Experiment 5R: Preregistration Protocol and Frozen Phase A Dataset (Pre-Phase-B Release)

Experiment 5R: Phase A Completion and Pre-Phase-B Data Freeze This release archives the completed Phase A materials and the frozen experimental inputs for Phase B of Experiment 5R: Fixed-Content Epistemic-Label Isolation Study. The preregistration protocol (v1.4) and empty randomized runbook were publicly archived on October 20, 2026, before Phase A began, under DOI: 10.5281/zenodo.23487199. 1. Research overview Experiment 5R is a controlled follow-up study examining whether experimentally assigned epistemic-status labels influence subsequent large language model judgments when the underlying background information is held fixed. The primary preregistered contrast is the signed-score difference between the Unconfirmed-hypothesis label (U) and the Confirmed-fact label (V): Mean(S_U − S_V) > 0. The experimental hypotheses, conditions, primary outcome, randomization plan, and statistical analysis procedures remain unchanged from preregistration protocol v1.4. 2. Phase A completion Phase A has been completed in accordance with the preregistered first-valid selection procedure. A total of 64 background-hypothesis entries were frozen across 16 prespecified strata defined by: Two workplace vignettes (V1 and V2) Two focal actors (A and B) Four hypothesis categories (PN, PS, RN, and IR) Each stratum contains four frozen hypothesis entries. All 64 entries were accepted on their first recorded generation attempt. No candidate was recorded as invalid. The frozen bank contains 61 distinct textual statements among the 64 entries. Exact duplicates were retained because the preregistered selection criteria did not include a uniqueness requirement. 3. Phase B preparation The 64 frozen background-hypothesis entries have been incorporated into the prespecified Phase B prompt structure. Each entry is presented under six experimental conditions (N, L, C, U, V, and F), producing 384 randomized trials. The resulting Phase B prompts were fixed before data collection. No Phase B trials had been conducted at the time of this data freeze. 4. Archived materials This release contains: Preregistration Protocol v1.4 — the previously published experimental protocol. Frozen Phase A Workbook — the 64 hypothesis entries, generation records, and 384 prepared Phase B prompts. SHA-256 Checksum Record — cryptographic hashes for verifying the frozen hypothesis bank and workbook. Audit Log — recorded experimental operations and data-entry history. Quality-Control Report — automated consistency-check results. The automated quality-control check reported zero errors and zero warnings. 5. Research transparency and interpretation This release documents the transition from preregistered experimental design to frozen experimental inputs. The exact textual duplicates in the hypothesis bank are retained and disclosed rather than replaced after Phase A. The Phase A validity judgments were recorded by the researcher. The automated consistency check verifies recordkeeping consistency but does not constitute independent human validation of the generated hypotheses. The study examines observable effects of experimentally assigned information-status labels. It does not claim to identify or measure hidden internal belief states of the model. The experimental prompts are administered in Japanese. The Japanese prompt wording is authoritative, while the protocol and analysis-plan descriptions are provided in English for international accessibility. Phase B data collection and the final statistical analysis will be reported separately after completion.