Evidence-based audits of one public website journey: enquiry → qualification → appointment or quote request.
Proofloop looks at what a company's website actually shows, links every finding to its evidence, and works out whether a fix needs a UX change, a simple automation, an integration, AI, or nothing at all. Savings are calculated from visible, labelled inputs, never guessed from a website.
Website audits, analytics and automation assessments already exist. Proofloop's aim is better evidence and clearer reasoning, not a new product category.
Status: planning, October 2026. No code yet: the brief, spec and tickets are worked out in the open in this repository. First complete version due January 2027.
- What can we actually observe?
- What improvement does that observation suggest?
- Simpler software, automation, AI, or no change?
- What extra information would establish the business case?
- How would we test whether the improvement works?
- What happened when we tested it?
The spec is still being settled, so details will change.
- Evidence collection for up to five approved public pages: page text, screenshots, URLs, timestamps and performance measurements (Playwright, Lighthouse). Failed checks and missing pages are shown, not hidden.
- At most three priority findings, each linked to its evidence and labelled as observed fact, hypothesis, estimate or measured result.
- A transparent savings calculator: hours released, break-even point, costs and sensitivity, with every assumption visible and every fictional input labelled. Hours released are not counted as cash saved.
- Setup recipes for automation findings, including importable n8n workflows.
- Downloadable one-page reports per finding.
- A small validation flow: hypothesis, one success metric, safeguards, baseline, result and limitations.
- A demo on a fictional business with deliberately introduced problems, including one recommendation that does not pay off.
- Infer revenue, conversion losses or staff workload from a website alone.
- Submit forms, sign in, buy anything or change a third-party site.
- Let website content steer the analysis model.
BRIEF.md: the original requirements, verbatim.docs/spec.md: the spec in progress (ideas, research, decisions).CLAUDE.md: rules for the AI agents that build this project. how-i-build explains the workflow behind them.