Open the Center
Create or open a workspace, then choose Chronicle Center from the landing page, Workspace menu, or toolbar. The Center reloads from the same controller used by Federation Explorer and Design federation.
Overview
Confirm the workspace file, root project, generation, fingerprint, trust state, federation counts, and pending proposal count. A trust decision is fingerprint-bound; any later non-trust command invalidates it.
History
Each command row shows its result generation, kind, audit description, and affected identities. Adding or editing workspace content appears here immediately. The workspace also archives exact prior JSON documents. History is immutable; the editor's Undo and Redo buttons create later compensating generations rather than silent rewinds.
Recovery
- Select a candidate and inspect its generation, command count, difference summary, and fingerprint tooltip.
- Choose Review and promote.
- Read the confirmation. Promotion proceeds only if the candidate’s base fingerprint still equals the current saved workspace.
- If the workspace moved, promotion stops without replacement. Reload and reassess instead of forcing it.
Agent trust
Trust a signer only after confirming its complete Ed25519 public-key ID through a separate channel. The key list shows active, expired, and permanently revoked records. Admission thresholds set the minimum independent-benchmark accuracy, maximum permitted regression, and minimum benchmark rows. Threshold changes immediately govern pending and previously accepted agents before a later proposal is accepted or applied. See Trust agent signers and set regression limits for the exact workflow.
Proposals
A proposal binds a subject to an artifact fingerprint and an evidence fingerprint. Select a pending proposal, enter the accountable review authority and note, then choose Accept, Reject, or Mark superseded.
| Status | Meaning | What it does not mean |
|---|---|---|
| Pending | Evidence awaits a named decision. | The artifact is not active. |
| Accepted | The displayed evidence was reviewed and accepted as a proposal. | Acceptance alone performs no model selection, workspace mutation, or build. |
| Applied | An accepted agent action plan was separately revalidated and applied through Chronicle commands. | No VFS image or Perl package is built automatically. |
| Rejected | The proposal is not suitable on the reviewed basis. | The underlying candidate files are not deleted. |
| Superseded | A newer or more appropriate proposal replaced its review relevance. | No replacement is automatically applied. |
Governed agent plans
- Choose Register agent bundle. Studio first verifies its trusted unexpired signature, independent benchmark, accuracy floor, and regression limit.
- Inspect the signer key ID and expiry, bundle fingerprint, quality, regression evidence, labels, artifacts, and authority block, then resolve its bundle proposal.
- Choose Import action proposal. Studio requires that exact signed agent to remain accepted and dry-runs every action against the current workspace.
- Inspect the plan summary, rationale, numbered actions, before/after counts, destructive count, and warnings. Accepting it remains review-only.
- For an accepted, current plan, enter a separate application authority and note, then choose Apply accepted plan. Studio validates the agent and plan again and saves the full batch atomically.
See Train an agent to propose workspace changes for the complete training-to-export walkthrough.
Perl NN linkage
When a workspace is active, explicit approval in AI Workbench registers the exact candidate as a pending perl-nn-main-candidate proposal. Repeating approval for the same fingerprint is idempotent. The machine-specific candidate directory is deliberately excluded from the portable evidence fingerprint.