Commit Graph

  • 68f84a59ff Kai: typed decisions under deterministic policy, and its results against public benchmarks main zeekay 2026-10-22 10:20:06 -07:00
  • dbb2012313 Grant hanzoai/ci the run's ID token [skip ci] z 2026-09-30 07:18:25 -07:00
  • aad1f306c5 Move the Enso and Kai papers to their private repos zeekay 2026-09-29 21:21:38 -07:00
  • ab5afcf18d Kai: stage a7's tables and text, and speed on ROCm beside Laya, with no Metal timings. fix/jev-compat-20260928 zeekay 2026-09-28 06:01:06 -07:00
  • b175771b53 Kai: the published PDF rebuilt from stage a6's tables and text; ci uploads artifacts with the action GitHub still runs. zeekay 2026-09-27 21:43:20 -07:00
  • ea96b16408 Kai: the speed table from stage a6 beside Laya on dbc, Metal and CPU. zeekay 2026-09-27 20:54:18 -07:00
  • aff739a295 Kai: stage a6 on the frozen harness, what it wins and loses and why, the score-head probe and its fix, the sibling rows for a5 and a6, and what a6 leaves open. zeekay 2026-09-27 20:30:01 -07:00
  • cf7d5d7fe8 Kai: the result tables from stage a6 on the frozen harness, the capability suites and speed on dgx. zeekay 2026-09-27 20:25:10 -07:00
  • 4ac2da9c22 papers.hanzo.ai publishes its static export from main: site: writes s3://hanzo-sites/hanzo/papers, the prefix its route serves. Without it CI built the image and published nothing. zeekayandHanzo Dev 2026-09-27 17:01:13 -07:00
  • 96f2140b4b papers site: kai.pdf rebuilt from main, with stage a5's tables and the programs pinned to Kai. zeekayandHanzo Dev 2026-09-27 16:54:23 -07:00
  • 1d41ff2908 Kai: the wire appendix's agent.command-risk@1 calls kai at stage a5's calibration, as the shipped program does. zeekayandHanzo Dev 2026-09-27 16:15:08 -07:00
  • 544627438d papers site: Kai links to its docs page, the public benchmarks and HIP-1332, not to Laya's weights or a private repository. zeekayandHanzo Dev 2026-09-27 15:59:06 -07:00
  • 4b15b53d35 Kai: harness and language tables from stage a5 (macro 0.851 over eleven suites, best of three in 8 of 12), which reads typed decisions' train split whole. zeekayandHanzo Dev 2026-09-27 14:08:18 -07:00
  • 8ed5a61078 Kai: the typed-decisions siblings measured apart; Laya scores alike on both halves, Kai falls on the states whose siblings its barrier removed. zeekayandHanzo Dev 2026-09-27 13:57:49 -07:00
  • a1f85455ba Kai: harness and language tables from the final stage-A checkpoint (runs/a4), the caption defines bold, dagger and the clean macro, capability and speed stay stubs until that checkpoint is timed. zeekayandHanzo Dev 2026-09-27 12:18:52 -07:00
  • 0a9310f7b6 make arxiv/<dir>/<name>.tar.gz: a paper's sources with ../shared/ flattened and its .bbl, compiled alone without bibtex before it is packed. zeekayandHanzo Dev 2026-09-27 12:08:52 -07:00
  • 8917c48fe1 Kai: the layout is a checkpoint field, the stage-A runs encode the state per question, and the speed prose claims only what they measure. zeekayandHanzo Dev 2026-09-27 10:31:30 -07:00
  • c1d6c0dff3 Kai: the speed section says which precision each runtime serves on CPU and why the Laya checkpoint timed matters. zeekayandHanzo Dev 2026-09-27 09:21:53 -07:00
  • 6df3d4401c Kai: result tables float within the evaluation section, which flushes them before the limits. zeekayandHanzo Dev 2026-09-27 09:21:29 -07:00
  • a4b63863a2 Kai: the invariance suite asks both noul swaps, and cardinality prose cites no label space the Rust port cannot reproduce. zeekayandHanzo Dev 2026-09-27 04:53:36 -07:00
  • 0ab6a3b8fc Kai: one model trained in stages, what stage A taught, and every result table written by bench tex. zeekayandHanzo Dev 2026-09-27 02:08:11 -07:00
  • f36c12640d Kai: Laya's latency at 1 to 1,000 questions, from the M1 Max run zeekayandHanzo Dev 2026-09-25 21:24:08 -07:00
  • 339819c169 Kai: native parity filled from the committed report, 21 states and 108/108 labels zeekayandHanzo Dev 2026-09-25 20:56:20 -07:00
  • f997ffaf84 Kai: capability cells filled from the committed runs, and Enso read where it writes zeekayandHanzo Dev 2026-09-25 20:55:33 -07:00
  • af5fccd8a8 Kai is one model against Laya and Jev, with a designed Enso systems evaluation. zeekayandHanzo Dev 2026-09-25 11:46:09 -07:00
  • 2325f97321 Kai: the final architecture, with every benchmark table filled from runs. zeekayandHanzo Dev 2026-09-25 11:13:42 -07:00
  • b79b939f11 We are based in San Francisco zeekayandHanzo Dev 2026-09-24 16:14:51 -07:00
  • f8e5384692 Kai: typed decision models as a control plane for agentic systems zeekayandHanzo Dev 2026-09-24 16:01:05 -07:00
  • c1372c1d14 LLM.md is the one committed guide; CLAUDE.md, AGENTS.md and GEMINI.md are local links to it zeekayandHanzo Dev 2026-09-24 15:47:35 -07:00
  • 97064b16a0 The index lists the papers the Makefile builds zeekayandHanzo Dev 2026-09-23 00:00:11 -07:00
  • 03c6289dfc A skeleton for sparse-basis indexes and part-level routing in a column store zeekayandHanzo Dev 2026-09-23 00:00:11 -07:00
  • 07694f8e88 A paper on the temporal assertion graph as it stands on main zeekayandHanzo Dev 2026-09-23 00:00:10 -07:00
  • 3647120209 The published abstract still had the numbers the paper stopped claiming zeekayandHanzo Dev 2026-09-16 14:56:30 -07:00
  • e57d01a884 The paper argued from a pooling tier the system does not have zeekayandHanzo Dev 2026-09-16 07:48:02 -07:00
  • 6c676f4d0b Build it the way this fleet builds, which is not with docker zeekayandHanzo Dev 2026-09-15 15:58:46 -07:00
  • e83a4a67b2 Ask for a machine that can build an image v59 zeekayandHanzo Dev 2026-09-15 15:49:34 -07:00
  • 5a0f823482 Say what the repair row cost and what it bought zeekay 2026-09-15 15:45:43 -07:00
  • ba960bd885 Wire the handoff deploy-site.yml says is wired zeekay 2026-09-15 15:43:39 -07:00
  • f67db9df83 Declare the one architecture this image can actually build zeekay 2026-09-15 11:50:58 -07:00
  • 1f5297718d runs-on names a machine, not a purpose zeekay 2026-09-15 11:31:50 -07:00
  • 219e7535aa Publish the memory paper, with the comparison it can now make zeekay 2026-09-15 10:52:42 -07:00
  • 34ba5b1cee hanzo-base-runtime: compare with Firebase, Supabase and Convex zeekayandHanzo Dev 2026-09-11 10:22:20 -07:00
  • f853262995 ci: re-run — the previous run was killed, not failed zeekayandHanzo Dev 2026-07-28 01:15:31 -07:00
  • bc1452e9f0 The zen claims that arithmetic settles zeekayandHanzo Dev 2026-09-10 23:54:00 -07:00
  • 5ef836794b The product papers, held to what was measured zeekayandHanzo Dev 2026-09-10 23:28:03 -07:00
  • d60d4b677e Three measured papers, and the pages that carry their numbers zeekayandHanzo Dev 2026-09-10 23:04:06 -07:00
  • 8264a97ab3 Name the builder this repo needs zeekayandHanzo Dev 2026-09-10 21:50:18 -07:00
  • d21e6abb84 Enso: the comparison table was running into the next column zeekayandHanzo Dev 2026-09-10 20:25:49 -07:00
  • 39770926be Enso: three of four best-arm cells come from the replay, and one does not zeekayandHanzo Dev 2026-09-10 19:57:34 -07:00
  • 7513388e56 Enso: two generated result files disagreed, and the paper now says so zeekayandHanzo Dev 2026-09-10 19:47:25 -07:00
  • c9d9ad5d3f Paper 1: the search over several plan sets, N-best against selection, and why 262K stops at 67 zooqueenandHanzo Dev 2026-09-10 13:23:10 -07:00
  • e099308433 LoCoMo: the empty-graph control, reported over everything it was run on zeekayandHanzo Dev 2026-09-09 14:03:08 -07:00
  • 2b594ab2c9 LoCoMo: every claim held to what the harness prints zeekayandHanzo Dev 2026-09-09 14:00:10 -07:00
  • 3c1a949bd1 A measurement of conversational memory on LoCoMo, and its negative controls zeekay 2026-09-09 12:45:36 -07:00
  • 21853c27cd Paper 1: the search ran over lexicon plans; no model at any point in those rows zooqueen 2026-09-09 11:49:21 -07:00
  • 59fa784e76 Paper 1: Pith's per-lane figures; no best-published claim on MemoryAgentBench zooqueen 2026-09-09 11:36:05 -07:00
  • b6d95aa959 Table: the current MemoryAgentBench runs only; earlier versions stay in the record zooqueen 2026-09-09 11:29:47 -07:00
  • 4183f76408 Paper 1: the typed search resolver, and answers on all four categories zooqueen 2026-09-09 11:28:59 -07:00
  • 89bf7a051f Paper 2: the first code lane is measured zooqueen 2026-09-09 05:43:01 -07:00
  • ccc3393829 Paper 1: the answers and the resolver, as measured zooqueen 2026-09-09 05:39:37 -07:00
  • 67816f32ac Paper 1 reads its retrieval tables from the runs zooqueen 2026-09-09 03:04:52 -07:00
  • 4048ba80b0 Two papers on Hanzo Context: linkage for memory, and the same engine for code zooqueen 2026-09-09 02:15:12 -07:00
  • 3544ae8fa6 Watch the path the workflows now live at zeekayandHanzo Dev 2026-09-08 19:39:11 -07:00
  • 405839ee77 Move the workflows to .github/workflows zeekayandHanzo Dev 2026-09-08 18:50:27 -07:00
  • 92818df184 ci: a runner label names the platform, not the org zeekayandHanzo Dev 2026-09-08 12:09:48 -07:00
  • 8cb34c6b7b Merge remote-tracking branch 'origin/main' zeekayandHanzo Dev 2026-09-08 09:51:44 -07:00
  • 0191067e18 fix: call the pipeline at the address it actually lives zeekayandHanzo Dev 2026-09-08 01:14:30 -07:00
  • 1d1e7be5ab Two measured papers: what an idle agent costs, and why extra hops do not help zeekayandHanzo Dev 2026-09-08 00:06:21 -07:00
  • 75a657b756 Merge branch 'paper/unified-cloud-inproc-results' zeekayandHanzo Dev 2026-09-07 21:53:09 -07:00
  • b5600f2d1a Merge branch 'land-bench-paper' zeekayandHanzo Dev 2026-09-07 21:52:54 -07:00
  • 46b274fb9b hanzod: what a node says about the compute it serves, and what that is worth zeekayandHanzo Dev 2026-09-05 23:36:34 -07:00
  • 23fbeb46b6 Attestation: what a confidential-compute tier proves, and what it inherits v58 zeekayandHanzo Dev 2026-09-04 14:22:17 -07:00
  • 5a1f3b777f Settlement: credits in, rewards out, and what admits a claim v57 zeekayandHanzo Dev 2026-09-04 14:16:51 -07:00
  • 2ded1a1fec Resolve the lockfile to the font the manifest asks for zeekayandHanzo Dev 2026-09-04 00:14:27 -07:00
  • 432b5de55b chore: support multi-arch amd64/arm64 builds for evo+spark zeekay 2026-09-03 18:51:23 -07:00
  • 911cca5cb5 Form controls and code take the site's faces 2026-08-24 13:25:09 -07:00
  • bd8ae02d19 chore: symlink AGENTS.md + GEMINI.md + CLAUDE.md -> LLM.md zeekayandHanzo Dev 2026-09-02 01:10:08 -07:00
  • 35ee89b992 every Hanzo dependency is the version npm serves zeekay 2026-09-01 16:15:45 -07:00
  • 5a1289e458 Credit the organisation on every published surface zeekay 2026-09-01 08:12:41 -07:00
  • f9db4f5737 Publish under the org, not a person zeekay 2026-09-01 02:29:53 -07:00
  • d7ee515030 State a version once zeekay 2026-08-29 13:46:58 -07:00
  • e720af4a19 the job asks for its runner the way the fleet answers zeekay 2026-08-29 12:59:54 -07:00
  • 3bcdf58195 the site's type is Zen, and it comes from us rather than Google v56 zeekay 2026-08-28 17:13:56 -07:00
  • 277439999d the corporate name is Lux Industries Inc zeekay 2026-08-28 16:05:03 -07:00
  • 55342e9ea7 One version per package, on the current release zeekay 2026-08-28 15:33:25 -07:00
  • 468724c996 Add coinage; add provenance-by-channel to the tier ladder zeekay 2026-08-26 14:14:25 -07:00
  • d10de53d99 Add mechanism and training papers; ground the series in what ships zeekay 2026-08-26 13:20:12 -07:00
  • 9b4cbda65d Add compute-market series: substrate, verification and economics zeekay 2026-08-26 13:01:37 -07:00
  • a3b2c12212 Type on the papers site is Zen, mounted from the design system mirror/main zooqueen 2026-08-24 13:25:09 -07:00
  • 3a0c53a96d site: move the type stack to Zen mirror/enso-paper-2fbd91c zeekay 2026-08-23 14:54:16 -07:00
  • f49e94bc85 say Hanzo Git, which is what it is zeekay 2026-08-19 22:28:06 -07:00
  • 0ffcd92e21 papers(poai): publish canonical A-Chain mining architecture zeekay 2026-08-18 22:53:32 -07:00
  • 36ad78b7ed Merge remote-tracking branch 'origin/main' zeekay 2026-08-18 20:39:28 -07:00
  • a051491a0a paper(ai-chain): define omnichain Proof of AI economics zeekay 2026-08-18 20:38:54 -07:00
  • 33aa65ee6e Publish papers from the shared lane only zeekay 2026-08-12 20:04:17 -07:00
  • 1680fe1783 ci: canonical pair — hanzo.yml + forge caller zeekayandHanzo Dev 2026-08-06 02:36:47 -07:00
  • de732755b8 papers: drop the poison-lineage vendor from digital-securities zeekayandHanzo Dev 2026-07-30 23:28:47 -07:00
  • 54f57136ab wip: preserve in-flight work zeekayandHanzo Dev 2026-07-29 16:33:10 -07:00
  • 1de2e6cfb4 docs: refs travel forge to GitHub, so a push to GitHub reaches no CI zooqueenandHanzo Dev 2026-07-28 17:44:20 -07:00
  • 44ec401a31 ci: a gate that can fail, and the 24 papers it catches zooqueenandHanzo Dev 2026-07-28 17:35:09 -07:00