Project Black Pearl
AI-Driven ECU Engineering — Proven on a 2005 WRX
Project Black Pearl is the structured development of a data-driven, AI-assisted ECU tuning framework — validated in real time on a rebuilt 2005 Subaru WRX.
This is not a typical “stage build.”
This is constraint-based calibration.
Every pull is logged.
Every revision is documented.
Every adjustment is small, deliberate, and reversible.
Target outcome:
A stable, repeatable ~305whp rally-capable WRX — built without sacrificing engine longevity.
Engineering First
Black Pearl operates on a strict iteration protocol:
Log → Analyze → Score → Adjust → Flash → Validate → Repeat
Each WOT pull is graded for:
- Knock behavior (IAM, FBKC, FLKC)
- Boost tracking vs target
- AFR accuracy
- Injector duty margin
- Thermal stability
Changes are limited by hard rules:
- Timing increases ≤ 0.5° per iteration
- WGDC adjustments ≤ 2% per iteration
- No boost increases if fueling or heat is unstable
- No power escalation without repeatable clean pulls
Power is the final variable — not the first.
Reliability is non-negotiable.
The Proving Ground
The platform:
2005 Subaru WRX (GD chassis)
Rebuilt EJ205
VF-series turbo system
Pump gas calibration
Performance bias:
- Strong midrange torque
- Smooth, controlled boost ramp
- Predictable throttle modulation
- Consistent output under heat
This is a rally-oriented calibration — usable power, not dyno theatrics.
The goal isn’t a spike.
It’s repeatable performance.
Why This Matters
Most builds show parts lists.
Few show process.
By completion, Project Black Pearl will deliver:
- A validated ~305whp calibration
- A repeatable AI-assisted tuning framework
- Structured log scoring templates
- A safe-iteration methodology
- Documented before/after validation data
The car proves the system.
The system becomes the product.
Built With Purpose
This project exists for builders who want:
- Power without gambling their engine
- Data-backed decision making
- Measured, defensible tuning methodology
- Engineering discipline over ego
If it cannot be measured, it does not get changed.
Explore the Framework
The System Behind the Power
Project Black Pearl runs on a structured, AI-assisted ECU calibration architecture designed to remove guesswork and eliminate ego tuning.
This is controlled iteration — not aggressive experimentation.
Layer 1 — Data Acquisition
No decisions are made without structured logs.
Each pull captures:
- RPM
- Load
- Boost (target vs actual)
- WGDC
- IAM
- FBKC / FLKC
- AFR (wideband)
- MAF voltage
- Injector duty cycle
- IAT & coolant temp
- Throttle angle
If it cannot be logged, it cannot be trusted.
Layer 2 — AI Log Intelligence
Each WOT pull is scored across four categories:
Stability Score
Knock Risk Score
Boost Control Score
Fueling Accuracy Score
Output classification:
SAFE
CAUTION
STOP
The AI identifies patterns such as:
- Knock clustering by RPM/load
- Heat soak sensitivity
- Boost overshoot or oscillation
- AFR deviation trends
- Injector duty risk zones
- MAF scaling inconsistencies
No emotional tuning. Only pattern detection.
Layer 3 — Controlled Change Generator
Adjustments are constrained by strict limits:
- Timing delta ≤ +0.5° per iteration
- WGDC delta ≤ 2% per iteration
- Fueling changes incremental only
No timing increases if:
- IAM unstable
- Active knock present
- Boost overshooting
No boost increases if:
- Injector duty exceeds margin
- AFR trends lean
- IAT elevated
Power is earned through stability.
Layer 4 — Validation Protocol
Before escalation:
- Two repeatable clean pulls
- Stable IAM
- No sustained knock events
- Boost tracking within tolerance
- AFR within target margin
If validation fails → revert.
No exceptions.
The Outcome
This framework produces:
- Predictable power delivery
- Heat-consistent performance
- Reduced engine risk
- Documented revision history
- A teachable, repeatable methodology
The car is the proof.
The framework is the asset.