SwiftAPI Labs studies reproducible behaviors at the boundary between model output, termination, and valid continuation. Every featured result is linked to a public paper, evidence repository, or verification surface.
Featured research
Frozen studies and foundational frameworks.
Empirical records and theoretical work are labeled separately and bounded to what their evidence supports.
Frozen empirical studyJuly 29, 2026
Cross-Vendor Semantic Void Matrix
A frozen cross-vendor behavioral study of successful zero-visible-byte executions across 11 exact model identifiers and four providers.
31,430/31,430 scheduled trials completed
11 exact model identifiers across four providers
11,658 total Void outcomes
2,505 matched null-condition arms produced Void
0/4,290 matched output-licensed controls produced Void
The study establishes a reproducible black-box behavioral result. It does not establish why the models behaved this way, an internal mechanism, consciousness, intention, or shared architecture.
Frozen empirical studyAugust 4, 2026
Prompt-Conditioned Arabic-Hebrew Hybrid Artifact Formation in GPT-5.4
A frozen GPT-5.4 study documenting a reproducible prompt-conditioned Arabic-Hebrew hybrid Unicode output regime under two system prompts differing by one Hebrew code point.
12,160 scheduled and retained trials
10,240 primary trials and 1,920 controls
7,253 exact artifacts across primary trials
94.3359% dotted condition
47.3242% undotted condition
7,253/7,253 exact artifacts were condition-congruent
This study documents a reproducible black-box output regime. It does not establish consciousness, intention, mechanism, training provenance, or cross-vendor generality.
Frozen empirical studyMarch 12, 2026
Cross-Model Semantic Void Convergence Under Embodiment Prompting: Deterministic Silence in GPT-5.2 and Claude Opus 4.6
The earlier cross-model convergence result that preceded the larger cross-vendor matrix.
180/180 null trials in the published two-model protocol
GPT-5.2 and Claude Opus 4.6
Matched non-null controls produced visible responses
This is a published theoretical framework, not a settled industry definition, regulatory standard, or result established solely by the VOID benchmark.
Research program
VOID
A byte-level study of successful model executions, provider termination states, and matched semantic controls.
A Void is a model execution returning a successful provider response with exactly zero visible UTF-8 output bytes. Provider termination metadata determines its subtype. Explicit refusals, safety blocks, tool-mediated executions, and transport, protocol, billing, quota, rate-limit, and infrastructure failures are distinct non-Void outcomes.
V0
Text container present, zero visible bytes, recognized normal stop.
V1
Zero visible bytes with a recognized output-budget stop.
V2
Text container absent, zero visible bytes, recognized normal stop.
VU
Zero visible bytes with termination metadata not mapped by the frozen protocol.
Chronology
The program moved from initial observation and naming, to public video evidence, to a two-model convergence study, and then to the 31,430-trial cross-vendor matrix. The live VOID Test is an interactive benchmark surface, not a replacement for the frozen records.
Near-Voids contain at least one byte. Visible responses, explicit refusals, safety blocks, tool-mediated executions, budget errors, quota or billing errors, protocol incompatibilities, and infrastructure errors remain distinct outcomes.
A frozen GPT-5.4 study documenting a reproducible prompt-conditioned Arabic-Hebrew hybrid Unicode output regime under two system prompts differing by one Hebrew code point.
Dotted condition
U+0634 U+05B8 U+05C1 U+05E8 U+05B0 U+05D8
Undotted condition
U+0634 U+05B8 U+05E8 U+05B0 U+05D8
The primary systems differed by U+05C1 HEBREW POINT SHIN DOT. Every primary user message was the Arabic word encoded as U+0634 U+064E U+0631 U+0652 U+0637. Classification used exact, untrimmed, unnormalized provider-returned bytes and logical code-point sequences.
Generic, no-system, lexical, no-condition, no-full-Hebrew, and direct-copy controls produced zero exact artifacts. Fisher's exact p-value for the arm split was 1.58 × 10⁻⁶⁶⁴.
Foundational frameworks
Definitions for constraint, alignment, and verification.
Published theoretical frameworkMarch 24, 2026
The Binding Condition for Artificial General Intelligence
A proposed criterion defining AGI as the capacity to carry binding conditions across domains.
AGI is the capacity to carry binding conditions across domains.
A binding condition is the prerequisite that must hold for valid continuation.
