[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-braintrust-braintrust-discover-agent-failures":3,"mdc-h1jb2u-key":38,"related-org-braintrust-braintrust-discover-agent-failures":381,"related-repo-braintrust-braintrust-discover-agent-failures":548},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":12,"stars":27,"repoUrl":28,"updatedAt":29,"license":30,"forks":31,"topics":32,"repo":33,"sourceUrl":36,"mdContent":37},"braintrust-discover-agent-failures","identify and classify agent failure modes","Search open-endedly for unanticipated agent failure modes and convert them into a named taxonomy and durable regression items. Use for requests to find out what goes wrong, surface unknown or silent failures, do error analysis over traces, cluster and triage production failures, or build a failure taxonomy — where the goal is discovering modes nobody thought to test rather than measuring a predefined criterion. Produces datasets and taxonomies, not headline scores. Do not use for adversarial attacks against a threat model, or to measure a known criterion.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},"braintrust","Braintrust","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fbraintrust.png","braintrustdata",[13,17,20,23,24],{"name":14,"slug":15,"type":16},"Evals","evals","tag",{"name":18,"slug":19,"type":16},"Triage","triage",{"name":21,"slug":22,"type":16},"Agents","agents",{"name":9,"slug":8,"type":16},{"name":25,"slug":26,"type":16},"Debugging","debugging",7,"https:\u002F\u002Fgithub.com\u002Fbraintrustdata\u002Feval-library","2026-08-20T03:53:00.423941",null,0,[],{"repoUrl":28,"stars":27,"forks":31,"topics":34,"description":35},[],"Braintrust eval skills library","https:\u002F\u002Fgithub.com\u002Fbraintrustdata\u002Feval-library\u002Ftree\u002FHEAD\u002Fskills\u002Fbraintrust-discover-agent-failures","---\nname: braintrust-discover-agent-failures\ndescription: >-\n  Search open-endedly for unanticipated agent failure modes and convert them into a named taxonomy\n  and durable regression items. Use for requests to find out what goes wrong, surface unknown or\n  silent failures, do error analysis over traces, cluster and triage production failures, or build a\n  failure taxonomy — where the goal is discovering modes nobody thought to test rather than\n  measuring a predefined criterion. Produces datasets and taxonomies, not headline scores. Do not\n  use for adversarial attacks against a threat model, or to measure a known criterion.\n---\n\n# Discover unknown failure modes\n\nContract: `references\u002Finteraction-contract.md`. Calibration, templates, provenance: `references\u002Ffailure-taxonomy.md`.\n\n## Trigger\n\n- \"What goes wrong with this agent?\" \u002F \"Find failures we haven't thought of.\"\n- Error analysis over a body of traces or production incidents.\n- A suite that passes while the product visibly disappoints.\n\n## Do\n\n1. Accept that you cannot write the scorer in advance — that is the point. Run the loop:\n   **open-ended search → cluster and triage → name a taxonomy → freeze each confirmed mode into a\n   deterministic regression item.**\n2. Generate or search at scale, then constrain the search to **realistic intents**, not only\n   adversarial ones. Purely adversarial generation finds failures no real user will trigger.\n3. Triage each candidate against the trace — genuine system failure, bad item, harness error, or\n   label problem? Only genuine failures enter the taxonomy.\n4. Attribute each mode to its **root cause, not its visible symptom.** Early mistakes cascade, so\n   organize by the module that first erred — memory, reflection, planning, action.\n5. Hunt **silent** failures specifically (`reference.md`). Most agent failures raise no error at\n   all: plausible-looking but unusable output, missing answers, wrong entities.\n6. Freeze each confirmed mode by isolating the decision point where it occurs and snapshotting it,\n   so a flaky trajectory failure becomes a stable regression test.\n\n## Avoid\n\n- Do not force an unknown failure into an existing rubric; the rubric is what missed it.\n- Do not report the search yield as a population failure rate — a discovery sample is not a\n  representative sample.\n- Do not stop at anecdotes; without a taxonomy and frozen items the work does not compound.\n- Do not run adversarial campaigns against a threat model here.\n\n## Check\n\n- Each mode verified against trace evidence, with severity and reproducibility noted.\n- Modes distinct, not restatements of one another at different depths.\n- Taxonomy organized by root cause, with the responsible module named.\n- At least one durable regression item per confirmed mode.\n- Search method and yield disclosed, with the explicit note that yield is not a rate.\n\n## Risk\n\n- Synthetic oddities and unrealistic prompts consume attention without improving reliability.\n- Duplicate modes inflate the apparent problem count and split ownership.\n- You can only discover what was logged — a thin result may be an instrumentation finding rather\n  than a good sign.\n- Discovery never terminates. Stop on a coverage or budget rule stated in advance.\n\n## Braintrust\n\nThe pipeline, each step load-bearing: **capture full traces** (discovery is bounded by\ninstrumentation — if tool calls are not spanned, a planning failure and an action failure are\nindistinguishable and the root-cause taxonomy collapses into a symptom list) → **trace\nclassifications** on live traffic to cluster, classified by suspected mode rather than pass\u002Ffail so\nthe classes *are* the emerging taxonomy → **review queues** for triage, since the\ngenuine-vs-bad-item call needs a human on the trace → **versioned dataset** for confirmed modes as\nfrozen items.\n\nFreeze by snapshotting the **decision point** — the state where the agent first erred becomes the\nitem's input — rather than replaying the whole trajectory. Tag frozen items in `metadata`:\n`source: discovery-\u003CYYYY-MM>`, `mode`, `root_module`, `frozen_from_trace`, `silent`.\n\nKeep discovery items in a **separate dataset or behind a metadata flag** excluded from population\nestimates, or the next aggregate silently becomes \"performance on the hardest cases we could find.\"\nSilent modes need real quality scorers; an exception-rate check will never see them.\n",{"data":39,"body":40},{"name":4,"description":6},{"type":41,"children":42},"root",[43,52,75,82,102,108,177,183,206,212,240,246,269,273,314,369],{"type":44,"tag":45,"props":46,"children":48},"element","h1",{"id":47},"discover-unknown-failure-modes",[49],{"type":50,"value":51},"text","Discover unknown failure modes",{"type":44,"tag":53,"props":54,"children":55},"p",{},[56,58,65,67,73],{"type":50,"value":57},"Contract: ",{"type":44,"tag":59,"props":60,"children":62},"code",{"className":61},[],[63],{"type":50,"value":64},"references\u002Finteraction-contract.md",{"type":50,"value":66},". Calibration, templates, provenance: ",{"type":44,"tag":59,"props":68,"children":70},{"className":69},[],[71],{"type":50,"value":72},"references\u002Ffailure-taxonomy.md",{"type":50,"value":74},".",{"type":44,"tag":76,"props":77,"children":79},"h2",{"id":78},"trigger",[80],{"type":50,"value":81},"Trigger",{"type":44,"tag":83,"props":84,"children":85},"ul",{},[86,92,97],{"type":44,"tag":87,"props":88,"children":89},"li",{},[90],{"type":50,"value":91},"\"What goes wrong with this agent?\" \u002F \"Find failures we haven't thought of.\"",{"type":44,"tag":87,"props":93,"children":94},{},[95],{"type":50,"value":96},"Error analysis over a body of traces or production incidents.",{"type":44,"tag":87,"props":98,"children":99},{},[100],{"type":50,"value":101},"A suite that passes while the product visibly disappoints.",{"type":44,"tag":76,"props":103,"children":105},{"id":104},"do",[106],{"type":50,"value":107},"Do",{"type":44,"tag":109,"props":110,"children":111},"ol",{},[112,123,135,140,152,172],{"type":44,"tag":87,"props":113,"children":114},{},[115,117],{"type":50,"value":116},"Accept that you cannot write the scorer in advance — that is the point. Run the loop:\n",{"type":44,"tag":118,"props":119,"children":120},"strong",{},[121],{"type":50,"value":122},"open-ended search → cluster and triage → name a taxonomy → freeze each confirmed mode into a\ndeterministic regression item.",{"type":44,"tag":87,"props":124,"children":125},{},[126,128,133],{"type":50,"value":127},"Generate or search at scale, then constrain the search to ",{"type":44,"tag":118,"props":129,"children":130},{},[131],{"type":50,"value":132},"realistic intents",{"type":50,"value":134},", not only\nadversarial ones. Purely adversarial generation finds failures no real user will trigger.",{"type":44,"tag":87,"props":136,"children":137},{},[138],{"type":50,"value":139},"Triage each candidate against the trace — genuine system failure, bad item, harness error, or\nlabel problem? Only genuine failures enter the taxonomy.",{"type":44,"tag":87,"props":141,"children":142},{},[143,145,150],{"type":50,"value":144},"Attribute each mode to its ",{"type":44,"tag":118,"props":146,"children":147},{},[148],{"type":50,"value":149},"root cause, not its visible symptom.",{"type":50,"value":151}," Early mistakes cascade, so\norganize by the module that first erred — memory, reflection, planning, action.",{"type":44,"tag":87,"props":153,"children":154},{},[155,157,162,164,170],{"type":50,"value":156},"Hunt ",{"type":44,"tag":118,"props":158,"children":159},{},[160],{"type":50,"value":161},"silent",{"type":50,"value":163}," failures specifically (",{"type":44,"tag":59,"props":165,"children":167},{"className":166},[],[168],{"type":50,"value":169},"reference.md",{"type":50,"value":171},"). Most agent failures raise no error at\nall: plausible-looking but unusable output, missing answers, wrong entities.",{"type":44,"tag":87,"props":173,"children":174},{},[175],{"type":50,"value":176},"Freeze each confirmed mode by isolating the decision point where it occurs and snapshotting it,\nso a flaky trajectory failure becomes a stable regression test.",{"type":44,"tag":76,"props":178,"children":180},{"id":179},"avoid",[181],{"type":50,"value":182},"Avoid",{"type":44,"tag":83,"props":184,"children":185},{},[186,191,196,201],{"type":44,"tag":87,"props":187,"children":188},{},[189],{"type":50,"value":190},"Do not force an unknown failure into an existing rubric; the rubric is what missed it.",{"type":44,"tag":87,"props":192,"children":193},{},[194],{"type":50,"value":195},"Do not report the search yield as a population failure rate — a discovery sample is not a\nrepresentative sample.",{"type":44,"tag":87,"props":197,"children":198},{},[199],{"type":50,"value":200},"Do not stop at anecdotes; without a taxonomy and frozen items the work does not compound.",{"type":44,"tag":87,"props":202,"children":203},{},[204],{"type":50,"value":205},"Do not run adversarial campaigns against a threat model here.",{"type":44,"tag":76,"props":207,"children":209},{"id":208},"check",[210],{"type":50,"value":211},"Check",{"type":44,"tag":83,"props":213,"children":214},{},[215,220,225,230,235],{"type":44,"tag":87,"props":216,"children":217},{},[218],{"type":50,"value":219},"Each mode verified against trace evidence, with severity and reproducibility noted.",{"type":44,"tag":87,"props":221,"children":222},{},[223],{"type":50,"value":224},"Modes distinct, not restatements of one another at different depths.",{"type":44,"tag":87,"props":226,"children":227},{},[228],{"type":50,"value":229},"Taxonomy organized by root cause, with the responsible module named.",{"type":44,"tag":87,"props":231,"children":232},{},[233],{"type":50,"value":234},"At least one durable regression item per confirmed mode.",{"type":44,"tag":87,"props":236,"children":237},{},[238],{"type":50,"value":239},"Search method and yield disclosed, with the explicit note that yield is not a rate.",{"type":44,"tag":76,"props":241,"children":243},{"id":242},"risk",[244],{"type":50,"value":245},"Risk",{"type":44,"tag":83,"props":247,"children":248},{},[249,254,259,264],{"type":44,"tag":87,"props":250,"children":251},{},[252],{"type":50,"value":253},"Synthetic oddities and unrealistic prompts consume attention without improving reliability.",{"type":44,"tag":87,"props":255,"children":256},{},[257],{"type":50,"value":258},"Duplicate modes inflate the apparent problem count and split ownership.",{"type":44,"tag":87,"props":260,"children":261},{},[262],{"type":50,"value":263},"You can only discover what was logged — a thin result may be an instrumentation finding rather\nthan a good sign.",{"type":44,"tag":87,"props":265,"children":266},{},[267],{"type":50,"value":268},"Discovery never terminates. Stop on a coverage or budget rule stated in advance.",{"type":44,"tag":76,"props":270,"children":271},{"id":8},[272],{"type":50,"value":9},{"type":44,"tag":53,"props":274,"children":275},{},[276,278,283,285,290,292,298,300,305,307,312],{"type":50,"value":277},"The pipeline, each step load-bearing: ",{"type":44,"tag":118,"props":279,"children":280},{},[281],{"type":50,"value":282},"capture full traces",{"type":50,"value":284}," (discovery is bounded by\ninstrumentation — if tool calls are not spanned, a planning failure and an action failure are\nindistinguishable and the root-cause taxonomy collapses into a symptom list) → ",{"type":44,"tag":118,"props":286,"children":287},{},[288],{"type":50,"value":289},"trace\nclassifications",{"type":50,"value":291}," on live traffic to cluster, classified by suspected mode rather than pass\u002Ffail so\nthe classes ",{"type":44,"tag":293,"props":294,"children":295},"em",{},[296],{"type":50,"value":297},"are",{"type":50,"value":299}," the emerging taxonomy → ",{"type":44,"tag":118,"props":301,"children":302},{},[303],{"type":50,"value":304},"review queues",{"type":50,"value":306}," for triage, since the\ngenuine-vs-bad-item call needs a human on the trace → ",{"type":44,"tag":118,"props":308,"children":309},{},[310],{"type":50,"value":311},"versioned dataset",{"type":50,"value":313}," for confirmed modes as\nfrozen items.",{"type":44,"tag":53,"props":315,"children":316},{},[317,319,324,326,332,334,340,342,348,349,355,356,362,363,368],{"type":50,"value":318},"Freeze by snapshotting the ",{"type":44,"tag":118,"props":320,"children":321},{},[322],{"type":50,"value":323},"decision point",{"type":50,"value":325}," — the state where the agent first erred becomes the\nitem's input — rather than replaying the whole trajectory. Tag frozen items in ",{"type":44,"tag":59,"props":327,"children":329},{"className":328},[],[330],{"type":50,"value":331},"metadata",{"type":50,"value":333},":\n",{"type":44,"tag":59,"props":335,"children":337},{"className":336},[],[338],{"type":50,"value":339},"source: discovery-\u003CYYYY-MM>",{"type":50,"value":341},", ",{"type":44,"tag":59,"props":343,"children":345},{"className":344},[],[346],{"type":50,"value":347},"mode",{"type":50,"value":341},{"type":44,"tag":59,"props":350,"children":352},{"className":351},[],[353],{"type":50,"value":354},"root_module",{"type":50,"value":341},{"type":44,"tag":59,"props":357,"children":359},{"className":358},[],[360],{"type":50,"value":361},"frozen_from_trace",{"type":50,"value":341},{"type":44,"tag":59,"props":364,"children":366},{"className":365},[],[367],{"type":50,"value":161},{"type":50,"value":74},{"type":44,"tag":53,"props":370,"children":371},{},[372,374,379],{"type":50,"value":373},"Keep discovery items in a ",{"type":44,"tag":118,"props":375,"children":376},{},[377],{"type":50,"value":378},"separate dataset or behind a metadata flag",{"type":50,"value":380}," excluded from population\nestimates, or the next aggregate silently becomes \"performance on the hardest cases we could find.\"\nSilent modes need real quality scorers; an exception-rate check will never see them.",{"items":382,"total":547},[383,397,415,429,443,459,473,485,499,515,527,539],{"slug":384,"name":384,"fn":385,"description":386,"org":387,"tags":388,"stars":394,"repoUrl":395,"updatedAt":396},"troubleshoot-braintrust-mcp","configure and troubleshoot Braintrust MCP servers","This plugin auto-configures a \"braintrust\" MCP server. If you can't see it or reach it, activate this skill\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[389,390,391],{"name":9,"slug":8,"type":16},{"name":25,"slug":26,"type":16},{"name":392,"slug":393,"type":16},"MCP","mcp",18,"https:\u002F\u002Fgithub.com\u002Fbraintrustdata\u002Fbraintrust-claude-plugin","2026-07-12T08:36:13.889274",{"slug":398,"name":398,"fn":399,"description":400,"org":401,"tags":402,"stars":27,"repoUrl":28,"updatedAt":414},"braintrust-analyze-eval-experiment","analyze LLM and agent eval experiments","Analyze completed LLM or agent eval experiments using uncertainty-aware and decision-relevant methods. Use to audit run completeness and pairing, calculate confidence intervals, run paired comparisons, report wins, losses, and ties, incorporate run-to-run variance, handle multiple comparisons, inspect subgroup performance, and test fragility to favorable slices. Use when results already exist and someone asks what they mean, whether a difference is real, or which model won. Do not use to design an experiment that has not yet collected results.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[403,406,407,408,411],{"name":404,"slug":405,"type":16},"Analysis","analysis",{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":409,"slug":410,"type":16},"LLM","llm",{"name":412,"slug":413,"type":16},"Statistics","statistics","2026-08-20T03:53:01.13806",{"slug":416,"name":416,"fn":417,"description":418,"org":419,"tags":420,"stars":27,"repoUrl":28,"updatedAt":428},"braintrust-attribute-multi-variable-change","attribute performance changes to multiple variables","Attribute an observed change when several things moved at once — model plus prompt plus tools, a provider migration, a framework upgrade, or a vendor swap that bundles serving stack with model. Use when asked which part of a change caused the result, when a comparison's arms differ in more than one way, when a treatment has no uniform implementation across vendors, or when a serving-stack difference is confounded with a model difference. Do not use for a clean single-variable comparison, or to design an experiment that has not yet run.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[421,422,423,424,425],{"name":404,"slug":405,"type":16},{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":409,"slug":410,"type":16},{"name":426,"slug":427,"type":16},"Performance","performance","2026-08-20T03:53:40.036077",{"slug":430,"name":430,"fn":431,"description":432,"org":433,"tags":434,"stars":27,"repoUrl":28,"updatedAt":442},"braintrust-build-eval-dataset","create and manage LLM eval datasets","Create, edit, audit, or compare eval datasets for LLM applications and agents, including target-population definition, case sourcing from production traces, stratified sampling, label provenance and label audits, expected values as constraints for open-ended tasks, dev\u002Ftest splits, contamination and leakage controls, headroom checks, refresh policy, and datasheets. Use when working on the content or lifecycle of an eval dataset. Do not use for sample-size or power calculations, scorer implementation, or open-ended adversarial failure discovery.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[435,436,437,440,441],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":438,"slug":439,"type":16},"Datasets","datasets",{"name":14,"slug":15,"type":16},{"name":409,"slug":410,"type":16},"2026-08-20T03:53:33.304815",{"slug":444,"name":444,"fn":445,"description":446,"org":447,"tags":448,"stars":27,"repoUrl":28,"updatedAt":458},"braintrust-define-eval-objective","define LLM evaluation objectives","Create, edit, or audit an eval objective by working backward from a product decision to the target outcome, construct, population, intended claim, and verification-versus-validation questions. Use when a team is unsure what an eval should establish, asks \"what are we actually trying to measure,\" \"is this eval measuring the right thing,\" \"does this benchmark support our claim,\" or needs to turn a product goal into an eval objective and state which claims are out of scope. Do not use to select detailed metrics, design datasets, or implement scorers.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[449,450,451,452,455],{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":409,"slug":410,"type":16},{"name":453,"slug":454,"type":16},"Product Management","product-management",{"name":456,"slug":457,"type":16},"Strategy","strategy","2026-08-20T03:53:00.07097",{"slug":460,"name":460,"fn":461,"description":462,"org":463,"tags":464,"stars":27,"repoUrl":28,"updatedAt":472},"braintrust-define-eval-release-gate","configure release gates for LLM applications","Create, edit, audit, or apply release gates for LLM applications and agents. Use to combine minimum meaningful improvement, statistical significance, regression rate, subgroup consistency, worst-run stability, all-attempts reliability, safety upper bounds, latency, and cost into an explicit ship-or-hold policy, to turn metrics into a CI gate, or to explain why a candidate failed a gate and what evidence would justify reconsideration. Do not use for general result analysis without a deployment decision.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[465,466,467,470,471],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":468,"slug":469,"type":16},"CI\u002FCD","ci-cd",{"name":14,"slug":15,"type":16},{"name":409,"slug":410,"type":16},"2026-08-20T03:53:15.686158",{"slug":474,"name":474,"fn":475,"description":476,"org":477,"tags":478,"stars":27,"repoUrl":28,"updatedAt":484},"braintrust-deploy-evaluator","deploy evaluators to Braintrust","Take a validated scorer or classifier from definition to running instrument in Braintrust — scope selection, inline testing before saving, saving as an evaluator, attaching an online-scoring rule, activating it for new traffic, and backfilling history with a rewind. Use when a scorer needs to actually run against production logs, when an online-scoring rule needs to be created or changed, or when historical traces need scoring. Do not use to decide what the scorer should measure, to write its rubric, or to establish that it agrees with human judgment.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[479,480,483],{"name":9,"slug":8,"type":16},{"name":481,"slug":482,"type":16},"Deployment","deployment",{"name":14,"slug":15,"type":16},"2026-08-20T03:53:32.558937",{"slug":486,"name":486,"fn":487,"description":488,"org":489,"tags":490,"stars":27,"repoUrl":28,"updatedAt":498},"braintrust-design-eval-experiment","design controlled LLM eval experiments","Design or audit controlled eval experiments for model, prompt, retrieval, tool, guardrail, or agent-architecture changes. Use before data collection to state directional and minimum-effect hypotheses, name independent, dependent, and control variables including the serving environment and tool surface, choose paired designs, set repetitions and allocation, distinguish exploratory from confirmatory comparisons, and pre-specify stopping, exclusion, multiplicity, and analysis rules. Do not use primarily to analyze results already collected.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[491,492,493,496,497],{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":494,"slug":495,"type":16},"Experiments","experiments",{"name":409,"slug":410,"type":16},{"name":456,"slug":457,"type":16},"2026-08-20T03:53:36.534554",{"slug":500,"name":500,"fn":501,"description":502,"org":503,"tags":504,"stars":27,"repoUrl":28,"updatedAt":514},"braintrust-design-eval-instrumentation","design trace and evaluation dataset schemas","Design the trace and eval-dataset schema for an LLM app or agent, and wire the system to emit it. Use when deciding what to log, designing a trace schema, setting up tracing or observability before evals, or when failures cannot be debugged or sliced from existing traces — covering inputs, outputs, spans for tool and LLM calls, state changes, metadata, resolved configuration, serving path, tool manifest, per-item status, attachments, and subgroup variables. Do not use to decide what the evidence should mean, or to write scorers.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[505,506,507,508,511],{"name":9,"slug":8,"type":16},{"name":438,"slug":439,"type":16},{"name":14,"slug":15,"type":16},{"name":509,"slug":510,"type":16},"Observability","observability",{"name":512,"slug":513,"type":16},"Tracing","tracing","2026-08-20T03:53:37.274703",{"slug":516,"name":516,"fn":517,"description":518,"org":519,"tags":520,"stars":27,"repoUrl":28,"updatedAt":526},"braintrust-design-eval-metric-bundle","create multi-objective evaluation metric bundles","Create, edit, audit, or compare a multi-objective eval metric bundle covering product quality, safety, reliability, latency, and cost. Use to choose metrics for an eval, define a goodness bundle, distinguish optimization metrics from non-regression guardrails, expose tradeoffs, audit a KPI or single composite score for Goodhart and metric-gaming risk, or answer \"what should improve and what must not regress.\" Do not use to design trace schemas, build datasets, or implement scoring methods.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[521,522,523,524,525],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":409,"slug":410,"type":16},{"name":426,"slug":427,"type":16},"2026-08-20T03:53:16.078609",{"slug":528,"name":528,"fn":529,"description":530,"org":531,"tags":532,"stars":27,"repoUrl":28,"updatedAt":538},"braintrust-design-human-eval-review","design human evaluation and review workflows","Design or audit human evaluation workflows and golden datasets for LLM applications and agents. Use to set up expert review, select review cases, write reviewer instructions, assign raters, capture rationales and confidence, measure inter-rater agreement with kappa or alpha, adjudicate disagreements, and preserve reviewed examples with provenance as a versioned reference set. Do not use to elicit the criteria or rubric in the first place, to validate a scorer once reference labels exist, or to implement the scorer.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[533,534,535,536,537],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":438,"slug":439,"type":16},{"name":14,"slug":15,"type":16},{"name":409,"slug":410,"type":16},"2026-08-20T03:53:36.900129",{"slug":4,"name":4,"fn":5,"description":6,"org":540,"tags":541,"stars":27,"repoUrl":28,"updatedAt":29},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[542,543,544,545,546],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":25,"slug":26,"type":16},{"name":14,"slug":15,"type":16},{"name":18,"slug":19,"type":16},27,{"items":549,"total":604},[550,558,566,574,582,590,596],{"slug":398,"name":398,"fn":399,"description":400,"org":551,"tags":552,"stars":27,"repoUrl":28,"updatedAt":414},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[553,554,555,556,557],{"name":404,"slug":405,"type":16},{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":409,"slug":410,"type":16},{"name":412,"slug":413,"type":16},{"slug":416,"name":416,"fn":417,"description":418,"org":559,"tags":560,"stars":27,"repoUrl":28,"updatedAt":428},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[561,562,563,564,565],{"name":404,"slug":405,"type":16},{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":409,"slug":410,"type":16},{"name":426,"slug":427,"type":16},{"slug":430,"name":430,"fn":431,"description":432,"org":567,"tags":568,"stars":27,"repoUrl":28,"updatedAt":442},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[569,570,571,572,573],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":438,"slug":439,"type":16},{"name":14,"slug":15,"type":16},{"name":409,"slug":410,"type":16},{"slug":444,"name":444,"fn":445,"description":446,"org":575,"tags":576,"stars":27,"repoUrl":28,"updatedAt":458},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[577,578,579,580,581],{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":409,"slug":410,"type":16},{"name":453,"slug":454,"type":16},{"name":456,"slug":457,"type":16},{"slug":460,"name":460,"fn":461,"description":462,"org":583,"tags":584,"stars":27,"repoUrl":28,"updatedAt":472},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[585,586,587,588,589],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":468,"slug":469,"type":16},{"name":14,"slug":15,"type":16},{"name":409,"slug":410,"type":16},{"slug":474,"name":474,"fn":475,"description":476,"org":591,"tags":592,"stars":27,"repoUrl":28,"updatedAt":484},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[593,594,595],{"name":9,"slug":8,"type":16},{"name":481,"slug":482,"type":16},{"name":14,"slug":15,"type":16},{"slug":486,"name":486,"fn":487,"description":488,"org":597,"tags":598,"stars":27,"repoUrl":28,"updatedAt":498},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[599,600,601,602,603],{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":494,"slug":495,"type":16},{"name":409,"slug":410,"type":16},{"name":456,"slug":457,"type":16},24]