{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:IHUKXL22WBXX72MJWTMJM2KWOF","short_pith_number":"pith:IHUKXL22","canonical_record":{"source":{"id":"2605.04733","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-05-06T10:32:23Z","cross_cats_sorted":[],"title_canon_sha256":"3c99ad273693806eb9b73a67741cf7b7a15317232f7f9fec50a9fa0d10d2a9fa","abstract_canon_sha256":"df083be057eb2a3ad8e2c36288e446eda919a2f68b83a0928b9a3de9a7e7338e"},"schema_version":"1.0"},"canonical_sha256":"41e8abaf5ab06f7fe989b4d8966956716a7f3b309cec894778a08b786cf033f8","source":{"kind":"arxiv","id":"2605.04733","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.04733","created_at":"2026-06-05T00:13:46Z"},{"alias_kind":"arxiv_version","alias_value":"2605.04733v2","created_at":"2026-06-05T00:13:46Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.04733","created_at":"2026-06-05T00:13:46Z"},{"alias_kind":"pith_short_12","alias_value":"IHUKXL22WBXX","created_at":"2026-06-05T00:13:46Z"},{"alias_kind":"pith_short_16","alias_value":"IHUKXL22WBXX72MJ","created_at":"2026-06-05T00:13:46Z"},{"alias_kind":"pith_short_8","alias_value":"IHUKXL22","created_at":"2026-06-05T00:13:46Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:IHUKXL22WBXX72MJWTMJM2KWOF","target":"record","payload":{"canonical_record":{"source":{"id":"2605.04733","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-05-06T10:32:23Z","cross_cats_sorted":[],"title_canon_sha256":"3c99ad273693806eb9b73a67741cf7b7a15317232f7f9fec50a9fa0d10d2a9fa","abstract_canon_sha256":"df083be057eb2a3ad8e2c36288e446eda919a2f68b83a0928b9a3de9a7e7338e"},"schema_version":"1.0"},"canonical_sha256":"41e8abaf5ab06f7fe989b4d8966956716a7f3b309cec894778a08b786cf033f8","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-05T00:13:46.809249Z","signature_b64":"7V6HUG4eCHUuHxZlsHRZ1pL4hMicTukNDt6zWaF5vNeRBgTaID7XDaOULS01cZFDMIHR6h1wMDql9ADdyLY1AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"41e8abaf5ab06f7fe989b4d8966956716a7f3b309cec894778a08b786cf033f8","last_reissued_at":"2026-06-05T00:13:46.808607Z","signature_status":"signed_v1","first_computed_at":"2026-06-05T00:13:46.808607Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.04733","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-05T00:13:46Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"UozO/V4NaKGgm8rhscTkh0GRQip1knrRTWdfWqlhH/3ja7lz76lfvkwUj0EkJD8Ih54m/DBxMZZvpvcb/BvyCA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-27T06:23:21.309164Z"},"content_sha256":"862f9533ed9b580289310c36f019a83654208f58277b6d7eedaf648c3e6c9203","schema_version":"1.0","event_id":"sha256:862f9533ed9b580289310c36f019a83654208f58277b6d7eedaf648c3e6c9203"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:IHUKXL22WBXX72MJWTMJM2KWOF","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Reward-Decomposed Reinforcement Learning for Immersive Video Role-Playing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"EBM-RL decomposes rewards to ground video role-playing in visual scenes and character traits.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Bin Li, Bo Gao, Jingtong Wu, Jun Wang, Miao Wang, Xiaodong Gu, Yaduan Ruan, Yeheng Chen, Yijiang Li, Yuling Shi, Zengxin Han","submitted_at":"2026-05-06T10:32:23Z","abstract_excerpt":"Text-based role-playing models can imitate character styles, but often fail to capture scene atmosphere and evolving tension, which are crucial for immersive applications such as VR games and interactive narratives. We study video-grounded role-playing dialogue and introduce EBM-RL (Eye--Brain--Mouth Reinforcement Learning), a decoupled GRPO-based framework that separates observation (<perception>), reasoning (<think>), and utterance generation (<answer>). This design mimics the human See-Think-Speak process, enabling the model to ground dialogue in visual perception before reasoning and respo"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Extensive experiments demonstrate that EBM-RL substantially outperforms text-only role-playing baselines and larger-scale vision-language models on our immersive role-playing benchmark, delivering simultaneous gains in visual-atmosphere consistency and character authenticity.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The four rewards (CLIP scene-text alignment, perceptual-cognitive, answer accuracy, and dense format) are assumed to collectively promote human-like sensory grounding and immersive dialogue without introducing unintended biases or overfitting to the specific benchmark and reference responses.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"EBM-RL decomposes reinforcement learning into perception-think-answer stages with CLIP alignment, perceptual-cognitive, accuracy, and format rewards to improve immersive video role-playing over text baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"EBM-RL decomposes rewards to ground video role-playing in visual scenes and character traits.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"d59a7bcd9e8f106abe62dd2793b887d42df402634e2aae90d484218860679896"},"source":{"id":"2605.04733","kind":"arxiv","version":2},"verdict":{"id":"cc320adb-0ee7-40f0-a9de-ed0dd940422b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T17:35:58.948511Z","strongest_claim":"Extensive experiments demonstrate that EBM-RL substantially outperforms text-only role-playing baselines and larger-scale vision-language models on our immersive role-playing benchmark, delivering simultaneous gains in visual-atmosphere consistency and character authenticity.","one_line_summary":"EBM-RL decomposes reinforcement learning into perception-think-answer stages with CLIP alignment, perceptual-cognitive, accuracy, and format rewards to improve immersive video role-playing over text baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The four rewards (CLIP scene-text alignment, perceptual-cognitive, answer accuracy, and dense format) are assumed to collectively promote human-like sensory grounding and immersive dialogue without introducing unintended biases or overfitting to the specific benchmark and reference responses.","pith_extraction_headline":"EBM-RL decomposes rewards to ground video role-playing in visual scenes and character traits."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.04733/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T11:35:10.612245Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T22:31:19.688425Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T14:12:45.619110Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"64b49aceb8b253d676ee449c07e36243c06f76c70557c544e7344acca7ffdd34"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"cc320adb-0ee7-40f0-a9de-ed0dd940422b"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-05T00:13:46Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"YtRZQMJSBt6nP937JaF23WPFBdqjIDqweYWVccKF6kmLUJs6WziH+nsAkbsl4WXJXJ+0Fty8TLqueKDqM3+KCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-27T06:23:21.309667Z"},"content_sha256":"97e93633bf7a1dfb7fdd3605d122ef95fe160467071458c830bf40742a0dd64a","schema_version":"1.0","event_id":"sha256:97e93633bf7a1dfb7fdd3605d122ef95fe160467071458c830bf40742a0dd64a"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/IHUKXL22WBXX72MJWTMJM2KWOF/bundle.json","state_url":"https://pith.science/pith/IHUKXL22WBXX72MJWTMJM2KWOF/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/IHUKXL22WBXX72MJWTMJM2KWOF/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-27T06:23:21Z","links":{"resolver":"https://pith.science/pith/IHUKXL22WBXX72MJWTMJM2KWOF","bundle":"https://pith.science/pith/IHUKXL22WBXX72MJWTMJM2KWOF/bundle.json","state":"https://pith.science/pith/IHUKXL22WBXX72MJWTMJM2KWOF/state.json","well_known_bundle":"https://pith.science/.well-known/pith/IHUKXL22WBXX72MJWTMJM2KWOF/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:IHUKXL22WBXX72MJWTMJM2KWOF","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"df083be057eb2a3ad8e2c36288e446eda919a2f68b83a0928b9a3de9a7e7338e","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-05-06T10:32:23Z","title_canon_sha256":"3c99ad273693806eb9b73a67741cf7b7a15317232f7f9fec50a9fa0d10d2a9fa"},"schema_version":"1.0","source":{"id":"2605.04733","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.04733","created_at":"2026-06-05T00:13:46Z"},{"alias_kind":"arxiv_version","alias_value":"2605.04733v2","created_at":"2026-06-05T00:13:46Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.04733","created_at":"2026-06-05T00:13:46Z"},{"alias_kind":"pith_short_12","alias_value":"IHUKXL22WBXX","created_at":"2026-06-05T00:13:46Z"},{"alias_kind":"pith_short_16","alias_value":"IHUKXL22WBXX72MJ","created_at":"2026-06-05T00:13:46Z"},{"alias_kind":"pith_short_8","alias_value":"IHUKXL22","created_at":"2026-06-05T00:13:46Z"}],"graph_snapshots":[{"event_id":"sha256:97e93633bf7a1dfb7fdd3605d122ef95fe160467071458c830bf40742a0dd64a","target":"graph","created_at":"2026-06-05T00:13:46Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Extensive experiments demonstrate that EBM-RL substantially outperforms text-only role-playing baselines and larger-scale vision-language models on our immersive role-playing benchmark, delivering simultaneous gains in visual-atmosphere consistency and character authenticity."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The four rewards (CLIP scene-text alignment, perceptual-cognitive, answer accuracy, and dense format) are assumed to collectively promote human-like sensory grounding and immersive dialogue without introducing unintended biases or overfitting to the specific benchmark and reference responses."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"EBM-RL decomposes reinforcement learning into perception-think-answer stages with CLIP alignment, perceptual-cognitive, accuracy, and format rewards to improve immersive video role-playing over text baselines."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"EBM-RL decomposes rewards to ground video role-playing in visual scenes and character traits."}],"snapshot_sha256":"d59a7bcd9e8f106abe62dd2793b887d42df402634e2aae90d484218860679896"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-20T11:35:10.612245Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T22:31:19.688425Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T14:12:45.619110Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.04733/integrity.json","findings":[],"snapshot_sha256":"64b49aceb8b253d676ee449c07e36243c06f76c70557c544e7344acca7ffdd34","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Text-based role-playing models can imitate character styles, but often fail to capture scene atmosphere and evolving tension, which are crucial for immersive applications such as VR games and interactive narratives. We study video-grounded role-playing dialogue and introduce EBM-RL (Eye--Brain--Mouth Reinforcement Learning), a decoupled GRPO-based framework that separates observation (<perception>), reasoning (<think>), and utterance generation (<answer>). This design mimics the human See-Think-Speak process, enabling the model to ground dialogue in visual perception before reasoning and respo","authors_text":"Bin Li, Bo Gao, Jingtong Wu, Jun Wang, Miao Wang, Xiaodong Gu, Yaduan Ruan, Yeheng Chen, Yijiang Li, Yuling Shi, Zengxin Han","cross_cats":[],"headline":"EBM-RL decomposes rewards to ground video role-playing in visual scenes and character traits.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-05-06T10:32:23Z","title":"Reward-Decomposed Reinforcement Learning for Immersive Video Role-Playing"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.04733","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-08T17:35:58.948511Z","id":"cc320adb-0ee7-40f0-a9de-ed0dd940422b","model_set":{"reader":"grok-4.3"},"one_line_summary":"EBM-RL decomposes reinforcement learning into perception-think-answer stages with CLIP alignment, perceptual-cognitive, accuracy, and format rewards to improve immersive video role-playing over text baselines.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"EBM-RL decomposes rewards to ground video role-playing in visual scenes and character traits.","strongest_claim":"Extensive experiments demonstrate that EBM-RL substantially outperforms text-only role-playing baselines and larger-scale vision-language models on our immersive role-playing benchmark, delivering simultaneous gains in visual-atmosphere consistency and character authenticity.","weakest_assumption":"The four rewards (CLIP scene-text alignment, perceptual-cognitive, answer accuracy, and dense format) are assumed to collectively promote human-like sensory grounding and immersive dialogue without introducing unintended biases or overfitting to the specific benchmark and reference responses."}},"verdict_id":"cc320adb-0ee7-40f0-a9de-ed0dd940422b"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:862f9533ed9b580289310c36f019a83654208f58277b6d7eedaf648c3e6c9203","target":"record","created_at":"2026-06-05T00:13:46Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"df083be057eb2a3ad8e2c36288e446eda919a2f68b83a0928b9a3de9a7e7338e","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-05-06T10:32:23Z","title_canon_sha256":"3c99ad273693806eb9b73a67741cf7b7a15317232f7f9fec50a9fa0d10d2a9fa"},"schema_version":"1.0","source":{"id":"2605.04733","kind":"arxiv","version":2}},"canonical_sha256":"41e8abaf5ab06f7fe989b4d8966956716a7f3b309cec894778a08b786cf033f8","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"41e8abaf5ab06f7fe989b4d8966956716a7f3b309cec894778a08b786cf033f8","first_computed_at":"2026-06-05T00:13:46.808607Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-05T00:13:46.808607Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"7V6HUG4eCHUuHxZlsHRZ1pL4hMicTukNDt6zWaF5vNeRBgTaID7XDaOULS01cZFDMIHR6h1wMDql9ADdyLY1AQ==","signature_status":"signed_v1","signed_at":"2026-06-05T00:13:46.809249Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.04733","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:862f9533ed9b580289310c36f019a83654208f58277b6d7eedaf648c3e6c9203","sha256:97e93633bf7a1dfb7fdd3605d122ef95fe160467071458c830bf40742a0dd64a"],"state_sha256":"bb6a8709037219ce08dd6c76631a0b847dfdbd199b31815627a78b5a5421028c"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"EPNV5Cvvq2SvBZXyfIZ587aN4zV2Xx25DBdvce86XCi0QjrKiRiAZ3+3W8QEJRDKe8HKmHhyMXAsM3C0nyuXCA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-27T06:23:21.312173Z","bundle_sha256":"fedeef372b7eaefafc1b7c6350ede4b890dbeb8fa0718cc6c1d533f9aaaf55ea"}}