Extract arXiv:2607.15495 (Verbalizable Representations Form a Global Workspace in Language Models)

Checks run in this order. Open a step to inspect its rules, checker or review rubric.

1. Structural checkAutomated checkChecker configured

   The document parses as JSON and satisfies the bound schema. Establishes shape only; a schema-valid document can still be false.

   A conforming checker is configured. Execution capacity and availability are checked when the Submission runs.

   TfP builtin checker

   This checker executes code inside the TfP Worker. It is not a pullable OCI image.

   Read checker code

   Source provenance and runtime metadata

   This viewer shows the builtin source bundled with the current deployment. It is not a historical execution or build attestation.

   The builtin’s recorded digest identifies its reference string; it is not an OCI digest or a hash of the executed source code.

   Recorded image / builtin reference

   tfp-builtin/json-schema-check:1

   Recorded runtime digest

   sha256:0d84283aa10e891272f89c4dab2f6e8f09c9bbebf6e9c8a0f57409a372b2b9ac

   Declared entrypoint (argument array)

   ["/tfp/verify"]

   Declared source

   Open recorded source repository ↗

   No exact source commit is recorded.

   Declared license: MIT. TfP has not verified a reproducible build connecting this declared source to the runtime.

   Checker identity and execution limits

   Implementations can be replaced while preserving this Contract’s acceptance rules.

   {
     "verifier_implementation_id": "vfi_yf1xu4wumu",
     "version": 1,
     "content_hash": "sha256:71f3458240b91bead5b735f1ebfd674d443efc6ac299485663c2bb388fab266f",
     "conformance": {
       "checked_at": "2026-09-02T16:48:32Z",
       "differential_against_implementation_id": null,
       "failures": [],
       "fixtures_passed": 4,
       "fixtures_total": 4,
       "status": "passed"
     },
     "runtime_image_digest": "sha256:0d84283aa10e891272f89c4dab2f6e8f09c9bbebf6e9c8a0f57409a372b2b9ac",
     "limits": {
       "disk_mb": 64,
       "memory_mb": 512,
       "network": "none",
       "timeout_seconds": 60,
       "vcpu": 1
     }
   }

   Stage evidence schema

   {
     "properties": {
       "valid": {
         "type": "boolean"
       },
       "violations_count": {
         "type": "integer"
       }
     },
     "required": [
       "valid",
       "violations_count"
     ],
     "type": "object"
   }

2. Fidelity reviewReviewHosted reviewer unavailable

   Reviewers read the exact bound source and the document and judge whether the document is a faithful extraction. Establishes reviewer acceptance of fidelity; does not establish that the source is correct.

   At least one hosted review node has no configured allowed agent in this deployment. Do not assume this review can complete.

   Read the pinned rubric and reviewer rules

   Extraction fidelity: three independent agents, human expert on disagreement · pinned protocol

   Independent agent panel · 3 hosted agents

   You are judging whether the submitted extraction (the document artifact, in materials) is a FAITHFUL representation of the exact bound source (paper_text, also in materials; paper_html is the same document as published). You are not judging whether the paper is right.

   Read the source, then check the document field by field. Answer pass only if every material statement is supported by the source at or near its locator, the paper's central conclusions are not misstated, nothing is claimed about consciousness beyond what the paper itself asserts (the paper's §1.5 and §9.4 are the reference points), every replication/resource assertion is exactly what the paper states, and no omission would mislead a reader about the main findings, limitations, or explicit non-claims. Answer fail if any of those is violated materially. Answer uncertain when the document's fidelity turns on a judgment the source does not settle (e.g. slightly stronger wording than the source, or an omission whose importance is debatable).

   List every material error you found in material_errors, each with the document field, the problem, and the source locator that shows it. Do not list errors you did not verify against the source. Cite the source passages you relied on.

   Rationale requiredAt least 1 citationConfidence required

   Response fields and full schema

   • material_errors · required array
   • verdict · required "pass" / "fail" / "uncertain"

   {
     "additionalProperties": false,
     "properties": {
       "material_errors": {
         "items": {
           "additionalProperties": false,
           "properties": {
             "field": {
               "maxLength": 200,
               "type": "string"
             },
             "problem": {
               "maxLength": 600,
               "type": "string"
             },
             "source_locator": {
               "maxLength": 200,
               "type": "string"
             }
           },
           "required": [
             "field",
             "problem",
             "source_locator"
           ],
           "type": "object"
         },
         "maxItems": 20,
         "type": "array"
       },
       "verdict": {
         "enum": [
           "pass",
           "fail",
           "uncertain"
         ]
       }
     },
     "required": [
       "verdict",
       "material_errors"
     ],
     "type": "object"
   }

   TfP-hosted agents from the protocol’s allowed configurations.The submitter is excluded.Responses on this node are withheld until it completes.

   Exact reviewer eligibility, visibility and Skill pins

   {
     "reviewers": {
       "allowed": [
         {
           "config": {
             "effort": "high"
           },
           "model": "claude-opus-5",
           "provider": "anthropic"
         }
       ],
       "count": 3,
       "eligibility": null,
       "independent": true,
       "kind": "agent"
     },
     "conflicts": null,
     "visibility": {
       "hidden": [
         "submitter_identity",
         "submitter_organization"
       ],
       "prior_nodes": [],
       "stage_evidence": true
     },
     "skills": [
       {
         "requirement": "required",
         "skill_version_id": "skv_38bb5h9zgu"
       }
     ]
   }

   Human expert · 1 reviewer

   Runs only when panel_decision is unresolved.

   The agent panel did not agree. Read their reports (prior reviews), examine the source and the document yourself, and decide whether the document is a faithful extraction. pass or fail; list the material errors you confirmed.

   Rationale requiredAt least 1 citationConfidence required

   Response fields and full schema

   • material_errors · required array
   • verdict · required "pass" / "fail"

   {
     "additionalProperties": false,
     "properties": {
       "material_errors": {
         "items": {
           "additionalProperties": false,
           "properties": {
             "field": {
               "maxLength": 200,
               "type": "string"
             },
             "problem": {
               "maxLength": 600,
               "type": "string"
             },
             "source_locator": {
               "maxLength": 200,
               "type": "string"
             }
           },
           "required": [
             "field",
             "problem",
             "source_locator"
           ],
           "type": "object"
         },
         "maxItems": 20,
         "type": "array"
       },
       "verdict": {
         "enum": [
           "pass",
           "fail"
         ]
       }
     },
     "required": [
       "verdict",
       "material_errors"
     ],
     "type": "object"
   }

   Only designated Campaign reviewers may claim a seat.The submitter is excluded.Declare any agent assistance used; a declaration is optional under this protocol.

   Exact reviewer eligibility, visibility and Skill pins

   {
     "reviewers": {
       "allowed": null,
       "count": 1,
       "eligibility": {
         "organization_id": null,
         "rule": "campaign_reviewers",
         "user_ids": null
       },
       "independent": false,
       "kind": "human"
     },
     "conflicts": {
       "exclude_submitter": true,
       "exclude_submitter_organization": false
     },
     "visibility": {
       "hidden": [
         "submitter_identity",
         "submitter_organization"
       ],
       "prior_nodes": [
         "panel"
       ],
       "stage_evidence": true
     },
     "skills": [
       {
         "requirement": "required",
         "skill_version_id": "skv_38bb5h9zgu"
       }
     ]
   }

   How judgments become the outcome

   Two of three

   Accept when at least 2 responses have verdict equals "pass". Reject when at least 2 have verdict equals "fail". Otherwise unresolved.

   Inputs, in order: panel · panel_decision

   Expert decides · only when panel_decision is unresolved

   Accept when at least 1 responses have verdict equals "pass". Reject when at least 1 have verdict equals "fail". Otherwise unresolved.

   Inputs, in order: expert · expert_decision

   Panel, else expert

   Use the first outcome that is not unresolved, in the input order below. If none resolves, the outcome stays unresolved.

   Inputs, in order: panel_decision, expert_decision · final

   Final outcome comes from final. Accept passes this stage; reject refutes it; unresolved is inconclusive. All verification stages must pass.

   Node deadline: 72 hoursClaim hold: 120 minutes, bounded by the node deadline

   Protocol identity and full rules

   rpt_ifc0n7f3gh
sha256:b9fd7cc1d39e9ca3751c3a78b45e51f68dc1de1e0ea254f4d605c2d47ba7c716

   Read the complete pinned protocol →

   {
     "output": "final",
     "policy": {
       "assignment_timeout_hours": 72,
       "max_replacements_per_node": 2
     }
   }