Skip to content

tasks.md

Latest commit

 

History

History
254 lines (182 loc) · 8.99 KB

tasks.md

File metadata and controls

254 lines (182 loc) · 8.99 KB

Task Backlog

Two work-streams: the Lua plugin (ExportTune/) and the Python tools (tools/).
They share the same JSON contract (tune_export.json / tune_import.json).

Stream A — ExportTune Lua Plugin

A1 — Modularise main.lua into sub-files [prerequisite for everything else]

package.path includes the plugin folder, so require("modulename") works for
sibling .lua files. Splitting the 1 250-line monolith makes each feature independently
editable and testable.

Proposed file split:

File Contents extracted from main.lua
json.lua esc, num, json.decode (no FlashPro API deps)
export_cal.lua dump_parameter/index/1d/2d/3d, dump_table, write_calibration, write_dtcs, print_cal_summary, KEY_PARAMS
export_datalog.lua SENSORS, TS_SENSORS, safe_sensor, stats, ask_sample_count, dump_datalogs, write_datalog
import.lua build_cal_index, apply_change, run_import
health.lua do_health_check
debug.lua copy_to_clip, INTERESTING_GLOBALS, dump_global_val, dump_global, do_debug
main.lua get_cal_dir, get_downloads_dir, get_out_file, banner, done, action wrappers, main(), callbacks

Also update deploy.bat:

xcopy /Y /I "%SRC%\*.lua" "%DST%\" >nul

A2 — Add Table3D import support

apply_change currently skips Table3D kinds.
Boost and traction-control maps are 3D; without this, AI-generated changes to them
cannot be applied.

  • Add elseif kind == "Table3D" branch in apply_change (in import.lua after split)
  • Validate pages array: #pages == ax3:length(), each page rows × cols
  • Call t:setvalue(val, col, row, page)
  • Update tune_import.json format docs in README.md

A3 — Per-cell (partial) table import

Currently Table1D and Table2D imports require the full array.
AI often only wants to change a handful of cells (e.g. knock retard at a specific RPM/load).

  • Accept an optional "cells" array instead of "values" / "rows": 
    { "name": "Ignition Low", "kind": "Table2D",
  "cells": [ {"col":3,"row":5,"value":28.0}, ... ] }
  • Validate each index is within bounds before writing
  • Mix with full-table replace is not required — one format per change entry is enough

A4 — Export device & ECU metadata

Currently the JSON has no information about the connected FlashPro device or ECU state.
Useful for AI context and for diagnosing support requests.

  • Add a "device" block to the export header: 
    "device": {
  "name": "...", "connected": true, "hardware": "...",
  "serial": "...", "vinlocked": false,
  "obdii_voltage": 12.4, "ignition_on": true
}
  • Write in write_header() (or a new write_device())
  • Wrap every Device.* and ECU.* call in pcall (device may not be connected)

A5 — Export calibration table categories

All 300+ tables are currently exported as a flat list.
Python and AI tools have to guess groupings from table names.

  • Add a "category" field to each table entry using a name-prefix lookup: 
    local CATEGORIES = {
  ["Injector"] = "fuel", ["AFM"] = "fuel", ["Fuel"] = "fuel",
  ["Ignition"] = "ignition", ["Cam"] = "cam",
  ["Idle"] = "idle", ["Rev"] = "limits",
  ["WOT"] = "fuel", ["Lean"] = "fuel",
  ["Boost"] = "boost", ["TC"] = "boost",
  ["AIRC"] = "torque", ["Trq"] = "torque",
  ["Knock"] = "knock", ["K."] = "knock",
}
  • Adds one string field per table entry; keeps the export backwards-compatible

A6 — Export histogram data per sensor

Currently only min / max / mean are exported per sensor.
Histograms allow the Python tool to do binning without the full time-series.

  • After computing stats, do a second pass building a 20-bucket histogram
  • Add "hist": {"bins":[...], "counts":[...]} inside each sensor stats object
  • Bucket boundaries derived from sensor:min() / sensor:max()
  • Make bucket count configurable (default 20) or ask via InputQuery

A7 — Health check enhancements

do_health_check currently only prints to console.

  • Show result in a ShowMessage dialog (truncated to 60 lines, full in console)
  • Flag warning conditions inline:
    • |S.TRIM mean| > 5 % → warn "AFM may need correction"
    • K.Retard max > 3° → warn "Knock detected"
    • Wide min < 11.0 → warn "Dangerously lean WOT excursion"
    • ECT max > 105°C → warn "Cooling system concern"
  • Return a structured result table so future Python bridge can consume it

Stream B — Python Tools (tools/)

B1 — Knock analysis tab

The most safety-critical analysis missing from the current tool.

  • New tab: Knock
  • Load K.Retard, K.Level, K.Retard.1–.4 from time-series
  • Scatter: K.Retard vs RPM, coloured by cylinder (1–4 + overall)
  • Heatmap: bin K.Retard mean into RPM × Load (MAP) grid — overlay on IGN map shape
  • Highlight cells with mean retard > 1° in red, > 3° in orange
  • Export: generate tune_import.json with IGN table reduced by the mean retard at
    each RPM/load cell (conservative, opt-in with a checkbox)

B2 — Fuel trim health overview

Companion to the existing AFM analyzer; broader picture.

  • New tab: Fuel Health
  • Panel 1: S.TRIM and L.TRIM distributions (histogram) with ±5% / ±10% reference lines
  • Panel 2: S.TRIM mean binned into RPM × MAP grid — shows where the car is running rich/lean
  • Panel 3: wideband Wide vs AFCMD scatter — shows closed-loop tracking accuracy
  • Flags: if |L.TRIM mean| > 8% suggest AFM correction; if |L.TRIM| > 15% warn ECU limit approaching

B3 — Ignition table visualiser

No current table visualisation beyond AFM flow.

  • New tab: Ignition
  • Load Ignition Low (or first Table2D named "Ignition*") from tables[]
  • Render as a 2D heatmap: RPM (X) × Load (Y), colour = degrees
  • Overlay datalog density: shade cells by how many time-series frames hit that RPM/load bin
  • Tooltip: hover shows RPM, load, timing value, frame count
  • No import generation needed for this tab (read-only visualisation)

B4 — Before / after diff viewer

Currently there is no way to compare two exports.

  • File menu: Open Reference Export (loads a second tune_export.json)
  • New tab: Diff
  • List all tables present in both exports
  • For Parameter tables: show old → new value, highlight changed
  • For 1D/2D tables: show max delta, colour cells that changed
  • Export: auto-generate tune_import.json from the diff (only changed tables)
  • Useful for reviewing what an AI actually changed before applying

B5 — Multi-log support

The analyzer currently assumes a single export file with one log.

  • File > Open Export can load any tune_export.json
  • If datalog.count > 1, show a log selector dropdown in the toolbar
  • All tabs filter their time-series to the selected log
  • Summary tab: show stats side-by-side for all logs in the file

B6 — Sensor range anomaly detection

No automated flagging of out-of-range datalog values.

  • Add a Sensor Sanity report (can live in the Health tab or a new tab)
  • Define expected physical ranges per sensor: 
    EXPECTED = {
  "ECT":   (−10, 120),   # °C
  "IAT":   (−20, 70),
  "MAP":   (0.0, 3.5),   # Bar (allow boost)
  "BAT":   (10.5, 15.5), # V
  "Wide":  (10.0, 18.0), # AFR
  "RPM":   (0, 10000),
  "FuelP": (20, 120),    # PSI
}
  • Flag any sensor whose min or max falls outside the expected range
  • Show as a colour-coded table (green/yellow/red)

B7 — Export summary report (text / clipboard)

Users frequently copy-paste tune data into AI chat prompts.

  • File > Copy Summary to Clipboard
  • Generates a compact human-readable text block:
    • Key parameters (injector size, closed loop state, fuel trim limits)
    • Sensor stats for the 10 most important sensors
    • Any anomalies from B6
    • AFM correction recommendations if available
  • Plain text, suitable for pasting into ChatGPT / Claude prompt

Stream C — Bridge (JSON Contract)

These are format changes that require coordinated updates to both the Lua plugin
and the Python tool.

C1 — Add "category" field to table entries (depends on A5)

Python tool uses category to filter which tables to show in each analysis tab.
No structural change; just an additive field.

C2 — Add "device" block (depends on A4)

Python tool can show device/ECU status in a status bar or info panel.
Additive; old exports without the field degrade gracefully.

C3 — Add "hist" to sensor stats (depends on A6)

Python tool uses pre-computed histogram buckets for faster rendering when
time-series is not loaded or was skipped.
Additive; tool falls back to computing its own histogram from ts rows if absent.

C4 — Standardise tune_import.json for partial cell edits (depends on A3)

Python tool must generate the "cells" format when only a subset of cells changes
(e.g. B1 knock correction only touches cells where mean retard > threshold).