Claude and Claude Code
For code, tests, research summaries and drafts, directed from a written project brief and a running state file. No paid APIs in the product. Netlify for hosting.
← Back to work
AI TRANSPARENCY
How I used AI.
Not hidden. Not exaggerated. Here is the split.
Every page on this site says “AI-assisted build.” This page says what that means. I am an oilfield operator, not a software engineer by training, and I used AI as the hands while I supplied the problem, the rules and the judgment. A hiring manager or a company evaluating a pilot deserves to know which is which.
01 / The split
What I did.
What AI did.
| Area | Me | AI (Claude and Claude Code) | How I checked |
|---|---|---|---|
| The problem | Lived it. The handover write-up, the data-van duties, the missed deliveries, the hitch rotation, the time nobody could prove. I wrote the “why” and decided what the tool is for and what it is not. | Nothing. AI did not pick the problem. | Read every page as the crew member I was, not as a developer. |
| Rules and field knowledge | Stage limits, what an alert should say, 14-on 7-off with day and night crews, how overtime and holidays actually get worked, which vendor tools crews really use, what finance needs from a timesheet. | Turned my rules into code and tests. Suggested textbook methods (net-pressure slope, hydraulic horsepower) and I accepted the ones that match what I used in the field. | Corrected rules AI got wrong: night-shift runs counted as 15 days instead of 14, a late order ranked as only at-risk, a page that returned one row when asked for none. |
| Research | Named the systems my crews use (Dastrac, FracPro, FieldPro) and the companies to look at. | Read public product pages and labor-rule sources and summarized them with links. | Anything AI could not find a public source for is marked unverified on the site, including vendor file formats. |
| Code | Reviewed behavior, not syntax. Decided what gets built next and what gets cut. | Wrote most of the HTML, CSS and JavaScript, the simulation, the command layer and the test files. | Ran the test suite after every change (136 tests at the time of writing). Clicked through each page on desktop and phone widths. |
| Writing | Wrote the personal story and approved every public sentence. Set the rule that nothing claims a customer, a deployment or a result that did not happen. | Drafted page copy and the comparison tables. | Struck claims AI could not back with a source or a test. |
| Design | Chose what each audience (crew, supervisor, back office) should see first. | Produced layouts, the charts and the share images. | Checked contrast and phone layout; fixed pale text AI shipped on light backgrounds. |
02 / Tools and limits
What I used.
Where I stop.
What you can inspect
The authorship record, the decision log in the AI Lab, the change log, and the tests. The AI Lab publishes one known failure instead of hiding it.
What I would not hand to AI here
Deciding fault from a record. Writing an employee’s time for them without a reason on file. Vendor file formats I have not seen. Overtime law. Each of these stays with a person.
If you are evaluating me for a role: I can explain every rule in ShiftLedger and why it is there, and I can run and read the tests. I cannot yet write the code from a blank file without help, and I am learning that. If you are evaluating ShiftLedger for a pilot: the field logic is mine and tested; the code was AI-written and should be reviewed by your engineers like any other contractor’s code.