Try a public pull request

Use the Lens Studio to read a real PR result without installing anything, then change its policy and see what that changes. Start with a retained teaching snapshot: its result does not depend on GitHub answering a new request, and its revisions stay fixed while you experiment. The browser can separately check whether that public PR has a newer head; the operating guide explains that request and its tab storage.

You need a browser. No account or token is required. Live acquisition accepts public GitHub.com PRs only; it is not a private-repository route or the managed App.

Open the lockfile lesson

Open the whole-change lockfile lesson. It selects the retained vitejs/vite#18968 teaching snapshot, head 48db5c9555f2d43cb88367506e79854fce8ed47d, captured September 18, 2026, with the standard catalogue policy (size@1).

The useful starting result is 22 Changed / 25 churn across 2 included files, classified as s. Both files and the evidence belong to that captured comparison, not to whatever the PR may contain when you open its GitHub page today.

Read Terminal for the shared human presentation. Agent · data exposes the tool-oriented result; Explanation connects facts to the policy. GitHub preview shows desired labels and an optional comment, not labels that the Lens Studio has applied or read back from GitHub.

Change one explicit exclusion

Under Controls, enter pnpm-lock.yaml in Exclude and press Enter. Alternatively, open the catalogue’s Exclude lockfile variant for the same complete policy change.

The result becomes 1 Changed / 2 churn, with 1 included file and 1 excluded file. Its size band is now xs. The observed comparison still contains two files; the lockfile did not stop changing because you excluded it from this question.

Selected policy	Included Changed	Included churn	Observed / included / excluded files	Band
Whole change (standard)	22	25	2 / 2 / 0	s
Exclude lockfile (vite-without-lockfile)	1	2	2 / 1 / 1	xs

Control edits re-evaluate the acquired report rather than refetching the PR. Policy shows the ordinary configuration behind them. A setting that the visual controls cannot represent remains in the policy; switching views is not permission to discard it. Invalid edits show diagnostics and distinguish the last valid result from the current invalid input.

This lesson uses complete source and policy records from the shared example catalogue. It is not a second website-only calculation. The snapshot retains normalized evidence, not raw source lines; policy replay is supported, but reconstructing or remeasuring a patch from that snapshot is not.

Distinguish the snapshot from a fresh acquisition

Analyze latest deliberately asks for the source PR’s current comparison. Inspect the returned revisions and evidence again; this is a new acquisition, not an update to the frozen lesson’s facts. Back to snapshot restores the retained teaching edition.

To try a different public PR, choose Public pull request, enter its full GitHub PR URL, and select Analyze public PR. The input must name a public github.com PR, not an arbitrary URL. Do not paste a token or expect the Lens Studio to inherit private repository access from your browser’s GitHub session.

The output may be bounded, unknown, unmeasurable or limited by an incomplete file set. An exact raw counter does not supply missing edit blocks. A new PR may also differ from a teaching example in both file population and policy. Compare identities and origins before comparing numbers.

Export the result you just inspected

Open Export. Select report.json / plan.json, choose the report.json artifact, use Copy, and save that single JSON document as report.json. Do not join a report and plan into one JSON file. A plan is a separate artifact selectable in the same tab.

After the exclusion, also copy the .diffdevil.yml tab into that file. The CLI tab gives read-only commands for the saved comparison. With Node.js and the CLI execution route available, the excluded lockfile example is:

npx @wolfsblvt/diffdevil analyze --report report.json --format human
npx @wolfsblvt/diffdevil query --report report.json --metric changed --format value
npx @wolfsblvt/diffdevil plan --report report.json --config .diffdevil.yml --target-repo vitejs/vite --target-pr 18968 --format human

The scalar prints 1. The plan desires size/XS and reports that nothing was applied; it does not contact or modify Vite’s repository. The report commands retain the exported comparison. Planning explicitly uses the saved policy rather than an unrelated policy discovered in your current checkout.

An Action step is not a complete workflow. The separate Complete workflow export describes a workflow for your own repository and states whether it writes effects. A preview is not evidence of permission, activation or provider readback. Continue with Label pull requests when reviewed repository automation is the job you want to adopt.

When the result cannot be obtained

Invalid or unavailable PR input: correct the URL and confirm that the PR is public. A private, absent or inaccessible result is not a zero-sized PR.

Read budget or provider failure: retain the input and last useful result, respect the shown retry information, or return to a saved lesson. Repeatedly changing thresholds will not repair acquisition. Local analysis is a separate route, not a claim that the failed request succeeded.

Policy error: repair the marked declaration in Policy. Do not mistake the visibly retained previous valid result for output from the invalid edit.

An export refuses an exact value: read its exit and evidence. Use canonical query JSON for unresolved values rather than coercing a range to a midpoint or an absent path set to an empty list. Use results in scripts gives complete consumers.

Explore a different lesson

The Examples gallery and these links use the same catalogue. Each link selects a declared example and variant:

Example	Lessons	Inspect
The lockfile is still part of the change	scope, counting, policies	Open in Lens Studio
A small helper can touch a wide surface	distribution, scope, counting	Open in Lens Studio
Snapshots are material, not invisible background	distribution, scope	Open in Lens Studio
Binary files still count as evidence	evidence, scope	Open in Lens Studio
Similar change size, radically different spread	distribution, counting	Open in Lens Studio
A bounded value can still prove a policy	evidence, policies	Open in Lens Studio
Observed files are not the whole inventory	evidence	Open in Lens Studio

Use Lens Studio for the deeper operating guide, Evidence and uncertainty for interpretation, and Security and data for the data boundary. A frozen result, a fresh analysis, a desired plan and an applied effect remain four different things.