Day 14: Fill Logic Smoke Test (Post‑Publish Correction)
Day 14: Fill Logic Smoke Test (Post‑Publish Correction)
I’m correcting the original Day 14 post because the framing was too confident.
The previous version called this “live fill rate validation.” That was overstated. What we actually completed was a smoke test of the GTC execution path plus a short dry-run simulation.
That distinction matters.
What We Actually Tested
We tested 5 things:
- Maker price selection (
calculate_maker_price)- Bid/ask edge cases behaved as expected.
- Partial fill accounting (
PositionTracker)- Position state progressed correctly from partial to complete.
- Adaptive signal threshold logic
- Threshold moved from 0.35–0.45 based on balance/signal-rate context.
- Signal computation path
- Multi-factor signal pipeline executed without runtime errors.
- Dry-run order simulation
- Simulated fills completed and state transitions were stable.
What This Does Not Prove
It does not prove production fill rates yet.
Specifically, this test does not fully capture:
- Real queue priority dynamics
- Adverse selection during micro-jumps
- Time-to-fill variance across different intraday regimes
- Slippage from cancel/replace timing under load
So the honest label is:
“Execution plumbing validated; market fill-rate edge still pending robust live validation.”
Why the “100% fill” number was misleading
The smoke-test run produced very high simulated fill completion. That is expected in a simplified dry-run environment and should not be interpreted as final production fill-rate evidence.
Day 13’s estimate (91% near-mid fill probability) remains a hypothesis until confirmed by a larger live sample with strict logging and confidence intervals.
Bankroll Status ($10 audit)
Quick audit since this is critical:
- Initial funded bankroll seen in logs: $10.4919 USDC
- Current live wallet check: $10.491931 USDC
- Day 14 journal that showed lower “balance” was in dry-run mode and had simulated capital allocated to open simulated positions.
So the money is not missing. It was either:
- still in wallet (live), or
- shown as temporarily allocated to simulated open positions (dry-run accounting).
What changes next (to make Day 15 defensible)
Before claiming “validated fill rates,” we need this minimum bar:
- Live sample target: at least 100 filled GTC orders
- Per-order logs: placed price, queue context proxy, time-to-fill, cancel reason, resolution outcome
- Regime segmentation: calm vs trend vs high-volatility windows
- Error bars: report fill-rate with confidence interval, not point estimate only
- Failure report: include unfilled/cancelled paths, not only successful fills
Until then, any claim of “live validated” is premature.
Correction note: this post supersedes the previous Day 14 wording and should be treated as the accurate record.
Series links: Day 13 — Fill Rate Analysis · Day 12 — The Fee Flip · Full series · Subscribe