Corpus
The reference layer the historical collection is measured against: what the reviewed analysis found, the section taxonomy every retrieved unit is classified into, and the defect classes that decide what may be retrieved. The collection itself is loaded — the live counts are on the Precedents screen.
Known corpus state
Ingestion reconciles against these figures rather than silently producing a new reality. Where the pipeline disagrees with the completed analysis, both are kept and the disagreement is surfaced for review.
- Source files
- 918
- Distinct documents
- 693
- Readable distinct documents
- 677
- Exact duplicate groups
- 225
- Near-duplicate groups
- 18
- Matters
- ≈489
- Documents sharing a filename
- 60
- Unreadable files
- 16
Section taxonomy
36 canonical section types across 5 blocks, plus 86 clause tags preserved verbatim from the Pass 3 analysis. One primary type per section; tags are multi-value.
GENERAL PROCEDURAL
- Captionstructural
- Introductory response
- Objection
- Objection then answer
- Privacy objection
- Judicial notice (CPLR 4511)
- Unavailable-information deferral
- Supplementation reservation
- Expert reservation
- Property-damage reservation
- Plaintiff's verification
- Attorney affirmation
- Signature blockstructural
- Affidavit of servicestructural
AUTO LIABILITY
- Accident mechanism
- Negligence recital
- Statutory allegation
PREMISES LIABILITY
- Accident mechanism
- Negligence recital
- Statutory allegation
- Dangerous condition
- Actual notice
- Constructive notice
- Creation of the condition
- Maintenance / inspection failure
LABOR LAW
- Labor Law allegation
INJURY DAMAGES
- Injuries
- Surgery
- Injections
- Future care
- Permanency
- Serious injury (Ins. Law 5102)
- Prior condition
- Confinement
- Treating providers
- Employment and lost earnings
- Special damages
- Collateral sources
- Loss of services
Defect registry
59 rules: the 55 Pass 3 template defects plus 3 document-level blocking conditions. Severity is how much a defect costs the firm; effect is what retrieval does about it. The two are independent.
- BLOCK DOCUMENT
- 4
- SUPPRESS SECTION
- 15
- DEMOTE
- 20
- WARN
- 20
Coverage
58 categories: the 35 documented gaps plus the core categories where the corpus is measurably strong or merely thin. Coverage stays visible in precedent search and becomes a drafting gate later — it never suppresses a result.
- Supported
- 12
- Limited
- 8
- Attorney review required
- 17
- Unsupported
- 21
Labor Law is unsupported in every category: 9 documents reciting one paragraph that pleads safety devices were provided. Pass 3 recommends switching it off entirely rather than queueing it behind the auto and premises work.