Sample EOD risk extract
A real extract, taken from a live run; not hand-written. It is here so the delivered contract can be read, loaded against, and verified without standing a cluster up.
| File | What |
|---|---|
risk-extract.csv | The fixture as delivered: # metadata preamble, then one un-netted row per (accountId, security) |
risk-extract.cut | The position cut it was built from; the leader's render at consensus sequence 1544719 |
Taken at sessionDate=2026-07-22, priceSnapshotVersion=8, consensusSequence=1544719, 18 rows.
Delivered as seq-1544719.csv alongside seq-1544719.cut.
It verifies itselfβ
The fixture's preamble records the SHA-256 of the cut it was built from, so the pair can be checked with nothing but a hash:
grep '^# cutSha256=' risk-extract.csv # 9ceeed3652486b71c919ed532e06f01a81a468dd2d23dc122bfcf30b56e3274d
sha256sum risk-extract.cut # ...the same value
And the fixture is a pure function of that cut, so it rebuilds byte-identically from it with no cluster involved (FR-RXT10):
java -cp '/opt/app/classes:/opt/app/lib/*' \
finos.traderx.ordermatcher.cluster.RiskExtractMain --rebuild risk-extract.cut /tmp/rebuild.csv
cmp risk-extract.csv /tmp/rebuild.csv
Loading itβ
The preamble is #-prefixed, so every mainstream CSV reader skips it with one argument:
import pandas as pd
df = pd.read_csv("risk-extract.csv", comment="#")
Read the preamble separately for the stamp and the conventions; it states the market-value and
cost-basis formulas, the netting position, and what each markSource means, so a tie-out
discrepancy starts from a written convention rather than a guess.
What to noticeβ
- Every row is
markSource=EOD_SNAPSHOT,markQuality=OK; equities and options alike are marked from the same published closing-price version. - Rows are un-netted. Accounts 22214 and 42422 hold the two sides of the same crosses and both
appear in full;
counterpartyIdandnettingSetIdare attributes, never an applied aggregation. marketValueis multiplier-aware.22214,AAPL260918C00240000is 10 contracts Γ $10.025 Γ 100 = $10,025; not $100.25. It equalseod_position_pnl.market_valuefor the same row exactly.- Flat positions are emitted, not dropped (
AAPL261218C00240000, quantity 0), so a consumer sees the whole(account, security)universe rather than only what happens to be non-zero. - Prices are exact decimals at scale 6 throughout, carried from integer ticks; no float rounding can differ between runs or architectures.
- Every row is state at one consensus sequence, so this is a portfolio the firm genuinely held simultaneously, which is what makes a VaR computed from it meaningful.