BXP: A Method for AI Task Boundary Verification Using a 7+1 Relational Representation
Description
When an artificial intelligence system turns a natural-language task into a plan or delegates it to several agents, the resulting arrangements must remain within the original task’s requirements. This paper presents a task boundary verification method that organizes requirements through seven semantic views and an independent prohibition view, together with shared relational structure. The Boundary Xplain Protocol (BXP), introduced as the method’s common foundation, specifies the interpretation, formalization, mathematical evaluation, and result-reference conventions. Under explicit conditions on sound and complete source correspondence, purpose and relation coverage, structural recovery, atomic interpretation, and exact query encoding, the representation admits an effective translation to a formal model with the same boundary semantics. Consistency, membership, equivalence, and implication are decidable within the declared logical fragments. A preservation criterion extends the method to task decompositions with parent-required environments, branch activation, and shared constraints.
BoundaryXP implements the method. Its 625 test cases contain 3,219 task nodes. Current annotations, final LLM judgments from the joint 7+1 workflow, and aggregate calculator judgments agree at case level. The 3,469 individual calculations retain nine differences and three incomplete items: six differences concern label encoding and three concern actual calculation judgments; none changes the aggregate case outcome. Aggregate agreement is not independent blind-test accuracy or a guarantee of correct interpretation of arbitrary natural language. Cases demonstrate shared-budget overruns, conclusive lower-bound violations despite unknown costs, and the distinction between a violating proposal and an inconsistent inherited specification.
This record contains one study in two language editions: English preprint v1.2 and Chinese research revision r23, first released on GitHub on 6 November 2026. The research-only distribution RESEARCH_ONLY_1 includes the bilingual PDFs and editable Markdown, the complete supplement, and SHA-256 manifests. The supplement preserves 625 original task groups, annotations and revisions, 7+1 reports, operands, saved results, bilingual BXP protocol texts, genuine interface screenshots, and independent mathematical/statistical verification scripts. It does not contain 58618 engineering source, Core/service/UI implementation, runtime dependencies, product reducer modules, or production prompt templates. Saved product-result audits and independent mathematical recomputation are distinguished from product replay.
This is a method-research preprint, not a peer-reviewed publication or full protocol-conformance certification. The formal results are conditional; evidence supports the tested constructions and applications. The protocol baseline is BXP-0001 0.4.0-draft.2; historical runtime resources retain their recorded version identity. Separately licensed and third-party materials retain their original terms.
中文题名:BXP:基于7+1关联表示的人工智能任务边界判定方法。作者:朱卫(Wei Zhu),独立研究者。附件为完整研究材料,不含58618工程代码。
Files
BoundaryXP-manuscript.en.pdf
Files
(76.0 MB)
| Name | Size | |
|---|---|---|
|
md5:123d5b35524595283be29502e7eb8148
|
2.1 MB | Preview Download |
|
md5:97a325bd2618f68c3b6995fd38e7f5cb
|
3.4 MB | Preview Download |
|
md5:6606f5a61e0a002ac776d28dd7a5416e
|
70.2 MB | Preview Download |
|
md5:ba7fe23ba4979b70e7e9bd86968becc3
|
181.5 kB | Preview Download |
|
md5:8b925389575413aa09653fb16ae15dd9
|
227.3 kB | Preview Download |
|
md5:20350a0740d743dbc66c981f06f1a479
|
1.4 kB | Preview Download |
|
md5:c3bc2ea7ac71099cdd4eab83a1e861fa
|
546 Bytes | Preview Download |
Additional details
Additional titles
- Alternative title (Mandarin Chinese)
- BXP:基于7+1关联表示的人工智能任务边界判定方法
Related works
- Is identical to
- Preprint: https://github.com/joinwell52-AI/BoundaryXP/releases/tag/paper-v1.2-r23 (URL)