req_pnl() callback returns cumulative notional values instead of daily P&L

[mdgi]

Bug Description

req_pnl() streaming callback returns values that appear to be cumulative notional trade values (qty × price) rather than actual daily P&L.

Environment

• IBX version: 0.4.4
• IBKR paper account
• Python 3.12, Linux

Observed Behavior

The daily_pnl field in the pnl() callback jumps by the notional value of each trade rather than the actual P&L:

│ 10:06 │ BUY 171 ASTS @ $87.38 (notional=$14,941)
│ 10:20 │ BUY 72 BE @ $205.93 (notional=$14,827)
│ 11:43 │ BUY 73 BE @ $203.40 (notional=$14,848)

Pattern: each trade adds ~$14,800-$14,940 (the notional size) to the reported "daily P&L". This suggests the field contains cumulative trade value rather than profit/loss.

Expected Behavior

daily_pnl should reflect the actual daily profit & loss (realized + unrealized), matching IBKR's reqPnL() API spec.

Possible Cause

FIX tag mapping issue — market value or trade notional may be read as P&L field.

[deepentropy]

Fixed in v0.4.5 (dc554db).

Root cause: Tag 6822 in the 6040=143 response is moneyTradedSinceMidnight (a midnight seed), not unrealized P&L. The previous code summed these values as P&L, so each trade added its notional to the reported daily P&L.

Fix: The engine now stores per-position midnight seeds and computes daily P&L client-side:

dailyPnL = Σ(qtyNow × priceNow - qtyMidnight × prevClose - moneyTraded)

This matches how the Java gateway computes it internally.

Note: Positions need active market data subscriptions for last and close prices. Positions without subscriptions are excluded from the computation.

@mdgi — could you verify the fix against your paper account? pip install --upgrade ibx should pull v0.4.5. The daily P&L values should now track actual profit/loss rather than trade notional.

[mdgi]

Hi .

Got v0.4.5 installed and verified the ABI surface against our setup — no callback signature changes needed on our side.

Quick note for anyone else landing here via pip: our install is editable Rust source, so we used git checkout v0.4.5 && maturin develop --release rather than pip install --upgrade ibx. Release build completed cleanly in ~20s.

I'll post before/after numbers once today's paper session produces its first bracket fill — comparing the streamed daily_pnl against the sum of realized_pnl from our per-fill CSV.

Will follow up with the post-fix numbers by end-of-day ET today.

[deepentropy]

Status update + architecture note

The v0.4.5 fix in 111ab5b resolved the notional-leak symptom reported here — for accounts that hold positions overnight, daily_pnl no longer accumulates trade notional. Verified by code review.

A live probe found a follow-on regression: for intraday-opened positions (not held at midnight) the pnl() callback no longer fires at all. Tracked at #166.

A protocol capture from deepentropy/ib-agent#146 then settled the architectural question: the upstream client treats reqPnL as a server-stream passthrough on the API socket, not a client-side reconstruction. The gateway already computes daily P&L and pushes the result event-driven whenever quotes tick or fills land. ibx's current CCP/FIX-side subscription + midnight-seed reconstruction is on the wrong side of the gateway boundary entirely.

Disposition

• This issue (req_pnl() callback returns cumulative notional values instead of daily P&L #142): original symptom is fixed in v0.4.5+. The architectural rework is tracked at req_pnl: pnl() callback never fires for intraday-opened positions #166 and will obsolete the current implementation; this issue can stay closed (or be reopened only if the v0.4.5 fix's specific symptom recurs).
• req_pnl: pnl() callback never fires for intraday-opened positions #166: tracks the passthrough rework that structurally prevents both the original symptom and the silent-stale regression.

No action requested on this issue — leaving the comment for context only.

[deepentropy]

Fixed across two follow-up merges on main:

• v0.4.5 — original notional-leak fix (daily_pnl no longer accumulates trade notionals per fill).
• Fix req_pnl() not firing for intraday-opened positions (#166) #169 — fixed the regression where pnl() stopped firing for positions opened intraday after the v0.4.5 fix. poll_pnl now iterates the union of midnight seeds and current positions; positions without a seed synthesize money_traded = qty_now × avg_cost, collapsing the formula to unrealized P&L.
• Fix pnl_single() defects: hardcoded zeros and wrong quote routing (#168) #171 — sibling fix for pnl_single() (daily_pnl/realized_pnl were hardcoded to 0.0; quote lookup ignored con_id).

Closing as fixed. Live RTH re-probe with scripts/.tmp/probe_pnl_142.py is the remaining verification step and will be tracked separately if it surfaces anything.