v3.9.0 - C.O.N.T.R.A. Phases A-G + Fresno Corpus + odia-v1
C.O.N.T.R.A. - Commercial Contract Asymmetry Framework
Ten new commercial-contract detectors (L-11 through L-20) analyzing consumer adhesion contracts for structural rights foreclosure:
Detector What It Catches
L-11 Arbitration Architecture Mandatory arbitration; AAA/JAMS/smaller provider stats; Wilson CI + HHI repeat-player concentration
L-12 Choice of Law / Forum Asymmetric forum selection relative to consumer's location
L-13 Unilateral Modification "We may update at any time" illusory-contract language
L-14 Data Collection Depth Breadth of collection categories; behavioral, biometric, inferred data
L-15 Data Retention Indefinite or absent retention limits
L-16 Onward Transfer Third-party sharing breadth; "partners / affiliates / service providers"
L-17 ML / AI Training User-content-as-training-data; opt-out absence
L-18 Remedy Foreclosure Class action waiver, injunctive relief waiver, jury trial waiver
L-19 Enforcement Asymmetry Fee-shifting, cost-bond, indemnification asymmetry
L-20 Dark Pattern UI/UX coercion; pre-checked boxes; roach-motel cancellation
CASI - Consumer Adhesion Severity Index
Five-axis 0–100 composite score per document: remedy_foreclosure · data_extraction_depth · modification_and_consent · procedural_adhesion · enforcement_cost_asymmetry
Bands: Baseline (0–19) / Elevated (20–39) / Substantial (40–59) / Severe (60–79) / Foreclosure Regime (80–100)
Entity Registry
Fuzzy-match resolution (rapidfuzz, threshold 0.88). 32 entities pre-seeded - SCE, PG&E, AT&T Mobility, JPMorgan Chase, BofA, Wells Fargo, T-Mobile, Comcast, and more. Auto-creates on first ingest of an unknown entity.
CCP § 1281.96 Pipeline
Consumer arbitration statistics retrieval from AAA, JAMS, ADRS, JUDICATE_WEST, FEDARB, NAM - normalized case records with prevailing-rate analysis and repeat-player HHI concentration.
Report Cards
T.C.A.M.S. (Targeted Contract Asymmetry Monitoring Summary) - corpus-level DOCX: CASI distribution, top-10 entities, axis dominance heatmap, doctrinal anchor frequency, L-detector firing counts
C.C.C.E.A. (Commercial Contract Clause Exposure Analysis) - Jaccard-clustered clause exposure (threshold 0.35) across 8 clause types with probable drafting firm attribution (Cooley, Wilson Sonsini, Fenwick & West, Latham & Watkins, Gibson Dunn, DLA Piper)
Analytical Card - per-document DOCX: entity profile, CASI axis breakdown, findings table, recommended actions
12-Step Commercial Ingest Pipeline
odia contra-ingest --source contract.pdf --entity "AT&T Mobility" --doc-type tos --output ./cards/
PDF/OCR extract → SHA-256 dedup → provenance → entity resolve → L-1–L-10 legal → L-11–L-20 C.O.N.T.R.A. → CASI score → Wayback archive → DB persist → Analytical Card DOCX
Wayback Machine Client
find_capture() / retrieve_prior_versions() - historical contract version tracking with 1 req/2 s rate limit.
Corpus Expansion
Addition Documents Findings
Fresno County (NSU complete) 32,340 73,547
Fresno PD 126 526
Total corpus (16 jurisdictions) 50,699 148,349
odia-v1 Fine-Tune
QLoRA fine-tuned Llama-3.1-8B-Instruct on 87,618 examples (13,498 MAS reports + 74,120 explanation pairs). Quantized to Q4_K_M GGUF (4.92 GB). Live as default RAG LLM via Ollama. Available at SynTechRev/odia-v1 (private) on HuggingFace.
RAG index rebuilt post-Fresno ingest: collection 282 MB · ace_collection 1,158 MB · jim_collection 2.8 MB all 50,699 docs indexed.
Tests
4,339 passing · 15 skipped · 0 failures
106 new tests for Phase F+G (T.C.A.M.S., C.C.C.E.A., Wayback client, 12-step ingest pipeline).