Commit graph

  • 47feeb32ee Linting, doc edits Lawrence Akka 2026-08-18 12:04:25 +0200
  • a5c4647005
    Merge pull request #550 from ggozad/feat/scalar-index-migration Yiorgis Gozadinos 2026-08-18 12:48:25 +0300
  • 11644c7f43
    Trim comments, docstrings and docs to what they need to say Yiorgis Gozadinos 2026-08-18 12:38:02 +0300
  • 3a9ded2218 Display all chunk metadata in inspector Lawrence Akka 2026-08-17 20:17:17 +0200
  • 09ff101349 Expose all chunk metadata on search results and citations through SearchResult.chunk_meta and Citation.chunk_meta Lawrence Akka 2026-08-15 15:21:09 +0200
  • 8e93b639bc
    Migrate existing databases to the full index set Yiorgis Gozadinos 2026-08-17 15:19:25 +0300
  • c184a25d68
    Index every hot lookup key from one shared definition Yiorgis Gozadinos 2026-08-17 14:54:01 +0300
  • 980b1985ab
    Merge pull request #545 from ggozad/feat/mtrag-bench Yiorgis Gozadinos 2026-08-17 11:28:49 +0300
  • 73dfc62f33
    Exclude fully unjudged conversations from the macro pass rate Yiorgis Gozadinos 2026-08-17 11:19:59 +0300
  • 587ba75a62
    Thread the document filter through the live QA runner Yiorgis Gozadinos 2026-08-17 11:03:52 +0300
  • 3b9d6ae2c6
    Publish the paired statistics behind the MTRAG compaction claims Yiorgis Gozadinos 2026-08-14 17:05:45 +0300
  • 0634964e64
    Point the mtrag reference config at the measured baseline model Yiorgis Gozadinos 2026-08-14 15:26:07 +0300
  • bf94efc655
    Simplify the eval harness and share the embed-fill path. Yiorgis Gozadinos 2026-08-14 15:03:41 +0300
  • f1d5918d43
    Update MTRAG benchmark numbers to the glimmer baseline Yiorgis Gozadinos 2026-08-14 15:11:41 +0300
  • db2b8fb883
    Add compaction arms and grounding status to MTRAG live runs Yiorgis Gozadinos 2026-08-13 22:52:36 +0300
  • 73d9d93db9
    Add MTRAG ClapNQ multi-turn evaluation Yiorgis Gozadinos 2026-08-06 16:42:54 +0300
  • cc04f92f28
    Batch embeddings across import_documents batches Yiorgis Gozadinos 2026-08-06 16:42:17 +0300
  • b07426a883
    Merge pull request #543 from Cwiesen/feat-evaluations-search-filter Yiorgis Gozadinos 2026-08-17 10:45:43 +0300
  • a75d89122e
    Drop the unused per-dataset filter default Yiorgis Gozadinos 2026-08-17 10:29:28 +0300
  • 9b2ae347d2
    Use filters that match the datasets they document Yiorgis Gozadinos 2026-08-17 10:19:51 +0300
  • 90e53b7348
    Trim the Muse-Glimmer benchmark footnote to the measured configuration Yiorgis Gozadinos 2026-08-16 14:19:53 +0300
  • 92cab7c846
    Record the Muse-Glimmer ORB multimodal numbers Yiorgis Gozadinos 2026-08-16 14:16:40 +0300
  • db6996b629 fix: revert changes to benchmarks.md file cwiesen 2026-08-14 16:36:51 -0500
  • a68cb23b6e fix: resolve incorrect dataset and column namings cwiesen 2026-08-14 16:27:04 -0500
  • 93c21272d1 feat: add search_filter to evaluations cwiesen 2026-08-13 17:05:04 -0500
  • 981353eb98
    Merge pull request #544 from ggozad/chore/remove-wix Yiorgis Gozadinos 2026-08-14 14:58:17 +0300
  • e522d8cfa8
    Remove the wix evaluation dataset Yiorgis Gozadinos 2026-08-14 14:46:51 +0300
  • 38446f9618
    Time the search pathway inside an agent run bryan davis 2026-08-13 14:33:06 -0500
  • d864cf8008
    vb 0.74.0 Yiorgis Gozadinos 2026-08-13 16:16:35 +0300
  • c5ecf718d9
    Merge pull request #541 from ggozad/fix/new-question-detection Yiorgis Gozadinos 2026-08-13 16:09:27 +0300
  • d1e3b828f1
    Judge each capability's evidence by its own record Yiorgis Gozadinos 2026-08-13 15:57:33 +0300
  • e1dd8517f9
    Refuse to compact evidence the host kept no record of Yiorgis Gozadinos 2026-08-13 15:46:23 +0300
  • 53bbf52697
    Round-trip the capability's whole state through the browser Yiorgis Gozadinos 2026-08-13 15:20:02 +0300
  • 771ac9c96c
    Register the optional capabilities where agents are composed Yiorgis Gozadinos 2026-08-13 15:04:05 +0300
  • 21e261f608
    Stop a partly resolved citation asking to be retried Yiorgis Gozadinos 2026-08-13 15:03:51 +0300
  • 3ec72363bc
    Let the record say whether a question is still being answered Yiorgis Gozadinos 2026-08-13 15:03:24 +0300
  • 7bdd11db39
    Update caps documentation Yiorgis Gozadinos 2026-08-13 14:08:09 +0300
  • 0a0ddbd3b6
    Merge pull request #538 from ggozad/feat/citation-policy Yiorgis Gozadinos 2026-08-13 13:55:19 +0300
  • cee0824744
    Say precisely which conversations go unenforced Yiorgis Gozadinos 2026-08-12 12:04:15 +0300
  • eb9934a6e5
    Notice the endings a structured answer arrives in Yiorgis Gozadinos 2026-08-12 11:46:27 +0300
  • a79f348ac7
    Recognise our own redirect by a tag, not by its wording Yiorgis Gozadinos 2026-08-12 11:20:11 +0300
  • ae2755e88f
    Enforce a declaration wherever there is something to declare Yiorgis Gozadinos 2026-08-12 10:04:49 +0300
  • 0f38417c60
    Require an answer to declare what grounds it, once per question Yiorgis Gozadinos 2026-08-12 09:32:53 +0300
  • e1e7936d15
    Let a model declare that nothing grounds its answer Yiorgis Gozadinos 2026-08-12 09:32:26 +0300
  • 89b3a9186c
    Merge pull request #535 from ggozad/feat/evidence-compaction Yiorgis Gozadinos 2026-08-13 13:38:22 +0300
  • f4d8e744a2
    Record this release's ORB multimodal numbers Yiorgis Gozadinos 2026-08-13 11:21:31 +0300
  • 97a29d168f
    Confine the ledger's epochs to the question that recorded them Yiorgis Gozadinos 2026-08-12 15:21:11 +0300
  • 11b1bfbc94
    Give the capsule to one return, and the run its own copy of state Yiorgis Gozadinos 2026-08-11 16:52:57 +0300
  • a69f3a8a98
    Document the evidence record and the compaction capability Yiorgis Gozadinos 2026-08-11 14:00:17 +0300
  • 2cd568847e
    Move the wire rewrite into the compaction capability Yiorgis Gozadinos 2026-08-11 13:47:30 +0300
  • 0aa6d79f88
    Build one capsule of cited evidence from the records Yiorgis Gozadinos 2026-08-11 12:59:02 +0300
  • 42d923fe4b
    Make the ledger's clocks defensible against the host Yiorgis Gozadinos 2026-08-11 10:08:10 +0300
  • 85594a1fd1
    Record what each evidence capability retrieved and cited Yiorgis Gozadinos 2026-08-10 15:09:53 +0300
  • fa0c4a4f50
    Treat an unfinished history tail as a continuation too Yiorgis Gozadinos 2026-08-10 14:13:19 +0300
  • 8eef0b3734
    Leave a resumed run's evidence alone Yiorgis Gozadinos 2026-08-10 12:15:47 +0300
  • 46f7ab8d97
    Name the search result each page image belongs to Yiorgis Gozadinos 2026-08-08 17:51:06 +0300
  • d9bd3a701f
    Take the compaction turn boundary from the run, not the message shape Yiorgis Gozadinos 2026-08-08 14:43:29 +0300
  • 5d76461b9f
    Trim earlier-question evidence off the wire, keeping the images Yiorgis Gozadinos 2026-08-07 12:40:05 +0300
  • 8762df5e75
    Merge pull request #540 from ggozad/chore/update-deps Yiorgis Gozadinos 2026-08-13 12:54:27 +0300
  • bf744fc72a
    Update dependencies for security advisories Yiorgis Gozadinos 2026-08-13 12:43:17 +0300
  • 10fd89f547
    Merge pull request #536 from ggozad/fix/reuse-docling-converter Yiorgis Gozadinos 2026-08-13 12:43:40 +0300
  • af77f836d5
    Reuse the local docling converter across documents Yiorgis Gozadinos 2026-08-11 16:18:34 +0300
  • 6186dae83f
    Move ingester telemetry setup behind argument parsing Yiorgis Gozadinos 2026-08-13 11:32:07 +0300
  • 614be8b2a7
    Merge pull request #537 from lawrenceakka/help-speedup Yiorgis Gozadinos 2026-08-13 11:34:42 +0300
  • d4d6414bc4 Similar fix for haiku.rag.ingester, and formatting Lawrence Akka 2026-08-12 15:20:11 +0100
  • 5e6284fd02 Stop eagerly importing Store from haiku.rag.store Lawrence Akka 2026-08-11 17:26:51 +0100
  • 432c81149c
    Merge pull request #529 from ggozad/fix/cross-encoder-activation Yiorgis Gozadinos 2026-08-07 14:17:49 +0300
  • d63199d96d
    Keep cross-encoder rerank scores apart when they saturate Yiorgis Gozadinos 2026-08-07 13:35:38 +0300
  • 96475bba8c
    vb 0.73.0 Yiorgis Gozadinos 2026-08-06 13:55:57 +0300
  • 57531f1653
    Merge pull request #528 from ggozad/chore/pin-eval-judge Yiorgis Gozadinos 2026-08-06 12:50:56 +0200
  • ba963864f3
    Pin the eval judge sampling and standardise on Qwen3-Reranker Yiorgis Gozadinos 2026-08-06 12:18:45 +0300
  • 605367ee5a
    Merge pull request #526 from ggozad/fix/light-document-lookups Yiorgis Gozadinos 2026-08-06 12:15:50 +0200
  • eb11a165b8
    Stop loading page rasters on the title and update paths Yiorgis Gozadinos 2026-08-06 12:28:18 +0300
  • ac9b2cbf81
    Load document blobs only when asked Yiorgis Gozadinos 2026-08-06 11:24:47 +0300
  • 788a2fe731
    Score retrieval evals by document URI Yiorgis Gozadinos 2026-08-06 11:06:13 +0300
  • a4839c0ad2
    Score retrieval evals from search results Yiorgis Gozadinos 2026-08-06 11:02:21 +0300
  • a2afdddc63
    Merge pull request #527 from ggozad/fix/s3-error-classification Yiorgis Gozadinos 2026-08-06 12:03:03 +0200
  • 292835190a
    Classify obstore config errors as permanent Yiorgis Gozadinos 2026-08-06 12:20:40 +0300
  • 5a023a24ad
    vb 0.72.1 Yiorgis Gozadinos 2026-07-31 15:25:28 +0300
  • c1cb9fe6c6
    Merge pull request #525 from ggozad/fix/no-page-loading-list Yiorgis Gozadinos 2026-07-31 11:07:32 +0300
  • 8b45f81464
    Load only content when listing documents with include_content Yiorgis Gozadinos 2026-07-31 10:00:01 +0300
  • adae9eff2e
    vb 0.72.0 Yiorgis Gozadinos 2026-07-30 19:37:45 +0300
  • a9afb176d6
    Merge pull request #524 from ggozad/fix/capability-citations Yiorgis Gozadinos 2026-07-30 19:36:40 +0300
  • c137305468
    Scope the limit notice and spend the cite window on own turns only Yiorgis Gozadinos 2026-07-30 18:50:05 +0300
  • 27ed0b3bb3
    Send analysis to the sandbox when only its search budget is spent Yiorgis Gozadinos 2026-07-30 18:17:26 +0300
  • b62d6e920b
    Drop budget_spent from eval attributes Yiorgis Gozadinos 2026-07-30 15:57:48 +0300
  • 721acbcf38
    Address review on PR #524 Yiorgis Gozadinos 2026-07-30 15:50:06 +0300
  • 68238f0b6a
    Name the readlines workaround when sandbox code iterates a file Yiorgis Gozadinos 2026-07-30 11:21:18 +0300
  • 485a8f901e
    Document the real max_searches default Yiorgis Gozadinos 2026-07-30 09:39:16 +0300
  • a528ab912f
    Record per-case retrieval diagnostics in evaluations Yiorgis Gozadinos 2026-07-29 15:30:26 +0300
  • 5752f61f2c
    Recover chunk ids mistyped from search results Yiorgis Gozadinos 2026-07-29 15:30:20 +0300
  • 685a7c393d
    Keep the cite tool past a capability's request limit Yiorgis Gozadinos 2026-07-29 11:29:31 +0300
  • 749752820c
    Document vacuum memory requirements Yiorgis Gozadinos 2026-07-30 14:32:40 +0300
  • 2ac2594d0d
    Merge pull request #523 from ggozad/fix/docling-dotted-names Yiorgis Gozadinos 2026-07-30 14:30:37 +0300
  • e2b930d469
    Use a neutral dotfile in the docling name examples Yiorgis Gozadinos 2026-07-30 14:13:23 +0300
  • 8e7e494fd1
    Strip leading dots from names handed to docling Yiorgis Gozadinos 2026-07-30 14:07:00 +0300
  • a6b19861ac
    add deadlock mitigations for vacuum bryan davis 2026-07-29 10:26:34 -0500
  • ebd1fc7f3e
    vb 0.71.0 Yiorgis Gozadinos 2026-07-29 09:06:55 +0300
  • de102b7e34
    Merge pull request #521 from ggozad/feat/monty-0.0.19 Yiorgis Gozadinos 2026-07-28 19:25:52 +0300
  • d1ad14f44b
    Point the limit comments at the right limits Yiorgis Gozadinos 2026-07-28 19:09:04 +0300