{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:GD7XVMZUXZJUKCX5TLTQ7JGSIH","short_pith_number":"pith:GD7XVMZU","canonical_record":{"source":{"id":"2605.13570","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-13T14:07:10Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"5d5d41614467562af644e939666e1efeb26e05eb63c4c938de155d8448196d04","abstract_canon_sha256":"de20f2527edf952771ca2ecd8823de3ff1adfcb0ef1c9905fc186554e3d6a570"},"schema_version":"1.0"},"canonical_sha256":"30ff7ab334be53450afd9ae70fa4d241d9b214c512bd3a054ec4a2dbd0ea6093","source":{"kind":"arxiv","id":"2605.13570","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13570","created_at":"2026-05-18T02:44:23Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13570v1","created_at":"2026-05-18T02:44:23Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13570","created_at":"2026-05-18T02:44:23Z"},{"alias_kind":"pith_short_12","alias_value":"GD7XVMZUXZJU","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"GD7XVMZUXZJUKCX5","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"GD7XVMZU","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:GD7XVMZUXZJUKCX5TLTQ7JGSIH","target":"record","payload":{"canonical_record":{"source":{"id":"2605.13570","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-13T14:07:10Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"5d5d41614467562af644e939666e1efeb26e05eb63c4c938de155d8448196d04","abstract_canon_sha256":"de20f2527edf952771ca2ecd8823de3ff1adfcb0ef1c9905fc186554e3d6a570"},"schema_version":"1.0"},"canonical_sha256":"30ff7ab334be53450afd9ae70fa4d241d9b214c512bd3a054ec4a2dbd0ea6093","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T02:44:23.380138Z","signature_b64":"Nw67VwTh5EvaMkVuEpNaeZAd3LIf+0oCq23v/AWDkBz8JFM2t7Dp36Mhe44zxo9ASIhx3ecU9iGZbOif+wevCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"30ff7ab334be53450afd9ae70fa4d241d9b214c512bd3a054ec4a2dbd0ea6093","last_reissued_at":"2026-05-18T02:44:23.379528Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T02:44:23.379528Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.13570","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T02:44:23Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"tI+msgoh38MHxAKbFP3Nys7hlMkZflHdQdes8bNdfGylP6/bPFFrHGRTtB1ayxfd7fVedVxt7i+I5SalBPT0Cg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T18:58:51.212037Z"},"content_sha256":"a2657a7e0a6adef32497af82501b968b08da72ec12b24b635f49f2be8efc79b1","schema_version":"1.0","event_id":"sha256:a2657a7e0a6adef32497af82501b968b08da72ec12b24b635f49f2be8efc79b1"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:GD7XVMZUXZJUKCX5TLTQ7JGSIH","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Learning Local Constraints for Reinforcement-Learned Content Generators","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Constraining an RL content generator with local patterns learned by Wave Function Collapse produces playable and visually satisfying puzzle-platform levels.","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Ahmed Khalifa, Debosmita Bhaumik, Georgios N. Yannakakis, Julian Togelius","submitted_at":"2026-05-13T14:07:10Z","abstract_excerpt":"Constraint-based game content generators that learn local constraints from existing content, such as Wave Function Collapse (WFC), can generate visually satisfying game levels but face challenges in guaranteeing global properties, such as playability. On the other hand, reinforcement-learning trained generators can guarantee global properties -- because such properties can easily be included in reward functions -- but the results can be visually dissatisfying. In this paper, we explore ways to combine these methods. Specifically, we constrain the action space of a PCGRL generator with constrai"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"the best of our trained generators produce visually satisfying and playable puzzle-platform game levels -- such as Lode Runner levels -- with desired global properties.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That WFC-learned local constraints remain compatible with the RL reward signal and do not overly restrict the policy to the point where global properties cannot be achieved during training.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Constraining a PCGRL generator's action space with locally learned WFC constraints yields visually satisfying and playable puzzle-platform levels with desired global properties.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Constraining an RL content generator with local patterns learned by Wave Function Collapse produces playable and visually satisfying puzzle-platform levels.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"83b20bf2f998f43a6f88a3cad28b749a3803a822210d55050c319da01c9ebcab"},"source":{"id":"2605.13570","kind":"arxiv","version":1},"verdict":{"id":"b755faf1-ef75-4c01-8117-3ded1de98dfe","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T18:32:41.490871Z","strongest_claim":"the best of our trained generators produce visually satisfying and playable puzzle-platform game levels -- such as Lode Runner levels -- with desired global properties.","one_line_summary":"Constraining a PCGRL generator's action space with locally learned WFC constraints yields visually satisfying and playable puzzle-platform levels with desired global properties.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That WFC-learned local constraints remain compatible with the RL reward signal and do not overly restrict the policy to the point where global properties cannot be achieved during training.","pith_extraction_headline":"Constraining an RL content generator with local patterns learned by Wave Function Collapse produces playable and visually satisfying puzzle-platform levels."},"references":{"count":40,"sample":[{"doi":"","year":2023,"title":"Hierarchical semantic wave function collapse","work_id":"f0808e2f-6f53-41de-b534-226b052cfa32","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Leveraging reinforcement learn- ing and wavefunctioncollapse for improved procedural level generation","work_id":"22ac18fb-0b37-43bb-996f-57e3e5dddd59","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and Wojciech Zaremba. Openai gym, 2016","work_id":"a9e00073-a37b-4b37-aee1-856d97eeeb8f","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Tile-based map generation using wave function col- lapse in ’caves of qud’, 2022","work_id":"464754ec-94be-42ac-92a3-0f4914c3039b","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Sturgeon-graph: Constrained graph generation from examples","work_id":"4f7c05bc-6207-49eb-9ff3-0dcafa14b5a9","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":40,"snapshot_sha256":"d85ad1f1102662e1f632985b3aeea395088acc16677cf018b37d135e60088e7d","internal_anchors":2},"formal_canon":{"evidence_count":2,"snapshot_sha256":"41aa4fdad3f6428b76700b4a07dde9661225471504e229a468dc09d027625df8"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"b755faf1-ef75-4c01-8117-3ded1de98dfe"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T02:44:23Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"kVOKN2lE2q12obl1kiIQypHQGCMDxODrB0jbi2QIXWwTVlH/6gEONWKNWSFSrrkZ/HE5Blpsdg9LSWFmk8UyBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T18:58:51.212684Z"},"content_sha256":"73304396d05f2a53cddbb87d50e5a1ffc0880948fa4eba545e5d4c79d721da03","schema_version":"1.0","event_id":"sha256:73304396d05f2a53cddbb87d50e5a1ffc0880948fa4eba545e5d4c79d721da03"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/GD7XVMZUXZJUKCX5TLTQ7JGSIH/bundle.json","state_url":"https://pith.science/pith/GD7XVMZUXZJUKCX5TLTQ7JGSIH/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/GD7XVMZUXZJUKCX5TLTQ7JGSIH/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":"2026-05-21T18:58:51Z","links":{"resolver":"https://pith.science/pith/GD7XVMZUXZJUKCX5TLTQ7JGSIH","bundle":"https://pith.science/pith/GD7XVMZUXZJUKCX5TLTQ7JGSIH/bundle.json","state":"https://pith.science/pith/GD7XVMZUXZJUKCX5TLTQ7JGSIH/state.json","well_known_bundle":"https://pith.science/.well-known/pith/GD7XVMZUXZJUKCX5TLTQ7JGSIH/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:GD7XVMZUXZJUKCX5TLTQ7JGSIH","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":"de20f2527edf952771ca2ecd8823de3ff1adfcb0ef1c9905fc186554e3d6a570","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-13T14:07:10Z","title_canon_sha256":"5d5d41614467562af644e939666e1efeb26e05eb63c4c938de155d8448196d04"},"schema_version":"1.0","source":{"id":"2605.13570","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13570","created_at":"2026-05-18T02:44:23Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13570v1","created_at":"2026-05-18T02:44:23Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13570","created_at":"2026-05-18T02:44:23Z"},{"alias_kind":"pith_short_12","alias_value":"GD7XVMZUXZJU","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"GD7XVMZUXZJUKCX5","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"GD7XVMZU","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:73304396d05f2a53cddbb87d50e5a1ffc0880948fa4eba545e5d4c79d721da03","target":"graph","created_at":"2026-05-18T02:44:23Z","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":"the best of our trained generators produce visually satisfying and playable puzzle-platform game levels -- such as Lode Runner levels -- with desired global properties."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That WFC-learned local constraints remain compatible with the RL reward signal and do not overly restrict the policy to the point where global properties cannot be achieved during training."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Constraining a PCGRL generator's action space with locally learned WFC constraints yields visually satisfying and playable puzzle-platform levels with desired global properties."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Constraining an RL content generator with local patterns learned by Wave Function Collapse produces playable and visually satisfying puzzle-platform levels."}],"snapshot_sha256":"83b20bf2f998f43a6f88a3cad28b749a3803a822210d55050c319da01c9ebcab"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"41aa4fdad3f6428b76700b4a07dde9661225471504e229a468dc09d027625df8"},"paper":{"abstract_excerpt":"Constraint-based game content generators that learn local constraints from existing content, such as Wave Function Collapse (WFC), can generate visually satisfying game levels but face challenges in guaranteeing global properties, such as playability. On the other hand, reinforcement-learning trained generators can guarantee global properties -- because such properties can easily be included in reward functions -- but the results can be visually dissatisfying. In this paper, we explore ways to combine these methods. Specifically, we constrain the action space of a PCGRL generator with constrai","authors_text":"Ahmed Khalifa, Debosmita Bhaumik, Georgios N. Yannakakis, Julian Togelius","cross_cats":["cs.LG"],"headline":"Constraining an RL content generator with local patterns learned by Wave Function Collapse produces playable and visually satisfying puzzle-platform levels.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-13T14:07:10Z","title":"Learning Local Constraints for Reinforcement-Learned Content Generators"},"references":{"count":40,"internal_anchors":2,"resolved_work":40,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Hierarchical semantic wave function collapse","work_id":"f0808e2f-6f53-41de-b534-226b052cfa32","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Leveraging reinforcement learn- ing and wavefunctioncollapse for improved procedural level generation","work_id":"22ac18fb-0b37-43bb-996f-57e3e5dddd59","year":2021},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and Wojciech Zaremba. Openai gym, 2016","work_id":"a9e00073-a37b-4b37-aee1-856d97eeeb8f","year":2016},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Tile-based map generation using wave function col- lapse in ’caves of qud’, 2022","work_id":"464754ec-94be-42ac-92a3-0f4914c3039b","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Sturgeon-graph: Constrained graph generation from examples","work_id":"4f7c05bc-6207-49eb-9ff3-0dcafa14b5a9","year":2023}],"snapshot_sha256":"d85ad1f1102662e1f632985b3aeea395088acc16677cf018b37d135e60088e7d"},"source":{"id":"2605.13570","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T18:32:41.490871Z","id":"b755faf1-ef75-4c01-8117-3ded1de98dfe","model_set":{"reader":"grok-4.3"},"one_line_summary":"Constraining a PCGRL generator's action space with locally learned WFC constraints yields visually satisfying and playable puzzle-platform levels with desired global properties.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Constraining an RL content generator with local patterns learned by Wave Function Collapse produces playable and visually satisfying puzzle-platform levels.","strongest_claim":"the best of our trained generators produce visually satisfying and playable puzzle-platform game levels -- such as Lode Runner levels -- with desired global properties.","weakest_assumption":"That WFC-learned local constraints remain compatible with the RL reward signal and do not overly restrict the policy to the point where global properties cannot be achieved during training."}},"verdict_id":"b755faf1-ef75-4c01-8117-3ded1de98dfe"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:a2657a7e0a6adef32497af82501b968b08da72ec12b24b635f49f2be8efc79b1","target":"record","created_at":"2026-05-18T02:44:23Z","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":"de20f2527edf952771ca2ecd8823de3ff1adfcb0ef1c9905fc186554e3d6a570","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-13T14:07:10Z","title_canon_sha256":"5d5d41614467562af644e939666e1efeb26e05eb63c4c938de155d8448196d04"},"schema_version":"1.0","source":{"id":"2605.13570","kind":"arxiv","version":1}},"canonical_sha256":"30ff7ab334be53450afd9ae70fa4d241d9b214c512bd3a054ec4a2dbd0ea6093","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"30ff7ab334be53450afd9ae70fa4d241d9b214c512bd3a054ec4a2dbd0ea6093","first_computed_at":"2026-05-18T02:44:23.379528Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T02:44:23.379528Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"Nw67VwTh5EvaMkVuEpNaeZAd3LIf+0oCq23v/AWDkBz8JFM2t7Dp36Mhe44zxo9ASIhx3ecU9iGZbOif+wevCA==","signature_status":"signed_v1","signed_at":"2026-05-18T02:44:23.380138Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13570","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:a2657a7e0a6adef32497af82501b968b08da72ec12b24b635f49f2be8efc79b1","sha256:73304396d05f2a53cddbb87d50e5a1ffc0880948fa4eba545e5d4c79d721da03"],"state_sha256":"b3db9cc4e109162b9f15bf886e250dc99bb74eb41b38a725b073327a4c84bc43"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"MNzTT4wlt+5EKxmeXAZxiDcPbfRxWjmlVW7+rBoHO9mnKL4WGdWfLLaHkORscL+79NKi4187s6QPb6OAgqymDw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-21T18:58:51.215407Z","bundle_sha256":"d5f274ea168010b73cd992102ee9311a2da3db309b6d3eb55ad8f0536690e6f0"}}