prooftrail

AI Reconstruction Side Road

ProofTrail’s AI reconstruction path is for the moment after artifacts already exist and a human or AI-adjacent workflow still wants help rebuilding or refining a flow.

It is not the public default road.

If you are still asking “Did the first canonical run even work?”, go back to:

  1. README.md
  2. docs/getting-started/human-first-10-min.md
  3. docs/reference/run-evidence-example.md

Why AI builders still care

AI reconstruction matters here because it gives an AI-relevant capability without redefining the repo into a generic AI browser bot.

The rule is simple:

Why this side road exists at all

ProofTrail is not trying to win on generic browser autonomy.

This side road exists because browser evidence is often rich enough to help an AI system rebuild intent, suggest a draft, or recover flow structure after the first run. That is useful. It is also exactly where teams can drift into wishful “let the model figure it out” behavior if the boundary is not made explicit.

So the product stance is intentionally narrow:

When to use this side road

Use AI reconstruction when:

Do not use it when:

Inputs, outputs, and guardrails

Layer What belongs here What does not belong here
Inputs retained HAR, HTML, video, session artifacts, evidence bundles a blank prompt with no artifact ground truth
Outputs reconstructed drafts, template candidates, profile hints, reviewable next steps silent autonomous promotion straight into default production use
Guardrails preview first, inspect artifacts, keep human review in the loop treating generated output as proof on its own

What this path does

In ProofTrail, AI reconstruction helps with:

  1. resolving a likely profile from artifacts
  2. previewing a reconstructed draft
  3. generating a flow or template candidate
  4. handing that draft back to the operator workflow for human review

The product rule is simple:

AI reconstruction is an optional assistant after proof exists, not a prerequisite before proof exists.

How it reconnects to the main product

This side road only makes sense when it flows back into the normal operator loop:

  1. evidence is retained from the canonical run
  2. reconstruction proposes a draft or likely profile
  3. Flow Workshop previews or refines that draft
  4. operators inspect the result before reuse, compare, or promotion

That handoff matters more than the generation step itself.

Product path

Use the path in this order:

  1. run the canonical path first
  2. inspect retained evidence in Task Center
  3. open Flow Workshop only when artifact-driven reconstruction is actually useful
  4. preview before you generate
  5. review before you reuse

Why this matters for AI agents and operators

Generic AI browser tools often optimize for open-ended autonomy.

ProofTrail optimizes for a different thing:

That is exactly why it matters to AI-agent builders:

Do not confuse this with a generic AI agent loop

If you want a system whose main value is open-ended browser autonomy, this page should probably push you toward the alternatives framing instead of trying to stretch ProofTrail into something else.

If you want AI help that starts from retained artifacts, stays reviewable, and feeds back into a governed operator workflow, this side road is the right fit.

Next pages