Field guide / Usage
From Run to Evidence.
Use an interface you own or can modify. Keep it on a stable route while you inspect one Scenario at a time.
Run, Baseline, and Mutation
Click the toolbar action on the current Target Page to create an Explicit Page Grant. The entire top-level light DOM is the Target Scope; the MVP does not offer a component picker. Choose Long Text, Unbreakable Text, or Large Text.
Each Run captures its own Baseline before applying temporary Mutations. Detectors acquire comparable Mutated Evidence using the same configuration and viewport. A Run stays active while you inspect it; collapsing, losing, or remounting the Extension UI does not Restore the page.
Read a Finding
A Finding requires a new or measurably worsened supported regression on the same connected Logical Element. Read the Baseline Evidence, Mutated Evidence, and Measured Delta. Possible Cause is a hypothesis for investigation, not a proven diagnosis.
For Text Clipping and Horizontal Containment Overflow, the Finding Subject is the responsible boundary, which may differ from the affected text or element. For Viewport Overflow, the subject is the Target Page; start inspection at its Primary Contributor.
Highlight or navigate to the live subject while the Run is active. A Locator is diagnostic metadata, not identity: disconnected or replaced nodes are never rematched to revive an action. Restore releases live Finding references.
Coverage and inconclusive results
Coverage reports eligible, mutated, excluded, skipped, safe-failed, ineffective, comparable, inconclusive, contributor, and Finding counts. These counts describe this Run and Scenario, not a percentage of application quality.
Unstable geometry, replaced nodes, unsupported clipping, or an ineffective style Mutation can make a comparison inconclusive. An inconclusive result is not a Finding. No Findings means no supported regression was demonstrated in the comparable measurements; it does not prove the page is correct.
Restore and reload-required
Choose explicit Restore when you finish inspection. The Engine processes its Mutation Journal in reverse order and restores only owned state that still exactly matches what it applied. External changes are preserved.
A Restore conflict or unverified cleanup enters reload-required and blocks another Run in that Document. An Unknown Mutation State aborts the Run, produces no Findings, and attempts best-effort Restore. Follow the displayed outcome and explicitly reload before continuing when required. There is no automatic reload or guaranteed recovery after abrupt runtime termination.
Navigation and interrupted connections
Observed top-level History, fragment, Navigation API, and back/forward entry changes end the grant. Restore is attempted immediately; the next route requires another toolbar action after verified cleanup. Navigation is never intercepted. Stable-route SPAs are supported; private router transitions without an observable entry change are outside this guarantee. Restore before those transitions. If the UI warns that navigation observation is incomplete, Restore before changing SPA routes.
Every Page Exit (pagehide) revokes the grant and attempts bounded synchronous local Restore without waiting for UI, messaging, storage, or another frame. Returning from BFCache never resumes the previous Run or authorization. Conflicts require a true reload; verified cleanup permits a fresh toolbar action.
A lost extension connection clears the Run indicator and displays a warning. Port loss does not prove Page Exit or BFCache: local Restore remains available for a live Run. Reopen the toolbar to reconnect. Abrupt tab, renderer, browser, or extension termination can prevent final cleanup or result delivery.
Current Run Result and Markdown Report
After completion, the Current Run Result is a reduced serialized result for this tab. Copy Report as Markdown copies exactly the visible deterministic preview on a user gesture, with no new page measurement. If the Clipboard API is denied or unavailable, select and copy the preview manually. There is no JSON, file, screenshot, or remote export.
After authorization ends, the saved result is a Previous-Target Snapshot. It remains readable and copyable, has no live DOM actions, and is not remeasured against the new page. Only one latest result is retained per tab session; a later completed Run replaces it. Clear saved result removes a previous snapshot, and closing the tab deletes its stored result. Active source values and Restore records stay in memory and are released at Run end.