Review the run evidence
The task detail surface keeps the price chart, artifact evidence, cashback, and delivery trail together.
Compare Preview
Compare first. Keep the right candidate alive.
Language
Compare evidence for AI-assisted decisions
Check the product target before you create durable state.
Compare Preview exists because ordinary price trackers assume you already know the right product URL. DealWatch starts one step earlier: validate the candidate, compare across stores, inspect deterministic evidence plus AI-assisted explanation, and only then decide whether the row deserves a watch task or a compare-aware watch group.
AI-assisted explanation stays on top of deterministic evidence. These public pages are read-only proof surfaces, not hosted automation.
2 URLs in, 0 durable writes
Compare Preview stays read-only until you intentionally create a task.
1 live API route
`POST /api/compare/preview` is the route behind the first product step.
AI explains, evidence stays primary
The current product uses AI-assisted compare explanations, but deterministic compare evidence still remains the source of truth.
Static evidence beats a cropped walkthrough here: this hero shows the same full Compare Preview state the sample fixture renders below.
What you can understand in under a minute
The page should make the product feel legible before you ever bring up the local stack.
What to look for
- Which URLs resolved into real candidate offers
- Whether the two pear listings look like the same product
- Why the chip listing should stay out of that match
What this sample proves
- Compare Preview can surface a strong cross-store match before any task is created.
- The same sample also shows an obvious distractor, so “different” is visible too.
- This page loads a fixed public fixture only. It does not save anything or call a live write path.
What this sample is not
- Not a live public compare API
- Not a write path that creates watch tasks or watch groups
- Not a promise that your own URLs will work until you run the local stack
- Not a shopping chatbot, and not the live local recommendation card. The shipped recommendation surface stays inside the local runtime Compare Preview flow.
Try the sample compare preview right now
This public sample loads a fixed compare fixture from the site itself. It is static, read-only, and saves nothing. Its only job is to show the first product step in under a minute.
Load the sample compare to inspect a fixed public fixture. Read-only. No data is saved.
Resolved comparisons
The UI shows whether each submitted URL was supported and fetched successfully.
Match reasoning
Match score and normalized candidate details make the compare decision visible instead of implicit.
What happens after a good compare result
The whole point is deciding whether the candidate deserves a durable watch task at all, or whether it should stay in a long-lived compare-aware group.
Choose the next door after the sample
Once the static sample makes sense, most visitors should read Proof next, then move into Quick Start when they want their own URLs. Builders is the specialist route for read-only agent/client wiring after the product story is already clear.
Read the proof trail
Use Proof when you want the command, verifier, and boundary map behind the public story before you install anything.
Run Quick Start
Use this path when the sample and proof already make sense and you want the local API, worker, and WebUI for your own grocery URLs.
Wire a builder client
Use Builders when your real job is wiring the catalog, starter pack, and read-only agent-facing surfaces for Codex, Claude Code, and similar clients.
What happens after a good compare result
Compare Preview is the first proof surface, not the whole product. A strong public story should also show where the picked candidate lands next.
Tune the alert policy
Notification settings explain how the compare-first decision turns into a sustainable alert loop instead of one-off clicks.