{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:Q54ABSZ3GIQGJCWNTEOIVCHJKV","short_pith_number":"pith:Q54ABSZ3","canonical_record":{"source":{"id":"2605.13687","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T15:42:26Z","cross_cats_sorted":["cs.AI","stat.ML"],"title_canon_sha256":"aad599750dd22b88303927a7baeecd30a60ca37ee333cd4ae88f04f2c0ad292c","abstract_canon_sha256":"c5346584ee4e8c5ba1eb6df317d00ddfb89ab9d1990303db47278d79f8bc7f43"},"schema_version":"1.0"},"canonical_sha256":"877800cb3b3220648acd991c8a88e95547e9984c594131dc1b0ac231fc644a4a","source":{"kind":"arxiv","id":"2605.13687","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13687","created_at":"2026-05-18T02:44:16Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13687v1","created_at":"2026-05-18T02:44:16Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13687","created_at":"2026-05-18T02:44:16Z"},{"alias_kind":"pith_short_12","alias_value":"Q54ABSZ3GIQG","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"Q54ABSZ3GIQGJCWN","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"Q54ABSZ3","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:Q54ABSZ3GIQGJCWNTEOIVCHJKV","target":"record","payload":{"canonical_record":{"source":{"id":"2605.13687","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T15:42:26Z","cross_cats_sorted":["cs.AI","stat.ML"],"title_canon_sha256":"aad599750dd22b88303927a7baeecd30a60ca37ee333cd4ae88f04f2c0ad292c","abstract_canon_sha256":"c5346584ee4e8c5ba1eb6df317d00ddfb89ab9d1990303db47278d79f8bc7f43"},"schema_version":"1.0"},"canonical_sha256":"877800cb3b3220648acd991c8a88e95547e9984c594131dc1b0ac231fc644a4a","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T02:44:16.984808Z","signature_b64":"sUYTXRaj7t6vnT1ZHB/RHbMxBt1bavQYWNHUjfqvHirKW4+QIzUYAXZfNSCqnntLgI7kuyc3ThBFEEiHEbtjDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"877800cb3b3220648acd991c8a88e95547e9984c594131dc1b0ac231fc644a4a","last_reissued_at":"2026-05-18T02:44:16.984366Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T02:44:16.984366Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.13687","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":"2026-05-18T02:44:16Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"mg49QZZ2NhUdZfXIeXQ4UK9HF7i/gFQsrUEwttTf33RYqsZaoH8ViFGm2ZeFa67y2S64kNHDSqoc99gpVIsRAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T16:16:53.441405Z"},"content_sha256":"a3db976975dda77d7279cc754a9c402819d4ba99333c1610041e0f26bb371346","schema_version":"1.0","event_id":"sha256:a3db976975dda77d7279cc754a9c402819d4ba99333c1610041e0f26bb371346"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:Q54ABSZ3GIQGJCWNTEOIVCHJKV","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"A Hierarchical Language Model with Predictable Scaling Laws and Provable Benefits of Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Bounded-context autoregressive models require linear context to sample hierarchical languages faithfully, while reasoning models succeed with only logarithmic memory.","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Allan Sly, Elchanan Mossel, Frederic Koehler, Jason Gaitonde, Joonhyung Shin","submitted_at":"2026-05-13T15:42:26Z","abstract_excerpt":"We introduce a family of synthetic languages with hierarchical structure -- generated by a broadcast process on trees -- for which the role of context length and reasoning in autoregressive generation can be analyzed precisely. At the heart of our analytic approach is an \\emph{exact $k$-gram ansatz} in place of transformers with context length $k$, a substitution we then validate empirically. Using this ansatz we derive explicit asymptotic predictions for distributional statistics of the sequences produced by a trained model, instantiated in two settings. For the \\emph{Ising broadcast process}"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"an autoregressive reasoning model with only Θ(log n) working memory can sample exactly from the true language — an exponential improvement.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The exact k-gram ansatz serves as a faithful substitute for transformers with context length k; this substitution is central to all derivations and is only validated empirically rather than proven.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Hierarchical synthetic languages require Ω(n) context length for faithful autoregressive sampling but only Θ(log n) working memory with reasoning for exact generation from the true distribution.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Bounded-context autoregressive models require linear context to sample hierarchical languages faithfully, while reasoning models succeed with only logarithmic memory.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"4bb58c3cb7ea43609c157dfdb502a9c4d9a6b1639c683874e408542eef89c18b"},"source":{"id":"2605.13687","kind":"arxiv","version":1},"verdict":{"id":"b5229d83-7c44-4c8d-8221-f8c01d158373","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:25:53.499916Z","strongest_claim":"an autoregressive reasoning model with only Θ(log n) working memory can sample exactly from the true language — an exponential improvement.","one_line_summary":"Hierarchical synthetic languages require Ω(n) context length for faithful autoregressive sampling but only Θ(log n) working memory with reasoning for exact generation from the true distribution.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The exact k-gram ansatz serves as a faithful substitute for transformers with context length k; this substitution is central to all derivations and is only validated empirically rather than proven.","pith_extraction_headline":"Bounded-context autoregressive models require linear context to sample hierarchical languages faithfully, while reasoning models succeed with only logarithmic memory."},"references":{"count":16,"sample":[{"doi":"","year":null,"title":"In this case, with probabilityρ, the root of the subtree remains the same as the root of the full tree and we get the same quantity of the subtree one level below","work_id":"5b1a98c4-69cd-4814-9cac-25e3111a9119","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"There ared·(d−1)choices for them (where the first choice is taken twice), and given this choice, there are three permutations","work_id":"5f8d1621-e475-4d44-8a9c-712fd353d3ee","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Conditioned on the root, each retains the signal with probabilityρindependently, and given they all do, the corresponding means areM d,ρ,h−1(1) = (dρ)h−1","work_id":"ad8f4467-6aa8-48a7-bbff-36d086a66d77","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"• Ifr= (1,· · ·,1), then we recursively sample through the broadcast channelκ X∅, X(1), X(1,1),· · ·, X r","work_id":"9ff93869-403d-47f6-a245-5a501b2ee43a","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"We sample a(d, w, κ, δ Xr)-language and setY ′ to be the sampled leaves","work_id":"5f65ae95-4dd2-4661-8b74-8318469ed252","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":16,"snapshot_sha256":"140c26597a2c6f5a028606965dc244751845280b0d23bb567ecc340750eb1017","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":"b5229d83-7c44-4c8d-8221-f8c01d158373"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T02:44:16Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"kWwifgKTiYXBl64bEeEmyklfgFnK0rBYz1xM8C5wgtImVMsYUk9B6qu7D8iDT7XN9ahiFfQ4QlBrIKbvan5PDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T16:16:53.442525Z"},"content_sha256":"f00f72e354853db981aa1fcfe2603f29ee1a850a808efd67666819e3b0796fb6","schema_version":"1.0","event_id":"sha256:f00f72e354853db981aa1fcfe2603f29ee1a850a808efd67666819e3b0796fb6"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/Q54ABSZ3GIQGJCWNTEOIVCHJKV/bundle.json","state_url":"https://pith.science/pith/Q54ABSZ3GIQGJCWNTEOIVCHJKV/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/Q54ABSZ3GIQGJCWNTEOIVCHJKV/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":"2026-05-28T16:16:53Z","links":{"resolver":"https://pith.science/pith/Q54ABSZ3GIQGJCWNTEOIVCHJKV","bundle":"https://pith.science/pith/Q54ABSZ3GIQGJCWNTEOIVCHJKV/bundle.json","state":"https://pith.science/pith/Q54ABSZ3GIQGJCWNTEOIVCHJKV/state.json","well_known_bundle":"https://pith.science/.well-known/pith/Q54ABSZ3GIQGJCWNTEOIVCHJKV/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:Q54ABSZ3GIQGJCWNTEOIVCHJKV","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":"c5346584ee4e8c5ba1eb6df317d00ddfb89ab9d1990303db47278d79f8bc7f43","cross_cats_sorted":["cs.AI","stat.ML"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T15:42:26Z","title_canon_sha256":"aad599750dd22b88303927a7baeecd30a60ca37ee333cd4ae88f04f2c0ad292c"},"schema_version":"1.0","source":{"id":"2605.13687","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13687","created_at":"2026-05-18T02:44:16Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13687v1","created_at":"2026-05-18T02:44:16Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13687","created_at":"2026-05-18T02:44:16Z"},{"alias_kind":"pith_short_12","alias_value":"Q54ABSZ3GIQG","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"Q54ABSZ3GIQGJCWN","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"Q54ABSZ3","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:f00f72e354853db981aa1fcfe2603f29ee1a850a808efd67666819e3b0796fb6","target":"graph","created_at":"2026-05-18T02:44:16Z","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":"an autoregressive reasoning model with only Θ(log n) working memory can sample exactly from the true language — an exponential improvement."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The exact k-gram ansatz serves as a faithful substitute for transformers with context length k; this substitution is central to all derivations and is only validated empirically rather than proven."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Hierarchical synthetic languages require Ω(n) context length for faithful autoregressive sampling but only Θ(log n) working memory with reasoning for exact generation from the true distribution."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Bounded-context autoregressive models require linear context to sample hierarchical languages faithfully, while reasoning models succeed with only logarithmic memory."}],"snapshot_sha256":"4bb58c3cb7ea43609c157dfdb502a9c4d9a6b1639c683874e408542eef89c18b"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"We introduce a family of synthetic languages with hierarchical structure -- generated by a broadcast process on trees -- for which the role of context length and reasoning in autoregressive generation can be analyzed precisely. At the heart of our analytic approach is an \\emph{exact $k$-gram ansatz} in place of transformers with context length $k$, a substitution we then validate empirically. Using this ansatz we derive explicit asymptotic predictions for distributional statistics of the sequences produced by a trained model, instantiated in two settings. For the \\emph{Ising broadcast process}","authors_text":"Allan Sly, Elchanan Mossel, Frederic Koehler, Jason Gaitonde, Joonhyung Shin","cross_cats":["cs.AI","stat.ML"],"headline":"Bounded-context autoregressive models require linear context to sample hierarchical languages faithfully, while reasoning models succeed with only logarithmic memory.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T15:42:26Z","title":"A Hierarchical Language Model with Predictable Scaling Laws and Provable Benefits of Reasoning"},"references":{"count":16,"internal_anchors":0,"resolved_work":16,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"In this case, with probabilityρ, the root of the subtree remains the same as the root of the full tree and we get the same quantity of the subtree one level below","work_id":"5b1a98c4-69cd-4814-9cac-25e3111a9119","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"There ared·(d−1)choices for them (where the first choice is taken twice), and given this choice, there are three permutations","work_id":"5f8d1621-e475-4d44-8a9c-712fd353d3ee","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Conditioned on the root, each retains the signal with probabilityρindependently, and given they all do, the corresponding means areM d,ρ,h−1(1) = (dρ)h−1","work_id":"ad8f4467-6aa8-48a7-bbff-36d086a66d77","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"• Ifr= (1,· · ·,1), then we recursively sample through the broadcast channelκ X∅, X(1), X(1,1),· · ·, X r","work_id":"9ff93869-403d-47f6-a245-5a501b2ee43a","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"We sample a(d, w, κ, δ Xr)-language and setY ′ to be the sampled leaves","work_id":"5f65ae95-4dd2-4661-8b74-8318469ed252","year":null}],"snapshot_sha256":"140c26597a2c6f5a028606965dc244751845280b0d23bb567ecc340750eb1017"},"source":{"id":"2605.13687","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T20:25:53.499916Z","id":"b5229d83-7c44-4c8d-8221-f8c01d158373","model_set":{"reader":"grok-4.3"},"one_line_summary":"Hierarchical synthetic languages require Ω(n) context length for faithful autoregressive sampling but only Θ(log n) working memory with reasoning for exact generation from the true distribution.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Bounded-context autoregressive models require linear context to sample hierarchical languages faithfully, while reasoning models succeed with only logarithmic memory.","strongest_claim":"an autoregressive reasoning model with only Θ(log n) working memory can sample exactly from the true language — an exponential improvement.","weakest_assumption":"The exact k-gram ansatz serves as a faithful substitute for transformers with context length k; this substitution is central to all derivations and is only validated empirically rather than proven."}},"verdict_id":"b5229d83-7c44-4c8d-8221-f8c01d158373"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:a3db976975dda77d7279cc754a9c402819d4ba99333c1610041e0f26bb371346","target":"record","created_at":"2026-05-18T02:44:16Z","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":"c5346584ee4e8c5ba1eb6df317d00ddfb89ab9d1990303db47278d79f8bc7f43","cross_cats_sorted":["cs.AI","stat.ML"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T15:42:26Z","title_canon_sha256":"aad599750dd22b88303927a7baeecd30a60ca37ee333cd4ae88f04f2c0ad292c"},"schema_version":"1.0","source":{"id":"2605.13687","kind":"arxiv","version":1}},"canonical_sha256":"877800cb3b3220648acd991c8a88e95547e9984c594131dc1b0ac231fc644a4a","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"877800cb3b3220648acd991c8a88e95547e9984c594131dc1b0ac231fc644a4a","first_computed_at":"2026-05-18T02:44:16.984366Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T02:44:16.984366Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"sUYTXRaj7t6vnT1ZHB/RHbMxBt1bavQYWNHUjfqvHirKW4+QIzUYAXZfNSCqnntLgI7kuyc3ThBFEEiHEbtjDQ==","signature_status":"signed_v1","signed_at":"2026-05-18T02:44:16.984808Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13687","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:a3db976975dda77d7279cc754a9c402819d4ba99333c1610041e0f26bb371346","sha256:f00f72e354853db981aa1fcfe2603f29ee1a850a808efd67666819e3b0796fb6"],"state_sha256":"dd3c1f42d4058c595c24c641dbff2bc31127eb3831987b847116219334541183"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"YMrYAWIhziQ94c0gyYX0TdkCv0gY6u8Y6wdynkpk9OZUsDtBBdfgXZVtVso5kWX2MagN7cC2dXtfA+zw4ELkBA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-28T16:16:53.447285Z","bundle_sha256":"b276b4abdd1e1af7664ed5132c71234b752424f9063749019c1cd56b1b1d5c00"}}