feat(ledger): book ccxt venue fees against their native asset (draft, #4277)

[TaprootFreak]

Draft for evaluation, not a merge candidate. Addresses #4277. The honest recommendation is hold / won't-fix — the benefit is immaterial and the semantic blast radius is high. Opening it so the change and its trade-offs are concrete.

What #4277 is, and the deeper finding

The ccxt Trade branch (exchange-tx.consumer.ts tradeSpec) books the venue fee as a CHF-only EXPENSE/spread-{exchange} leg valued at the persisted, 2-dp-rounded feeAmountChf. A Binance micro-fill's BNB commission is worth < 0.005 CHF → feeAmountChf = 0 → no leg (the #4277 symptom, and the benign cause of the pre-#4268 checkNativeBalance noise).

Investigating the fix surfaced a bigger point: the quote-spread plug (residualChf = -Σ amountChf) makes total spread-{exchange} = quoteChf − baseChf independent of feeAmountChf, so the venue fee is algebraically neutralized on every ccxt fill (not just sub-cent ones) — it is neither recognized as a cost nor tracked as a native outflow. The fee-currency (BNB) journal balance therefore carries a growing phantom that §7 reconciliation only observes and alarms (§7 never writes), so nothing is silently corrupted today.

The change

Mirror payout-order.consumer.appendDistinctFeeLegs: book the fee against its native ASSET/{exchange}/{feeCurrency} account (mark-valued), so the native outflow survives even when its CHF rounds to 0. New private appendCcxtFeeLeg replaces the two-line CHF push. Conservative by design: the native leg is booked only for a fee in a distinct asset (≠ base/quote — those are already in cost, native booking there would double-count) that has a bootstrapped account and a historical mark; otherwise it falls back to today's exact CHF spread leg, so no trade that books today starts deferring. Sign-aware off the signed feeAmount (maker rebate → Dr / INCOME). Scrypt's market-spread branch is untouched. No new account names, no CoA/bootstrap change, no migration; forward-only.

Blast radius (why hold)

• Because the plug neutralizes the fee for material fills too, this doesn't merely recover sub-cent legs — it recognizes the venue fee as real spread-{exchange} expense on all ccxt trades (Binance/MEXC/Kraken) and starts native fee-asset tracking. Cumulative P&L/native impact is the total ccxt fees over the ledger period (still small vs total balance, but larger than the sub-cent-only ~tens-of-CHF/20-months in Ledger: sub-cent fee-less exchange micro-fills omit the EXPENSE fee leg (feeAmountChf rounds to 0) #4277).
• It shifts the §7 Binance/BNB diff and equity-parity baselines toward correct (§7 is a pure observer → no double-correction), and the content-change scan will reverse+rebook the recently-synced trade window on first deploy (append-only, bounded burst).
• Requires Binance/BNB (and every ccxt fee currency) to be a bootstrapped ASSET account for the fix to bite; otherwise it safely no-ops to the CHF fallback.

Bottom line: correct and buildable, but the cost/benefit doesn't justify a merge on its own. Nothing is lost today that §7 doesn't already surface.

Does the pattern generalize? (the survey)

A repo-wide audit of the ledger for the same anti-pattern (a crypto-native amount booked via a pre-rounded *AmountChf field, losing sub-cent precision) found three confirmed sites; fiat/CHF-denominated 2-dp values are correct and were not flagged:

1. exchange-tx.consumer.ts (this PR) — ccxt venue fee. Immaterial.
2. trading-order.consumer.ts (book, network-fee leg) — DfxDex arbitrage on-chain gas (txFeeAmount, native) booked only as CHF txFeeAmountChf; no native gas-coin leg. Materiality plausibly non-trivial but unquantified (L1 gas). Higher fix risk (TradingOrder has no fee-currency column → the gas token must be derived from the swap chain). → own issue, not this PR.
3. payout-order.consumer.ts (appendDistinctFeeLegs early-return gate if (feeChf === 0) return) — short-circuits before the correct native-leg machinery, so a distinct non-payout fee asset whose CHF sum rounds to 0 loses its native leg. Immaterial, narrowest edge, lowest fix risk (un-gate the existing native legs, gate only the CHF EXPENSE). → small follow-up issue.

So: the approach is not exchange-tx-specific — the same "book fees natively, not via a pre-rounded CHF scalar" principle applies to trading-order and (narrowly) payout-order. Each is an independent mechanism on a different entity with different materiality and risk, so they belong in separate issues rather than one bundled PR.

Verification (Node 20)

prettier / eslint / type-check / build clean · DI graph resolves · full accounting suite 780/780 (incl. 3 new: native distinct-asset fee; sub-cent fee preserved natively; CHF fallback when the fee account is missing). The existing ccxt fee test (feeAmountChf only) now exercises the fallback path unchanged, as a regression guard.

Staying draft intentionally. If the team decides native fee tracking is a principle worth the churn, I'll take it through the full review-to-ready gate and file the two sibling issues; merge stays with the maintainer.

[TaprootFreak]

Folded into #4279, which packages all three native-fee sites (exchange-tx + trading-order + payout-order) from the #4277 audit into one PR. Closing this exchange-tx-only draft.