callback, and file-position state is observable when an async request is cancelled or its descriptor is closed while pending?
Competing explanations: cancellation wins; completion wins; close defers resource release; callback is suppressed or delivered once.
Observable: documented scalar return, callback count/order, descriptor state, and source-owned file hash. Hardware needed: PSP-3000 and one independent model if possible. Safety: disposable probe-owned Memory Stick files only.
EXP-DISPLAY-LATCH-001 — Display latch and pixel-format timing
Question: When multiple SetFrameBuf calls queue around VBLANK, which pointer/stride/format becomes visible and when?
Competing explanations: immediate latch; next-VBLANK latch; last-call-wins queue; format-specific behavior.
Observable: numeric framebuffer state and VBLANK count, not screenshots or video. Hardware needed: PSP-3000 plus second model. Safety: owned framebuffer memory only.
Publication rule
Until a planned record has a valid transport, intact capture, and complete provenance, it remains PLANNED, TRANSPORT FAILED, CAPTURE MISSING, CAPTURE CORRUPT, or PROBE INVALID as applicable. Never upgrade it from an expected behavior.
Related: Open Research Questions · Hardware Oracle Methodology · Experiments registry.
Last reviewed: 2026-08-11 · Publication safety: source-owned, non-destructive proposals only.
Index: The planned records below are intentionally separate from completed experiments and remain result-free until a public-safe protocol and evidence review exist.
Additional planned discriminator
EXP-DMAC-BOUNDARY-V6-001 — corrected v6 boundary probe for high-size sceDmacMemcpy tail loss. State: PLANNED. Use fresh instrumentation and publish only aggregate sizes/output digests; no result or exact maximum is implied by this record.
Planned coverage and supersession aliases
EXP-VFPU-WIDE-COVERAGE-001 — complete the 498 unmeasured wide-register encodings. State: PLANNED; no result is implied.
EXP-DMAC-BOUNDARY-001 — legacy boundary-probe label retained for provenance and SUPERSEDED by corrected EXP-DMAC-BOUNDARY-V6-001. No result is implied by either planned record.
Take a planned record into the field: choose a safe target through Hardware Replication or Replication Wanted, then follow How to Run a PSP Research Experiment. A protocol stays PLANNED until a valid transport, intact capture, and complete provenance are reviewed.