GDP Desktop can collect bounded local loop evidence, turn it into a compact reviewer context pack, preview a provider payload with credential references masked, and rehearse the reviewer-to-queue handoff locally. These tools prepare and inspect evidence; they do not themselves apply patches, run commands, contact a provider, or prove a real queue write.
Read the active repository's ProofLoop Evidence
ProofLoop Evidence is a read-only local report for the active repository. It brings together the current automation-loop summary, recent patch-intake state, browser-relay context, and bounded check-run evidence so a reviewer can reason from GDP-owned state instead of a conversation-only recollection.
Copying or inspecting this report does not execute a check, apply a patch, change automation policy, or advance a queue item. It is an evidence view of what GDP currently knows.
Use it to reconcile what has already happened before proposing the next action.
Treat missing or unknown evidence as missing or unknown rather than filling the gap from model confidence.
Read the exact detailed run or queue record when the compact summary is not enough.
Build a compact Closed-Loop Context Export
Desktop Context Export packages the current ProofLoop evidence into a compact reviewer-oriented JSON context with a bounded continuation instruction. It is intended to help a reviewer continue from the evidence without re-sending an unlimited working history.
This Desktop Closed-Loop Context Export is not the same artifact as Browser Companion Full Context Export. The Companion export preserves conversation/continuation material for its own handoff and audit workflow; the Desktop export is a compact reviewer pack derived from local ProofLoop evidence.
Neither export grants repository authority. Imported or copied context remains evidence/context and must not be treated as permission to execute.
Preview the provider payload without exposing the credential reference
Provider Preview loads the exported reviewer context and builds the bounded provider payload that would be handed to the reviewer adapter. The displayed preview masks the configured credential reference instead of exposing the underlying reference value.
Preview is still local inspection. It does not contact the external provider, execute repository commands, write Patch Intake, or prove that a downstream service is available.
The provider/model/system/user payload shape is bounded before transport.
Credential references are represented by a masked label in the preview.
A useful preview says what would be sent; it is not a receipt that anything was sent.
Treat Provider Dry Run as a local handoff rehearsal
Provider Dry Run exercises the reviewer-turn parser and queue-only handoff contract with a local mock reviewer and a local mock enqueue callback. The mock can return an actionable patch-shaped reviewer response so GDP can prove that parsing and handoff wiring behave as expected without reaching an outside provider.
A dry-run report may contain queue-like fields because the mock callback returns a synthetic handoff result. The current contract explicitly reports that no external call was made and no real Patch Intake queue item was written; the synthetic incoming identity is dry-run-only.
external_call_made=false means the dry run did not contact a real provider.
real_queue_written=false means a mock queued result is not an Incoming patch receipt.
Reviewer transport is queue-only: it does not directly apply, commit, push, reset, clean, or run repository commands.
Oversized or truncated patch output is rejected rather than silently queueing a partial patch.
Use real GDP intake and postcondition evidence after an actual reviewer/provider turn. Dry Run proves the local rehearsal contract only.
