AI Test AutomationCapital markets AI testingFIX workbench
Capital markets AI testing Preview
FIX workbench
Three ways in. Paste a message and see what a counterparty's engine would reject, and why. Paste a session log and see the gaps, the resends and each order's story. Or stand in the middle of a live session, drop or edit a message in transit, and watch the two sides find the gap and recover.
Preview. The FIX workbench is new and still growing: it reads FIX 4.4 order, execution and session messages, the ones the trading lab sends. It runs entirely in your browser; nothing you paste is sent anywhere.
| Tag | Field | Value | Meaning |
|---|
What went wrong
Each order's story
Client CLIENT
- Next out
- 1
- Expects in
- 1
Broker KITESTRING
- Next out
- 1
- Expects in
- 1
Market orders fill at $42.10; limit orders rest. A cancel for a filled order gets a 35=9.
Ask an AI
The rules find what is wrong; an assistant can tell the story
The checks on this page are plain rules, so the same log always gets the same findings, and every finding names the field and the rule. They do not guess at causes. For that, Ask an AI on Read a session copies a prompt with the log and the findings, ready to paste into Claude or the assistant your team uses: it asks for the story, the most likely cause, what each side should do next, and which findings it disagrees with. Nothing leaves this page unless you paste it somewhere.
That split is the point: rules for what must be true, a model for what probably happened, and a person deciding. Before you paste a production log anywhere, take out account numbers, client IDs and anything your firm treats as confidential.
The rules
What the checker knows
| Area | What it checks |
|---|---|
| The frame | BeginString (8), BodyLength (9) and MsgType (35) first, in that order; CheckSum (10) last; BodyLength and CheckSum computed again from the bytes; no empty fields, no tag repeated outside a group. A bad checksum means the receiver drops the message unread, so the next one shows a gap. Mend the frame puts the header in order and computes both again. |
| The header | SenderCompID (49), TargetCompID (56), MsgSeqNum (34) and SendingTime (52) on every message; UTC times as YYYYMMDD-HH:MM:SS.sss; OrigSendingTime (122) on a possible duplicate. |
| Orders | Required fields for 35=D, F, G, 8 and 9; Price (44) for limit and stop-limit, StopPx (99) for stops; a market order that carries a price; ExpireDate (432) for Good Till Date; a cancel or change needs a ClOrdID of its own; every enumerated value against FIX 4.4. |
| Execution reports | CumQty plus LeavesQty equal OrderQty on a live order; nothing left open on a filled, cancelled, expired or rejected one; LastPx and LastQty on a trade; OrigClOrdID on a replace. |
| The session | Per direction: the expected MsgSeqNum, gaps and whether they were recovered, resends with PossDupFlag, gap fills that move forward only, a sequence number too low without PossDupFlag (a serious error: the receiver logs out). Each order followed through ClOrdID and OrigClOrdID: a fill after a cancel, a cancel never answered, a ClOrdID used twice. |
FIX: CheckSum and BodyLength (OnixS dictionary, appendix B) · FIX 4.4 Sequence Reset (35=4): gap fill and reset modes · FIX 4.4 Order Cancel Request (F) · FIX 4.4 Order Cancel/Replace Request (G) · FIX 4.4 Order Cancel Reject (9) · FIX 4.4 TimeInForce (59) · FIX 4.4 OrdType (40) · FIX 4.4 ExecType (150) · FIX 4.4 CxlRejReason (102)
The checker is data/testlab/trading/feed/fixlint.js, tested with the rest of the FIX code in feed_tests.js. The orders it reads are the ones the trading lab sends; its FIX log downloads open here as they are.