KAIS Online Resource 2 - Reproducibility Materials
==================================================

Article: Forecasting Fine-Tuning Break-Even Label Budgets in Text Classification from Frozen Embedding Geometry
Author: Emin Talip Demirkiran
Journal: Knowledge and Information Systems

Contents
--------
protocol/
  prespecified_protocol_publication_copy.md
  prespecification_record.json

code/
  reproduce_task_level_analysis.py

results/tables/
  consolidated_task_level_results.csv
  primary_regression_results.csv

diagnostics/
  KAIS_Supplementary_Table_S3_HC3_Diagnostics.csv
  KAIS_Supplementary_Table_S3_HighK_Sensitivity.csv

task_metadata/
  KAIS_Supplementary_Table_S1_Task_Corpus.csv
  KAIS_Supplementary_Table_S2_Configuration.csv

environment/
  software_environment.json

The publication-facing analysis script reproduces the task-level primary OLS regressions and selected post hoc diagnostics from the included derived data table. It does not require raw third-party benchmark datasets or pretrained model weights.

The prespecified protocol copy removes project-internal file paths and internal labels while preserving the substantive pre-outcome analysis decisions. The accompanying prespecification record retains the original cryptographic hashes and lock timestamp for provenance.

Raw third-party datasets and pretrained model weights are intentionally not redistributed. Dataset repositories and source attributions are provided in Online Resource 1 / Table S1 and in the task-metadata CSV included here.
