# Epistemological method — public v04

# Epistemological framework and ACGR

The epistemological framework distinguishes claims, assumptions, evidence and uncertainty. It informs the ACGR methodology used here: formulate explicit claims, challenge them through proof inspection, counterexamples, primary-source comparison and reproducible checks, then record the revision and its reasons. Author revision and adversarial review are distinct roles within one inquiry. Role separation and repeated tool-assisted checking do not establish independent human acceptance.

## Research context and provenance

The research originated in a modular QBP implementation developed in Simulink/QUARC during a first internship in the second year of an engineering program. Shokoufeh Naderi's supplied centralized MATLAB simulation was the practical algorithmic reference. Maude J. Blondin supervised the internship work but did not approve the extrapolation to real-time agent addition and removal, which had not been mathematically demonstrated. The supplied simulation, subsequent implementation, mathematical model and browser illustration are distinct objects.

The implementation identified a communication-delay problem; Naderi developed the subsequent refinement. The present manuscript addresses membership-state maps and conditional priority restoration, not a combined delayed-and-open-membership protocol. Integration of the final delay refinement before the internship ended remains unresolved in the available record. Detailed conversation history is retained in the source provenance record rather than repeated in the public research description.

## AI use

AI assisted coding in the original internship implementation; research direction and implementation decisions were not delegated to AI. In the later research process, AI assistance extended to archive analysis, literature work, mathematical formulation and proof drafting, counterexample and test generation, adversarial review, manuscript preparation, the browser simulation and bilingual website. The supplied algorithm and earlier implementation records were inputs to that later process. No claim of independent human proof approval follows from these uses. Responsibility for the manuscript and its claims remains with the author.

## Inspectable revision

Record what changed and why: identify the prior claim, relevant new evidence, the inference it supports, the affected scope and remaining uncertainty. Use primary text for structural comparisons. Distinguish a change to each agent's coordinate space from a change to the number of fixed-dimensional blocks in a global state. Metadata-only comparisons remain provisional.

Use concise research prose in public. State supported facts directly and isolate actual uncertainty; repeated personal attribution is not a substitute for precise evidence boundaries. Do not erase real distinctions by merging known artifacts into vague wording. Keep detailed records and immutable prior versions for auditability.

## Publication and language

The manuscript is a working preprint. English/French interface availability does not imply that both manuscript translations exist. State current PDF availability explicitly, even in notices anticipating a future translation. A public preview is not an invitation to formal review. Each numbered package retains its checksum binding and predecessor history. Translations change presentation, not mathematical assumptions, formulas or computational state.

## Evolution

### v01 — Evidence and role boundaries

Mathematical proof, software behavior, literature and recollection were separated because evidence for one claim type does not establish another. Author revision and adversarial review were recorded as distinct roles.

### v02 — Provenance and public readiness

Algorithm dependence, implementation-led exploration, supervision and approval were distinguished. Draft notices were placed inside public PDFs because files can circulate without the website’s status notice.

### v03 — Source specificity and AI disclosure

Primary-source comparison now identifies carrier spaces and the semantic role of weights. Public prose states supported facts concisely rather than using a biographical narrative. AI use is disclosed by activity and phase; validation records bind each execution to exact bytes and revision context.
## Publication clarification on physical testing

The original internship QBP implementation, before Naderi’s delay refinement, ran in simulation and on physical QBot robots. The current formula and browser implementation have not yet been validated on physical robots. That validation is planned separately; no future result is presumed.


## v04 — Supervision and approval

The previous summary joined supervision and membership exploration in one sentence, which could imply approval of the extension. This correction restores the distinction: internship supervision did not extend approval to an extrapolation that lacked mathematical demonstration. The supervisor’s awareness of the claim and demand for proof remain distinct from approval. The mathematics and physical-testing chronology are unchanged.