Potter's Wheel

Ask an agent to

• Verify the selected claim against its linked evidence
• Attach this experiment run as evidence for a figure
• List claims that still need review

37 tools

• add_experiment_metricuntrusted content

  Add or update a named metric on a research run.

  experimentId: string, key: string, value

• add_experiment_parameteruntrusted content

  Add or update a named configuration value on a research run.

  experimentId: string, key: string, value

• attach_evidenceuntrusted content

  Attach structured evidence to the selected or explicitly identified tracked manuscript object. This visibly updates the same provenance state shown in the integrity panel.

  objectId: string?, type: string?, label: string, artifactId: string?, experimentId: string?, commit: string?, metric: string?, uri: string?, … (list cut; the site may define more)

• attach_experiment_artifactuntrusted content

  Attach a result, figure, table, configuration, or other artifact to a research run.

  experimentId: string, label: string, type: string?, uri: string?, id: string?

• compare_experimentsread-onlyuntrusted content

  Compare the configuration, metrics, datasets, artifacts, and provenance impact of two experiment runs.

  baseExperimentId: string, candidateExperimentId: string

• create_claimuntrusted content

  Create a first-class provenance claim from the researcher's current explicit text selection. Requires the exact text previously read by the agent so stale selections fail safely.

  expectedText: string

• create_experimentuntrusted content

  Create a run in the same Experiments workspace the researcher uses.

  id: string?, name: string, status: string?, method: string?, sourceCommit: string?, notes: string?, startedAt: string?

• create_experiment_evidenceuntrusted content

  Turn a run metric, configuration, result, figure, or table artifact into durable evidence for the paper. A superseded evidence id rewires its manuscript links and creates Research Diffs.

  experimentId: string, label: string, type: string?, evidenceKind: string?, metric: string?, value?, baseline?, artifactId: string?, … (list cut; the site may define more)

• get_claimread-onlyuntrusted content

  Return the currently selected claim, or a tracked claim by id, including its verification state and linked evidence.

  claimId: string?

• get_current_sectionread-onlyuntrusted content

  Return the named manuscript section containing the researcher's current selection together with that section's live text.

• get_current_selectionread-onlyuntrusted content

  Return the researcher's live manuscript selection, section, object kind, and verification state without changing the document.

• get_evidence_catalogread-onlyuntrusted content

  Return evidence between experiments and the paper, including its originating run and manuscript usage.

  experimentId: string?, type: string?

• get_experimentread-onlyuntrusted content

  Return one run with its parameters, metrics, artifacts, produced evidence, and dependent manuscript objects.

  experimentId: string

• get_experimentsread-onlyuntrusted content

  Return the same persisted experiment runs visible in Potter’s Wheel, optionally filtered by run status.

  status: string?, limit: integer?

• get_figureread-onlyuntrusted content

  Return the currently selected figure, or a tracked figure by id, including its verification state and linked evidence.

  figureId: string?

• get_integrity_statusread-onlyuntrusted content

  Return live manuscript integrity counts and optional summaries of tracked provenance objects without changing researcher state.

  includeObjects: boolean?, limit: integer?

• get_manuscript_contextread-onlyuntrusted content

  Return live manuscript text and integrity context for the current selection, current section, or manuscript.

  scope: string?, maxChars: integer?

• get_navigation_targetsread-onlyuntrusted content

  List tracked manuscript objects that an agent may ask the researcher to inspect or navigate to, optionally filtered by object kind or verification state.

  kind: string?, verificationState: string?, limit: integer?

• get_provenanceread-onlyuntrusted content

  Return provenance and linked evidence for the currently selected tracked manuscript object, or for an explicit object id.

  objectId: string?

• get_research_diffsread-onlyuntrusted content

  Return evidence-driven Research Diff entries from live provenance alongside currently pending CriticMarkup manuscript proposals.

• get_tableread-onlyuntrusted content

  Return the currently selected table, or a tracked table by id, including its verification state and linked evidence.

  tableId: string?

• insert_commentuntrusted content

  Insert an additive visible comment on the current selection line or a tracked object's line. Existing comment threads are never overwritten; the tool fails if that line already has a comment.

  text: string, objectId: string?, line: integer?

• link_evidence_to_manuscriptuntrusted content

  Connect existing experiment evidence to an existing claim, figure, table, or method in the same visible workspace.

  evidenceId: string, objectId: string, relation: string?

• log_experiment_metric_pointuntrusted content

  Append a numeric metric point at a training step and update the run's latest value, like an MLflow metric history.

  experimentId: string, key: string, value: number, step: number?

• navigate_to_objectuntrusted content

  Move the visible editor selection to a tracked claim, method, figure, or table. This changes navigation state only; it does not modify manuscript or provenance content.

  objectId: string

• navigate_to_research_diffuntrusted content

  Move the visible editor selection to an evidence-driven Research Diff or pending CriticMarkup proposal so the researcher can inspect it without changing manuscript content.

  from: integer?, diffId: string?

• propose_claim_changeuntrusted content

  Stage a replacement for a tracked claim as CriticMarkup. This never silently accepts the new manuscript text: the researcher receives the editor's normal accept/reject controls.

  claimId: string?, expectedText: string, replacementText: string

• record_verification_resultuntrusted content

  Record an evidence-backed conclusion from an external research agent on a tracked claim. This stores the visible result and reasons but never changes manuscript text; use propose_claim_change separately for a…

  claimId: string?, outcome: string, reason: string, evidenceIds: array

• replace_selected_contentuntrusted content

  Stage a replacement for the researcher's current explicit selection. Despite the action name, the new text is not silently accepted: it is inserted as CriticMarkup and remains pending until the researcher accepts or…

  expectedText: string, replacementText: string

• review_research_diffuntrusted content

  Record a defer or reject decision for an evidence-driven Research Diff without changing manuscript text. Agents cannot accept manuscript rewrites.

  diffId: string, decision: string, reason: string?

• set_experiment_statusuntrusted content

  Mark a run planned, running, completed, or failed in the shared experiment workspace.

  experimentId: string, status: string, completedAt: string?

• set_verification_stateuntrusted content

  Set the verification state of a tracked manuscript object. Verified, stale, contradicted, and needs-review states require linked evidence; unlinked is only valid when no evidence is attached.

  objectId: string?, verificationState: string

• stage_research_diffuntrusted content

  Stage the proposed manuscript text for one evidence-driven Research Diff as CriticMarkup. The update remains unapplied until the researcher uses the visible accept/reject controls.

  diffId: string

• supersede_experimentuntrusted content

  Identify a newer run as replacing an earlier run and mark dependent manuscript objects stale for review.

  olderExperimentId: string, newerExperimentId: string

• update_evidenceuntrusted content

  Update structured evidence already linked to a tracked manuscript object. The evidence must belong to that object; unrelated evidence cannot be edited through this call.

  objectId: string?, evidenceId: string, type: string?, label: string?, artifactId: string?, experimentId: string?, commit: string?, metric: string?, … (list cut; the site may define more)

• update_experimentuntrusted content

  Update run metadata such as its name, method, notes, timestamps, datasets, or source commit.

  experimentId: string, name: string?, method: string?, sourceCommit: string?, notes: string?, datasets: array?, startedAt: string?, completedAt: string?, … (list cut; the site may define more)

• verify_claimuntrusted content

  Run the editor's deterministic Verify This comparison for a selected or identified tracked claim. It compares only unambiguous manuscript and evidence metric values and returns an honest review-required outcome…

  claimId: string?