Auto Tuning

Every sample is classified into a zone (Idle, Cruise or Boost) from MAP, RPM and VSS.

Sensor glitches, deceleration and unstable samples (RPM, MAP or VSS moving faster than the stability windows allow) are excluded, so corrections only come from clean, settled data.

Attribution uses a weighted algorithm: each sample is split across the four VE cells bracketing its operating point with bilinear weights, so a sample between cells contributes to each in proportion to how close it sits.

Alternatively the cell zone gate credits only samples near a cell's centre (Link-style active zone area) and drops the rest.

The engine then runs three passes, one per zone, and where passes overlap on a cell, Boost wins over Cruise and Cruise over Idle.

Per cell, the error against target is averaged over every hit and the VE value is scaled toward target, damped by the preset's smoothing:

where $\bar{e}$ is the cell's average error, $k_{smooth}$ the preset smoothing factor and $k_{var}$ the variance multiplier. Noisy cells are protected by the spread of their error:

• Above the reduce threshold the correction runs at half strength; above the skip threshold the cell is not touched at all
• Cells already within the well-tuned threshold of target are left alone
• Cells without enough hits for their zone are not corrected
• Results always clamp to the valid VE range (5.0 to 100.0)
• Presets (Conservative, Moderate, Aggressive, Raw) scale the smoothing, hit requirements and variance thresholds from heavily damped to full-strength single-shot, with Custom exposing every parameter

Targets come from the AFR target table in the loaded BIN when one is tuned (blended over MAP and RPM the same way as cell attribution), otherwise from the per-zone targets in Settings.

The target used for each sample is written into the log, so playback shows exactly what the correction aimed at.

The learning database

• Accumulates weighted error history per cell across sessions, per profile; every processed log and live session feeds it
• After a log's own corrections are applied, the database steps in on the cells that log could not cover
• A database correction only fires once a cell holds enough weighted evidence in a zone, and it plays by the same rules as Auto Tuning: same zone priority, same smoothing, same variance protection, same VE clamps
• Cells already marked well tuned are skipped, and cells that keep getting worse across three or more logs are flagged and left alone, since that pattern points at hardware rather than the tune
• Once a database correction is applied, that cell's stored history is zeroed so nothing ever corrects twice

The numbers

Zones

Stability windows

Samples moving faster than these are excluded as unstable.

Presets

Custom exposes every parameter.

Default targets

Used when the loaded BIN has no tuned AFR target table. All configurable per profile.

Learning database confidence

Graded on accumulated weighted hits per cell.