Pin the world to any past instant and re-run your strategy, prompt, or screen against only what was published then. PRIORI.SH serves your agent clean, point-in-time retrieval — full provenance on every fact, and no leaked future in what we return.
Agent and RAG evals leak: your retrieval pulls
whatever's indexed today, so a "June 2022" run quietly
sees 2024 filings and a passing score proves nothing.
Point your agent at PRIORI.SH
with an as_of and its context holds only
what was published by that date — clean retrieval,
every run.*
* We gate the retrieval layer, not model weights — a model can still recall facts from its own training. For strict evals, also pick a model whose training cutoff predates your test window; see the methodology note.
# MCP server claude --mcp-server priorish \ "What did the Fed signal in May 2023 about persistent inflation?"
Macro releases get revised; as-reported EPS gets
restated. Pin every feature to its as_of
vintage and your eval sees the figure a desk actually
had that day — not the value quietly revised in
months later.
client.search( query="payrolls", as_of="2024-03-08", # what the tape saw sources=["alfred"], )
No leaked future in what we serve — and every fact proves where it came from.
Each result is filtered to what was published at your
as_of, then stamped with a provenance chain:
a BLAKE3 content hash, the source filing, and a sealed
extraction lock. Open any fact and trace it back to the
original document.
curl -H "X-API-Key: $KEY" \ -d '{"query":"CPI","as_of":"2024-06-01"}' \ https://api.priori.sh/search
from priorish_sdk import PriorishClient client = PriorishClient(api_key="...") client.search( query="unemployment", as_of="2024-03-15", )
# claude_desktop_config.json { "mcpServers": { "priorish": { "url": "https://mcp.priori.sh" } } }