Public Access

Files

Codex d4b73b236a PINK DITAv2 Sprint 2-3: accounting parity + multi-leg groundwork

Sprint 2 (accounting + observability parity, PINK scope):
- Verified pink_clickhouse.py writes the 8 BLUE-legacy row families at
  matching schema and that capital authority in pink_direct.step() is
  solely kernel.account (no balance-poll overwrite in the hot loop).
- Report: prod/clean_arch/dita_v2/SPRINT2_ACCOUNTING_PARITY.md.

Sprint 3 offline groundwork (no exchange contact):
- Add _write_trade_exit_leg to pink_clickhouse.py: one BLUE-schema-faithful
  trade_exit_legs row per exit leg, with isolated (non-cumulative) per-leg
  deltas tracked via _leg_state (reset on ENTER). Closes the docstring gap.
- New offline suite test_pink_multi_exit_groundwork.py (3 passed):
  * Flaw 4 — two-leg exit closes once, realized accrues per leg, closed
    slot rejects further EXIT (no double-close).
  * Overshoot invariant — a final EXIT requesting more than the remaining
    size CLAMPS (size to 0, no oversell), retiring the Sprint 0 cumulative-
    ratio risk empirically.
  * trade_exit_legs delta + full BLUE column-set assertions.
- Persistence regression after edits: 10 passed.

BLUE untouched: no changes to dolphin.* / DOLPHIN_*_BLUE / nautilus_event_trader.py.
Live VST multi-leg run remains deferred pending explicit authorization.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

2026-05-30 19:21:45 +02:00

4.8 KiB

Raw Blame History

Sprint 2 — Accounting + observability parity verification

Date: 2026-05-30 Scope: Verify (no behaviour change) that the DITAv2 PINK runtime preserves BLUE-legacy-compatible ClickHouse row shapes in dolphin_pink, and that capital authority in the hot loop is solely the kernel's AccountProjection. Offline only (MockVenue / unit), no exchange contact. Continues [SPRINT0_FLAW_VERIFICATION.md].

1. Row-shape parity — `clean_arch/persistence/pink_clickhouse.py`

BLUE-legacy row families written, same schema / no new columns:

Row family	Writer	Status
`policy_events` + `v7_decision_events`	`_write_policy_event`	✅
`account_events`	`_write_account_event`	✅
`position_state`	`_write_position_state`	✅
`status_snapshots`	`_write_status_snapshot`	✅
`trade_events`	`_write_trade_event`	✅ (terminal close)
`trade_reconstruction`	`_write_trade_reconstruction`	✅ (ENTRY/PARTIAL/EXIT)
`anomaly_events`	`_write_anomaly` / `record_anomaly`	✅
`trade_exit_legs`	—	⚠️ listed in docstring, no writer

trade_exit_legs has no emitter. It is a multi-leg row family → relevant to Sprint 3 (DOLPHIN_PINK_PHASE=multi_exit), not single-leg MARKET. Not a Sprint 1/2 blocker. Action: add the writer when Sprint 3 is taken up, or confirm BLUE TUI/observability does not require it for single-leg trades.

2. Capital authority — single source = kernel `AccountProjection`

clean_arch/runtime/pink_direct.py hot loop (step, L309-408):

Capital is read only from kernel.snapshot()["account"] (L320, L370, L395).
Capital is mutated only by kernel.process_intent() → account.settle() on fill.
No balance-poll overwrite anywhere in step(). ✅

External capital writes (all outside the hot loop, by design):

_reconcile_position_slot (L188-194) — the single place an exchange balance snapshot seeds account.snapshot.capital; called at startup/recovery only.
connect() (L230) seeds from the env default initial_capital, not an exchange poll (per code comment L228-229).
recover_account() (L431) re-seeds from kernel.account.snapshot.capital (the kernel's own value) — not an exchange poll.

Doc/code note (no change made): reconcile_account() (L453) docstring says it "re-seeds capital from the exchange balance as a guard against drift," but the code path (recover_account) actually re-seeds from the kernel's own capital — i.e. it does not overwrite from an exchange poll. Behaviour is the safe one; only the comment overstates. Flagged for accuracy; not edited (no behaviour change w/o auth).

pink_clickhouse.py reads capital/peak/seq solely from account.snapshot (_capital/_peak_capital/_trade_seq, L193-201) — no duplicate tracking. ✅

3. Offline test results

siloqy_env, PYTHONPATH=/mnt/dolphinng5_predict, run from repo root.

Suite	Result
`test_pink_clickhouse_persistence.py`	✅ pass
`test_pink_ditav2_accounting_invariants.py`	✅ pass
`test_pink_direct_runtime.py`	✅ pass
DITAv2 PINK Sprint-2 scope	14 passed
`test_bingx_capital_accounting_battery.py`	❌ 2 failed — legacy path, out of scope

The 2 failures are in the legacy Nautilus BingX execution/journal path (prod/bingx/execution.py + prod/bingx/journal.py, imported via launch_dolphin_live) — not a DITAv2 PINK file, untracked/pre-existing, not modified by this engagement. Root cause: the fuzz/equivalence tests reuse fingerprint="fp" across iterations, so bingx_journal.write_snapshot fingerprint- dedup short-circuits the sink and captured["row"] is never set (KeyError). This lives on the legacy side of the BLUE do-not-touch boundary → not fixed here.

GATE decision

PASS (DITAv2 PINK scope). Row-shape parity holds for single-leg MARKET; capital authority is single (kernel AccountProjection) with no hot-loop balance overwrite; all PINK-scoped offline suites green.

Carry-forward (Sprint 3)

✅ CLOSED (offline groundwork, 2026-05-30): trade_exit_legs writer added to pink_clickhouse.py (_write_trade_exit_leg, BLUE-schema-faithful, isolated per-leg deltas tracked via self._leg_state, reset on ENTER). Fires once per exit leg.
✅ CLOSED (offline groundwork): cumulative-ratio exit sizing overshoot validated — test_pink_multi_exit_groundwork.py::test_final_leg_overshoot_does_not_oversell proves a final EXIT requesting more than the remaining size clamps (size→0, no oversell, closes once). Validation suite: 3 passed; persistence regression: 10 passed.
⏳ PENDING (live): the on-exchange multi-leg run (successive MARKET exits on VST to confirm Flaw 4 end-to-end) is deferred — requires explicit authorization for additional live testnet orders beyond the single Sprint 1 round trip.

4.8 KiB Raw Blame History