Files

T

jackyu66gitandCursor e2fbe1c2b3 Snapshot nautilus_mm after EXP_002 Phase 1 stop.

Keep frozen research conclusions and code; raw ledgers and secrets stay out of git.

Co-authored-by: Cursor <cursoragent@cursor.com>

2026-09-10 16:53:22 +08:00

1.7 KiB

Raw Permalink Blame History

Testnet Limitations — MM_EDGE_EXP_001

Observed on Binance USDT-M Futures Testnet during Data Collection (Aug 2026).

userTrades history cutoff

Endpoint:   GET /fapi/v1/userTrades
Observed max trade time (UTC): 2026-08-17T03:08:23
Behavior:   Pagination returns 3890 rows; no further trades via time/fromId
            after cutoff, even while probe continues to produce fills until
            2026-08-18.

Impact: Local jsonl fill count can exceed paginated userTrades count. This is not evidence of duplicate local logging or fake fills.

Order API remains available

Endpoint:   GET /fapi/v1/order?orderId=
Behavior:   Post-cutoff orders return status=FILLED, executedQty, avgPrice
            while userTrades?orderId= returns 0 rows for the same orderId.

RECON-03 classifies these as:

VENUE_CONFIRMED_NO_TRADE_HISTORY

Evidence grade: Order only (not dual Order+Trade).

Income ledger continues

GET /fapi/v1/income continues to record COMMISSION / REALIZED_PNL after the userTrades cutoff. Account reconciliation (RECON-01) uses income, not userTrades alone.

Implications for future runs

Real-time immutable ledger — persist on every OrderFilled: venue_trade_id, venue_order_id, liquidity_side, commission, exchange_ts, local_ts. Do not rely on post-hoc userTrades backfill.
Reports must use evidence taxonomy — never compare raw fill count to userTrades count without cutoff annotation.
Strict trade-level closure may remain FAIL on Testnet while order-level closure can still PASS.

Maker-only constraint

Post-only orders rejected with -5022 when they would take. Verified: TAKER_FILLED_COUNT = 0 on all 3890 trades inside userTrades window.

1.7 KiB Raw Permalink Blame History