Where Legal Stress Resides: Uneven but Un-trending Supreme Court Engagement with India’s Insolvency and Bankruptcy Code, 2017-2026_ Supplementary Material
Description
Full Section-Level Dataset (269 Section Identifiers, 2017–March 2026) This annexure reports the complete raw and case-normalized engagement dataset underlying Figures 2 and 3, extending coverage beyond the top 30 sections discussed in the main text to all 269 distinct section identifiers extracted from the corpus. Columns report the IBC section, functional cluster assignment, total raw engagement count, number of contributing cases, and case-normalized engagement rate for each year in the 2019–2025 stable window. Sections with fewer than five years of non-zero engagement — excluded from the Objective O3 trend-testing battery in Section 4.3 — are marked accordingly in the final column. Full Spearman Trend-Test Results (58 Sections) This annexure reports the complete Benjamini-Hochberg-corrected results for all 58 sections tested in Objective O3, extending Table 7's illustrative subset to the full battery, including Spearman ρ, uncorrected p-value, and FDR-corrected p-value for every tested section.
Files
Steps to reproduce
# IBC Supreme Court Heatmap Paper — Single-Notebook Pipeline (Objectives O1 + O2) **Version 1.0 | Author: Manish Kumar Meena | Corpus: 225 Supreme Court IBC judgments (Manupatra .txt exports)** ## What this notebook produces | Output | Description | |---|---| | `paragraphs_raw.csv` | One row per judgment-body paragraph, with year, zone diagnostics | | `section_mentions_long.csv` | One row per (paragraph x IBC section) mention, with functional cluster | | `issues_final.csv` | Issue-framing extraction: one row per distinct legal issue per case | | `paragraphs_with_struct.csv` | Paragraphs + InLegalBERT structural role + confidence + fallback source | | `cases_summary.csv` | One row per case: year, sections, clusters, issues | | `O1_section_year_*.png / .csv` | **Objective 1** — Section x Year heatmap (paragraph counts + case counts) | | `O2_cluster_year_*.png / .csv` | **Objective 2** — Functional Cluster x Year heatmap with 2019/2021 amendment markers | | `IBC_heatmap_workbook.xlsx` | All pivots + validation cross-checks in one Excel workbook | ## Locked design decisions (from prior experiments) 1. **Raw individual paragraphs** — no clustering. One row = one court paragraph. 2. **Three-zone parsing** — HEADER / SUMMARY / JUDGMENT, with JUDGMENT/ORDER markers as primary boundary detector. 3. **Year extraction** — 5-signal cascade: override CSV → MANU citation → "Decided On" → "Date of Judgment" → header-scan fallback. 4. **Substantive filter** — InLegalBERT NB2_v22 structural head (`HOLDING_OUTCOME` + `INTERPRETATION_REASONING`), threshold 0.45, TF-IDF-calibrated keyword fallback for low-confidence paragraphs. 5. **IBC-only sections** — numeric range 1–255 plus statute-context exclusion (Companies Act, NI Act, CrPC, Limitation Act etc. are dropped). 6. **Issue framing** — explicit-marker regex (high confidence) + enumerated sub-issue splitting (moderate) + model-inferred candidates (exploratory, flagged). 7. **Heatmap counts = doctrinal engagements**, not raw word frequency: a section is counted only when it appears inside a paragraph classified as holding or interpretation. ## Before you run 1. Runtime → Change runtime type → **GPU** (T4 is enough). 2. Upload / confirm on Drive: the 225-case zip, `Training_Gold_Final.csv`, the `inlegalbert_ibc_model_v12` folder (weights + tokenizer), and `label_maps_v12.json` (optional but recommended). 3. Edit the paths in **Cell 1** only. Nothing else needs editing. 4. Run cells strictly in order. Diagnostic cells (marked D) tell you whether it is safe to proceed.
Institutions
- Indian Institute of Management LucknowUttar Pradesh, Lucknow