In case you missed me, haha!
I've been neglecting the Ranchero in favor of learning about coding, AI, and the pitfalls of trusting AI not to hallucinate or go wandering off into the weeds like a puppy with adhd. All the while making my '69 Chevy C10 with a Sniper EFI-d 350 run better.
I'm no coding or tuning guru, but I have been using this on my own vehicle as I developed it, and it's already helped me fix 2 nagging issues I had for awhile and a couple hidden issues that I had but couldn't see - and the truck is running pretty dang well, if I do say so myself!
Also, I built this with Holley Sniper 1 in mind 'cause that's what I have, but it seems to me that this would be useful with the other Holley systems, as well as non-Holley EFI - if you can get key tune data and datalogs into it, it should work - but ymmv.
+++++++++++++++++++++++++++++++++++++++
TL;DR: I've built an Excel-Python-AI pipeline that takes the data from your Holley Sniper calibration tables and datalogs, and turns them into a structured AI tuning session. It gives you specific VE table corrections, Target AFR sanity checks, idle diagnostics, cold start analysis, accel enrichment review, and a drive script for your next session — all referenced back to your actual data.
Looking for feedback on the design and people willing to test it on their own vehicles.
If you're interested, let me know via the email at the bottom of this post, and I'll shoot you a link to my google drive so you can download everything (including one of my recent tuning session files so you can play around with it.)
++++++++++++++++++++++++++++++++++++++
What it is
This is a tuning assistant system built around the Holley Sniper EFI. The core idea: the Holley software gives you the data, but it doesn't tell you what to do with it. Getting from "here's my datalog" to "here's what to change and why" requires either experience or a lot of forum digging. This system tries to close that gap using computers.
It's not a magic button. It doesn't connect to your ECU. It doesn't replace your judgment. What it does is process your calibration data and datalogs into a structured format, run a set of pre-computed checks, and then conduct a disciplined AI tuning session that produces specific recommendations with the reasoning behind every one.
The three-part stack
1. Excel workbook (.xlsx, no macros, no VBA)
You paste your calibration tables from the Holley software into the workbook — VE table, Target AFR table, accel enrichment, cold start tables, idle settings, etc. Then you enter your vehicle metadata, hardware configuration, and tune state into one sheet. The workbook becomes a dumb data container - a copy of your config file + context.
2. Python 3.8+, and a single script
A GUI launcher wraps all of this into a simple GUI.
All the logic and calculations live in the Python script, which reads the workbook and your datalog CSV files and does the heavy lifting before the AI ever sees the data. It:
Classifies every datalog row by operating state (cranking, warmup, idle, cruise, accel transient, decel, WOT)
Maps actual vs. target AFR across the RPM/MAP grid and computes cell-level error
Detects lean spikes and closed-loop anomalies
Builds coverage maps (which cells in the VE table actually have enough data to recommend changes)
Runs hardware checks, data integrity checks, and parsing verification
If you have speed/gear data logged, builds per-gear AFR statistics and detects gear-conflict cells where corrections shouldn't be applied
Checks your Target AFR table against community-validated danger and warning ranges by operating region (idle, cruise, WOT, transition, decel)
Exports a self-contained JSON file that the AI reads — no raw CSVs go to the AI, just structured pre-computed results
3. AI tuning session (Claude-primary, designed to work cross-platform)
You load a Core Prompt plus your JSON into the AI. The AI confirms what it sees, asks you up to three short questions about symptoms and driving context, identifies your current tuning stage, and then requests the matching Step Module — a focused instruction set for that specific stage of tuning. Structural gates prevent the AI from jumping ahead: it literally doesn't have the analysis instructions until you upload the right module.
There are seven step modules covering the full tuning sequence: Target AFR sanity and idle setup → Base Fuel VE (warm engine) → optional Speed/Gear calibration → optional Target AFR optimization → cold start and warm-up → acceleration enrichment → final refinement and regression check.
The output is a structured report with pre-analysis checks, vehicle context, operating summary, step-specific findings, and specific table change recommendations — plus a ready-to-paste tuning log entry and a drive script for the next session.
The AI can also write recommended VE corrections directly back into the workbook (with a dry-run preview first), so you're not hand-entering numbers (this is new, so don't run off and blow up your config files)
Why it's built this way
A few design decisions that might not be obvious:
* Pre-computation in Python, not in the AI. Early versions asked the AI to compute cell-level VE corrections, average AFR errors, and coverage statistics from raw data. AI models are inconsistent at arithmetic under long prompts. Moving numeric work to Python makes the results deterministic and auditable. The AI sees pre-computed results and does what it's actually good at: pattern recognition, diagnosis, prioritization, and explanation.
* Modular prompts instead of one big prompt. The previous version used a single 1,100-line prompt that covered all tuning stages. Testing showed that on some AI platforms, long stateful prompts cause the AI to skip steps, invent its own structure, or fabricate compliance. Breaking the system into a Core Prompt plus per-stage Step Modules keeps each session load under 1,200 lines, which is within the reliable instruction-following range of most capable models.
* No ECU connection, manual paste workflow. The Holley software is proprietary and Windows-only. Rather than trying to reverse-engineer the file format or build an API bridge, the workbook uses a manual copy-paste workflow. It's a real friction point, but it's also the most durable approach — it shouldn't break when Holley updates their software.
Current state
This was built and tested on a 1969 Chevrolet C10 with a 350 V8, 4L60E automatic, and a basic Holley Sniper 4bbl (2019). The system is on version 5.0, has gone through about 16 development sessions and 10 tuning sessions, and is in active use. The architecture is stable. The step modules covering idle/Target AFR, base fuel VE, speed/gear calibration, Target AFR optimization, cold start, accel enrichment, and final refinement are all complete.
What hasn't been tested: any vehicle other than mine. That's the gap.
What I'm looking for:
1. Design feedback — Does the architecture make sense? Are there obvious problems with the approach, missing pieces, things that would make it useless for your setup?
2. Testers — If you have a Holley Sniper (any variant, or even another EFI system and are willing to play) and are willing to go through the setup process, I'd like to know how it performs on a different vehicle. The setup requires Python 3.8+ and the ability to paste calibration tables from the Holley software. It's not plug-and-play, but it's documented. Claude.ai is the recommended AI platform (free tier works for a session, Pro is more comfortable). The system is designed to work on other capable AI platforms as a cross-check, though results vary.
** If you're interested in being a tester, email me at: jbmyers1963@gmail.com and I'll send you a link to my google drive.
Post any questions here, I'll answer them as best I can.
I've been neglecting the Ranchero in favor of learning about coding, AI, and the pitfalls of trusting AI not to hallucinate or go wandering off into the weeds like a puppy with adhd. All the while making my '69 Chevy C10 with a Sniper EFI-d 350 run better.
I'm no coding or tuning guru, but I have been using this on my own vehicle as I developed it, and it's already helped me fix 2 nagging issues I had for awhile and a couple hidden issues that I had but couldn't see - and the truck is running pretty dang well, if I do say so myself!
Also, I built this with Holley Sniper 1 in mind 'cause that's what I have, but it seems to me that this would be useful with the other Holley systems, as well as non-Holley EFI - if you can get key tune data and datalogs into it, it should work - but ymmv.
+++++++++++++++++++++++++++++++++++++++
TL;DR: I've built an Excel-Python-AI pipeline that takes the data from your Holley Sniper calibration tables and datalogs, and turns them into a structured AI tuning session. It gives you specific VE table corrections, Target AFR sanity checks, idle diagnostics, cold start analysis, accel enrichment review, and a drive script for your next session — all referenced back to your actual data.
Looking for feedback on the design and people willing to test it on their own vehicles.
If you're interested, let me know via the email at the bottom of this post, and I'll shoot you a link to my google drive so you can download everything (including one of my recent tuning session files so you can play around with it.)
++++++++++++++++++++++++++++++++++++++
What it is
This is a tuning assistant system built around the Holley Sniper EFI. The core idea: the Holley software gives you the data, but it doesn't tell you what to do with it. Getting from "here's my datalog" to "here's what to change and why" requires either experience or a lot of forum digging. This system tries to close that gap using computers.
It's not a magic button. It doesn't connect to your ECU. It doesn't replace your judgment. What it does is process your calibration data and datalogs into a structured format, run a set of pre-computed checks, and then conduct a disciplined AI tuning session that produces specific recommendations with the reasoning behind every one.
The three-part stack
1. Excel workbook (.xlsx, no macros, no VBA)
You paste your calibration tables from the Holley software into the workbook — VE table, Target AFR table, accel enrichment, cold start tables, idle settings, etc. Then you enter your vehicle metadata, hardware configuration, and tune state into one sheet. The workbook becomes a dumb data container - a copy of your config file + context.
2. Python 3.8+, and a single script
A GUI launcher wraps all of this into a simple GUI.
All the logic and calculations live in the Python script, which reads the workbook and your datalog CSV files and does the heavy lifting before the AI ever sees the data. It:
Classifies every datalog row by operating state (cranking, warmup, idle, cruise, accel transient, decel, WOT)
Maps actual vs. target AFR across the RPM/MAP grid and computes cell-level error
Detects lean spikes and closed-loop anomalies
Builds coverage maps (which cells in the VE table actually have enough data to recommend changes)
Runs hardware checks, data integrity checks, and parsing verification
If you have speed/gear data logged, builds per-gear AFR statistics and detects gear-conflict cells where corrections shouldn't be applied
Checks your Target AFR table against community-validated danger and warning ranges by operating region (idle, cruise, WOT, transition, decel)
Exports a self-contained JSON file that the AI reads — no raw CSVs go to the AI, just structured pre-computed results
3. AI tuning session (Claude-primary, designed to work cross-platform)
You load a Core Prompt plus your JSON into the AI. The AI confirms what it sees, asks you up to three short questions about symptoms and driving context, identifies your current tuning stage, and then requests the matching Step Module — a focused instruction set for that specific stage of tuning. Structural gates prevent the AI from jumping ahead: it literally doesn't have the analysis instructions until you upload the right module.
There are seven step modules covering the full tuning sequence: Target AFR sanity and idle setup → Base Fuel VE (warm engine) → optional Speed/Gear calibration → optional Target AFR optimization → cold start and warm-up → acceleration enrichment → final refinement and regression check.
The output is a structured report with pre-analysis checks, vehicle context, operating summary, step-specific findings, and specific table change recommendations — plus a ready-to-paste tuning log entry and a drive script for the next session.
The AI can also write recommended VE corrections directly back into the workbook (with a dry-run preview first), so you're not hand-entering numbers (this is new, so don't run off and blow up your config files)
Why it's built this way
A few design decisions that might not be obvious:
* Pre-computation in Python, not in the AI. Early versions asked the AI to compute cell-level VE corrections, average AFR errors, and coverage statistics from raw data. AI models are inconsistent at arithmetic under long prompts. Moving numeric work to Python makes the results deterministic and auditable. The AI sees pre-computed results and does what it's actually good at: pattern recognition, diagnosis, prioritization, and explanation.
* Modular prompts instead of one big prompt. The previous version used a single 1,100-line prompt that covered all tuning stages. Testing showed that on some AI platforms, long stateful prompts cause the AI to skip steps, invent its own structure, or fabricate compliance. Breaking the system into a Core Prompt plus per-stage Step Modules keeps each session load under 1,200 lines, which is within the reliable instruction-following range of most capable models.
* No ECU connection, manual paste workflow. The Holley software is proprietary and Windows-only. Rather than trying to reverse-engineer the file format or build an API bridge, the workbook uses a manual copy-paste workflow. It's a real friction point, but it's also the most durable approach — it shouldn't break when Holley updates their software.
Current state
This was built and tested on a 1969 Chevrolet C10 with a 350 V8, 4L60E automatic, and a basic Holley Sniper 4bbl (2019). The system is on version 5.0, has gone through about 16 development sessions and 10 tuning sessions, and is in active use. The architecture is stable. The step modules covering idle/Target AFR, base fuel VE, speed/gear calibration, Target AFR optimization, cold start, accel enrichment, and final refinement are all complete.
What hasn't been tested: any vehicle other than mine. That's the gap.
What I'm looking for:
1. Design feedback — Does the architecture make sense? Are there obvious problems with the approach, missing pieces, things that would make it useless for your setup?
2. Testers — If you have a Holley Sniper (any variant, or even another EFI system and are willing to play) and are willing to go through the setup process, I'd like to know how it performs on a different vehicle. The setup requires Python 3.8+ and the ability to paste calibration tables from the Holley software. It's not plug-and-play, but it's documented. Claude.ai is the recommended AI platform (free tier works for a session, Pro is more comfortable). The system is designed to work on other capable AI platforms as a cross-check, though results vary.
** If you're interested in being a tester, email me at: jbmyers1963@gmail.com and I'll send you a link to my google drive.
Post any questions here, I'll answer them as best I can.
