CONFIDENTIAL DATA AND CODE FOR PEER REVIEW
British Journal of Guidance & Counselling
Teaching Higher Vocational Students to Evaluate AI Career Advice: TRACE Scaffolding and Appropriate Reliance in a Cluster-Randomised Trial

REPORTING PACKAGE — 18 SEPTEMBER 2026
Use with the accompanying manuscript, responses and Online Resource 1. Participant-level records are supplied for confidential review; public release remains subject to consent, ethics approval and institutional policy.

CONTENTS
01_core_data — participant/class identifiers, task totals and T2B component scores.
02_scoring — rating, adjudication, scoring-key and reconciliation files.
03_primary_analysis — primary inputs, model specifications, code and outputs.
04_reviewer_requested — factorial and missing-outcome analyses.
05_ai_and_implementation — implementation records, configuration and prompts.
08_version_reconciliation — historical input comparisons and numerical corrections.
09_r_fresh_run — retained R model objects, outputs and execution records.
13_located_response_materials — response-level records and associated files.
14_BJGC_revision — the current Python reproduction script and results.
15_T2B_review_materials — additional assessment and scoring documents.

REPRODUCTION
From the extracted package root:
  python 14_BJGC_revision/code/reproduce_bjgc.py .
Dependencies are listed in requirements_current.txt. This command reproduces exact tests, alternative scoring, numerical judgement/confidence summaries, missing-outcome analyses and direct transfer-change comparisons. It does not run R. R execution instructions are in 09_r_fresh_run/README.md.

ANALYSIS AND MATERIAL NOTES
Online Resource 1, S1.1 documents the corrected H2 ANCOVA and historical-input comparison; S5.1 describes the scoring inputs and additional task documents; S8 records the retrospective registration and development chronology. S9 explains the numerical/text-field distinction. Original-writing-length claims are excluded. Earlier character-count diagnostics remain in 08_version_reconciliation/length_diagnostics_not_used and are not part of the reported analysis. Additional task documents have not been used to rescore stored outcomes.

FILE INTEGRITY
This editorial update leaves data values, scripts, numerical outputs and source documents unchanged. SHA256_MANIFEST.txt covers every package member except itself. The manifest records this distributed version, not an independent pretrial timestamp.
