50k-document e-discovery under budget
At corpus scale, compress or hand off — full-dump is off the table.
Open the live lab · pre-loaded to this scenario
Context & Memory Engineering
Context
Reviewing a 50,000-document e-discovery corpus under a fixed budget. Full-dump is impossible; the real choice is between semantic compression and handing slices to sub-agents that each carry only their brief.
The decision
At corpus scale the dial is compress vs sub-agent handoff — both stay in budget; the choice is whether cross-document reasoning (compress) or parallel throughput (handoff) matters more.
What most miss
People debate summarize vs full-dump; at 50k docs that's not the axis. The scale question is compression fidelity vs handoff coordination cost.
Stakes
The wrong strategy at corpus scale either blows the review budget or drops the document that mattered.
Studied · Agent & Protocol · verified 2026-07-03
Sources: E-discovery review at corpus scale (compression, sub-agent partitioning); Context budgeting for large-corpus tasks