Reading the whole class record in one context.

An employment class action lives or dies on the records — and the records arrive as a thousand-spreadsheet mess of payroll exports, punch logs, schedules, and personnel files that rarely agree. JustineAI™ is built to ingest the full production and reconcile it inside one context: the reasoning core will read every pay period for every member at once, align timekeeping against payroll against personnel, and surface where they diverge. The reconciled ledger — not the raw export pile — becomes the substrate every later mechanism reasons over.

What EL is designed to do.

1. 01

   Ingestion will normalize heterogeneous payroll exports into one canonical per-member, per-pay-period earnings-and-hours record with the source file cited on every field.

2. 02

   The 10M-token context lets the reasoning core hold the full class ledger at once, so reconciliation is a single pass over all members, not a per-custodian batch job that loses cross-member pattern.

3. 03

   Timekeeping-versus-payroll divergence is the engine of the case: the platform will flag punches that exceed paid hours, auto-deducted meal breaks never taken, and shifts that cross the overtime threshold unpaid.

4. 04

   Every reconciled value will carry provenance back to the exact cell, page, or punch it derives from — so a number that reaches a certification brief or damages model can be traced to its source on one click.

5. 05

   Reconciliation is attorney-supervised: cross-source conflicts route to a confirmation surface where counsel resolves the discrepancy, and the resolution is audit-logged as work product, not silently overwritten.

The AI reasons; the attorney decides.