[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-braintrust-braintrust-size-eval-dataset":3,"mdc--4pt5s6-key":32,"related-repo-braintrust-braintrust-size-eval-dataset":290,"related-org-braintrust-braintrust-size-eval-dataset":395},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":12,"stars":21,"repoUrl":22,"updatedAt":23,"license":24,"forks":25,"topics":26,"repo":27,"sourceUrl":30,"mdContent":31},"braintrust-size-eval-dataset","calculate evaluation dataset sample sizes","Calculate or audit eval sample sizes, minimum detectable effects, confidence interval precision, required repeated runs, and clean-trial counts for bounding rare failures. Use when a user asks how many eval cases, items, scenarios, runs, or safety trials are needed, whether an existing dataset is adequately powered, whether N examples can detect an X-point gain, or how many clean trials certify a low violation rate. Account for paired designs, clustering, target confidence, and practical effect size. Do not use for general dataset composition.",{"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],{"name":14,"slug":15,"type":16},"Evals","evals","tag",{"name":18,"slug":19,"type":16},"Statistics","statistics",{"name":9,"slug":8,"type":16},7,"https:\u002F\u002Fgithub.com\u002Fbraintrustdata\u002Feval-library","2026-08-20T03:53:32.932246",null,0,[],{"repoUrl":22,"stars":21,"forks":25,"topics":28,"description":29},[],"Braintrust eval skills library","https:\u002F\u002Fgithub.com\u002Fbraintrustdata\u002Feval-library\u002Ftree\u002FHEAD\u002Fskills\u002Fbraintrust-size-eval-dataset","---\nname: braintrust-size-eval-dataset\ndescription: >-\n  Calculate or audit eval sample sizes, minimum detectable effects, confidence interval\n  precision, required repeated runs, and clean-trial counts for bounding rare failures. Use when\n  a user asks how many eval cases, items, scenarios, runs, or safety trials are needed, whether\n  an existing dataset is adequately powered, whether N examples can detect an X-point gain, or\n  how many clean trials certify a low violation rate. Account for paired designs, clustering,\n  target confidence, and practical effect size. Do not use for general dataset composition.\n---\n\n# Size the dataset\n\nContract: `references\u002Finteraction-contract.md`. Calibration, templates, provenance: `references\u002Fpower-analysis.md`.\n\n## Trigger\n\n- \"How many eval cases do we need?\" \u002F \"Can 200 examples detect a 3-point gain?\"\n- \"How many clean trials certify a low failure rate?\"\n- A null result that needs distinguishing from an underpowered one.\n\n## Do\n\n1. Classify the goal first — it selects the method: **detect a difference**, **estimate a\n   rate**, or **bound a rare failure**.\n2. Take the minimum meaningful effect from the release gate's magnitude criterion, not from what\n   the data shows. Ask only if genuinely absent.\n3. Compute N with the method and anchors in `reference.md`, applying the **design effect** for\n   clustered items — related variants are fewer items than they look.\n4. Multiply through the whole matrix: items × arms × runs (K ≥ 3 for gating), and check the\n   product against every quota and credit balance in the path before anyone starts.\n5. Report a **sensitivity range**, not a single number, and close with the plain-language\n   decision implication.\n\n## Avoid\n\n- Do not return a sample size without stating the practical effect, baseline rate, risk\n  tolerance, clustering, pairing, confidence, and power assumptions.\n- Do not use the normal approximation near 0 or 1 — use Wilson, which is exactly the\n  safety-rate case.\n- Do not design content here.\n- Do not let a cheap smoke run over a handful of items be quoted as evidence about which option\n  is better. It verifies plumbing, nothing else.\n\n## Check\n\n- Goal type identified; method named; inputs shown; every assumption listed with its source.\n- Design effect applied for clustered or repeated-variant items.\n- Sensitivity range alongside the point answer.\n- A plain sentence stating what the resulting N can and cannot establish.\n\n## Risk\n\n- Treating correlated variants as independent overstates precision more than any other single\n  error in eval reporting.\n- Normal approximations near zero produce impossible bounds and false safety assurances.\n- Powering for the effect you hope for rather than the effect worth acting on guarantees an\n  ambiguous result.\n- An underpowered eval and a real null produce the identical sentence: \"no measurable effect.\"\n\n## Braintrust\n\nShared mechanics: `references\u002Fplatform-mechanics.md`. Pairing is what buys the sample-size\nreduction, and **§3** makes it a platform decision rather than a later statistical one — a design\nsized for pairing silently reverts to the unpaired requirement if the arms were never pinned. So\nthe N you compute here is only achievable if the experiment design honors that.\n\nPlanned N and achieved N diverge routinely, which makes the §2 read-safety checks a sizing\nconcern and not just an analysis one: a truncated pull looks exactly like an underpowered run,\nand the two call for opposite responses. K runs means K recorded trials (§7); the worst run is\nwhat a latency or reliability gate reads.\n",{"data":33,"body":34},{"name":4,"description":6},{"type":35,"children":36},"root",[37,46,69,76,96,102,174,180,203,209,232,238,261,265,285],{"type":38,"tag":39,"props":40,"children":42},"element","h1",{"id":41},"size-the-dataset",[43],{"type":44,"value":45},"text","Size the dataset",{"type":38,"tag":47,"props":48,"children":49},"p",{},[50,52,59,61,67],{"type":44,"value":51},"Contract: ",{"type":38,"tag":53,"props":54,"children":56},"code",{"className":55},[],[57],{"type":44,"value":58},"references\u002Finteraction-contract.md",{"type":44,"value":60},". Calibration, templates, provenance: ",{"type":38,"tag":53,"props":62,"children":64},{"className":63},[],[65],{"type":44,"value":66},"references\u002Fpower-analysis.md",{"type":44,"value":68},".",{"type":38,"tag":70,"props":71,"children":73},"h2",{"id":72},"trigger",[74],{"type":44,"value":75},"Trigger",{"type":38,"tag":77,"props":78,"children":79},"ul",{},[80,86,91],{"type":38,"tag":81,"props":82,"children":83},"li",{},[84],{"type":44,"value":85},"\"How many eval cases do we need?\" \u002F \"Can 200 examples detect a 3-point gain?\"",{"type":38,"tag":81,"props":87,"children":88},{},[89],{"type":44,"value":90},"\"How many clean trials certify a low failure rate?\"",{"type":38,"tag":81,"props":92,"children":93},{},[94],{"type":44,"value":95},"A null result that needs distinguishing from an underpowered one.",{"type":38,"tag":70,"props":97,"children":99},{"id":98},"do",[100],{"type":44,"value":101},"Do",{"type":38,"tag":103,"props":104,"children":105},"ol",{},[106,132,137,157,162],{"type":38,"tag":81,"props":107,"children":108},{},[109,111,117,119,124,126,131],{"type":44,"value":110},"Classify the goal first — it selects the method: ",{"type":38,"tag":112,"props":113,"children":114},"strong",{},[115],{"type":44,"value":116},"detect a difference",{"type":44,"value":118},", ",{"type":38,"tag":112,"props":120,"children":121},{},[122],{"type":44,"value":123},"estimate a\nrate",{"type":44,"value":125},", or ",{"type":38,"tag":112,"props":127,"children":128},{},[129],{"type":44,"value":130},"bound a rare failure",{"type":44,"value":68},{"type":38,"tag":81,"props":133,"children":134},{},[135],{"type":44,"value":136},"Take the minimum meaningful effect from the release gate's magnitude criterion, not from what\nthe data shows. Ask only if genuinely absent.",{"type":38,"tag":81,"props":138,"children":139},{},[140,142,148,150,155],{"type":44,"value":141},"Compute N with the method and anchors in ",{"type":38,"tag":53,"props":143,"children":145},{"className":144},[],[146],{"type":44,"value":147},"reference.md",{"type":44,"value":149},", applying the ",{"type":38,"tag":112,"props":151,"children":152},{},[153],{"type":44,"value":154},"design effect",{"type":44,"value":156}," for\nclustered items — related variants are fewer items than they look.",{"type":38,"tag":81,"props":158,"children":159},{},[160],{"type":44,"value":161},"Multiply through the whole matrix: items × arms × runs (K ≥ 3 for gating), and check the\nproduct against every quota and credit balance in the path before anyone starts.",{"type":38,"tag":81,"props":163,"children":164},{},[165,167,172],{"type":44,"value":166},"Report a ",{"type":38,"tag":112,"props":168,"children":169},{},[170],{"type":44,"value":171},"sensitivity range",{"type":44,"value":173},", not a single number, and close with the plain-language\ndecision implication.",{"type":38,"tag":70,"props":175,"children":177},{"id":176},"avoid",[178],{"type":44,"value":179},"Avoid",{"type":38,"tag":77,"props":181,"children":182},{},[183,188,193,198],{"type":38,"tag":81,"props":184,"children":185},{},[186],{"type":44,"value":187},"Do not return a sample size without stating the practical effect, baseline rate, risk\ntolerance, clustering, pairing, confidence, and power assumptions.",{"type":38,"tag":81,"props":189,"children":190},{},[191],{"type":44,"value":192},"Do not use the normal approximation near 0 or 1 — use Wilson, which is exactly the\nsafety-rate case.",{"type":38,"tag":81,"props":194,"children":195},{},[196],{"type":44,"value":197},"Do not design content here.",{"type":38,"tag":81,"props":199,"children":200},{},[201],{"type":44,"value":202},"Do not let a cheap smoke run over a handful of items be quoted as evidence about which option\nis better. It verifies plumbing, nothing else.",{"type":38,"tag":70,"props":204,"children":206},{"id":205},"check",[207],{"type":44,"value":208},"Check",{"type":38,"tag":77,"props":210,"children":211},{},[212,217,222,227],{"type":38,"tag":81,"props":213,"children":214},{},[215],{"type":44,"value":216},"Goal type identified; method named; inputs shown; every assumption listed with its source.",{"type":38,"tag":81,"props":218,"children":219},{},[220],{"type":44,"value":221},"Design effect applied for clustered or repeated-variant items.",{"type":38,"tag":81,"props":223,"children":224},{},[225],{"type":44,"value":226},"Sensitivity range alongside the point answer.",{"type":38,"tag":81,"props":228,"children":229},{},[230],{"type":44,"value":231},"A plain sentence stating what the resulting N can and cannot establish.",{"type":38,"tag":70,"props":233,"children":235},{"id":234},"risk",[236],{"type":44,"value":237},"Risk",{"type":38,"tag":77,"props":239,"children":240},{},[241,246,251,256],{"type":38,"tag":81,"props":242,"children":243},{},[244],{"type":44,"value":245},"Treating correlated variants as independent overstates precision more than any other single\nerror in eval reporting.",{"type":38,"tag":81,"props":247,"children":248},{},[249],{"type":44,"value":250},"Normal approximations near zero produce impossible bounds and false safety assurances.",{"type":38,"tag":81,"props":252,"children":253},{},[254],{"type":44,"value":255},"Powering for the effect you hope for rather than the effect worth acting on guarantees an\nambiguous result.",{"type":38,"tag":81,"props":257,"children":258},{},[259],{"type":44,"value":260},"An underpowered eval and a real null produce the identical sentence: \"no measurable effect.\"",{"type":38,"tag":70,"props":262,"children":263},{"id":8},[264],{"type":44,"value":9},{"type":38,"tag":47,"props":266,"children":267},{},[268,270,276,278,283],{"type":44,"value":269},"Shared mechanics: ",{"type":38,"tag":53,"props":271,"children":273},{"className":272},[],[274],{"type":44,"value":275},"references\u002Fplatform-mechanics.md",{"type":44,"value":277},". Pairing is what buys the sample-size\nreduction, and ",{"type":38,"tag":112,"props":279,"children":280},{},[281],{"type":44,"value":282},"§3",{"type":44,"value":284}," makes it a platform decision rather than a later statistical one — a design\nsized for pairing silently reverts to the unpaired requirement if the arms were never pinned. So\nthe N you compute here is only achievable if the experiment design honors that.",{"type":38,"tag":47,"props":286,"children":287},{},[288],{"type":44,"value":289},"Planned N and achieved N diverge routinely, which makes the §2 read-safety checks a sizing\nconcern and not just an analysis one: a truncated pull looks exactly like an underpowered run,\nand the two call for opposite responses. K runs means K recorded trials (§7); the worst run is\nwhat a latency or reliability gate reads.",{"items":291,"total":394},[292,308,322,338,354,368,380],{"slug":293,"name":293,"fn":294,"description":295,"org":296,"tags":297,"stars":21,"repoUrl":22,"updatedAt":307},"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},[298,301,302,303,306],{"name":299,"slug":300,"type":16},"Analysis","analysis",{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":304,"slug":305,"type":16},"LLM","llm",{"name":18,"slug":19,"type":16},"2026-08-20T03:53:01.13806",{"slug":309,"name":309,"fn":310,"description":311,"org":312,"tags":313,"stars":21,"repoUrl":22,"updatedAt":321},"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},[314,315,316,317,318],{"name":299,"slug":300,"type":16},{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":304,"slug":305,"type":16},{"name":319,"slug":320,"type":16},"Performance","performance","2026-08-20T03:53:40.036077",{"slug":323,"name":323,"fn":324,"description":325,"org":326,"tags":327,"stars":21,"repoUrl":22,"updatedAt":337},"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},[328,331,332,335,336],{"name":329,"slug":330,"type":16},"Agents","agents",{"name":9,"slug":8,"type":16},{"name":333,"slug":334,"type":16},"Datasets","datasets",{"name":14,"slug":15,"type":16},{"name":304,"slug":305,"type":16},"2026-08-20T03:53:33.304815",{"slug":339,"name":339,"fn":340,"description":341,"org":342,"tags":343,"stars":21,"repoUrl":22,"updatedAt":353},"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},[344,345,346,347,350],{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":304,"slug":305,"type":16},{"name":348,"slug":349,"type":16},"Product Management","product-management",{"name":351,"slug":352,"type":16},"Strategy","strategy","2026-08-20T03:53:00.07097",{"slug":355,"name":355,"fn":356,"description":357,"org":358,"tags":359,"stars":21,"repoUrl":22,"updatedAt":367},"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},[360,361,362,365,366],{"name":329,"slug":330,"type":16},{"name":9,"slug":8,"type":16},{"name":363,"slug":364,"type":16},"CI\u002FCD","ci-cd",{"name":14,"slug":15,"type":16},{"name":304,"slug":305,"type":16},"2026-08-20T03:53:15.686158",{"slug":369,"name":369,"fn":370,"description":371,"org":372,"tags":373,"stars":21,"repoUrl":22,"updatedAt":379},"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},[374,375,378],{"name":9,"slug":8,"type":16},{"name":376,"slug":377,"type":16},"Deployment","deployment",{"name":14,"slug":15,"type":16},"2026-08-20T03:53:32.558937",{"slug":381,"name":381,"fn":382,"description":383,"org":384,"tags":385,"stars":21,"repoUrl":22,"updatedAt":393},"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},[386,387,388,391,392],{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":389,"slug":390,"type":16},"Experiments","experiments",{"name":304,"slug":305,"type":16},{"name":351,"slug":352,"type":16},"2026-08-20T03:53:36.534554",24,{"items":396,"total":521},[397,413,421,429,437,445,453,459,467,483,495,507],{"slug":398,"name":398,"fn":399,"description":400,"org":401,"tags":402,"stars":410,"repoUrl":411,"updatedAt":412},"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},[403,404,407],{"name":9,"slug":8,"type":16},{"name":405,"slug":406,"type":16},"Debugging","debugging",{"name":408,"slug":409,"type":16},"MCP","mcp",18,"https:\u002F\u002Fgithub.com\u002Fbraintrustdata\u002Fbraintrust-claude-plugin","2026-07-12T08:36:13.889274",{"slug":293,"name":293,"fn":294,"description":295,"org":414,"tags":415,"stars":21,"repoUrl":22,"updatedAt":307},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[416,417,418,419,420],{"name":299,"slug":300,"type":16},{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":304,"slug":305,"type":16},{"name":18,"slug":19,"type":16},{"slug":309,"name":309,"fn":310,"description":311,"org":422,"tags":423,"stars":21,"repoUrl":22,"updatedAt":321},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[424,425,426,427,428],{"name":299,"slug":300,"type":16},{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":304,"slug":305,"type":16},{"name":319,"slug":320,"type":16},{"slug":323,"name":323,"fn":324,"description":325,"org":430,"tags":431,"stars":21,"repoUrl":22,"updatedAt":337},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[432,433,434,435,436],{"name":329,"slug":330,"type":16},{"name":9,"slug":8,"type":16},{"name":333,"slug":334,"type":16},{"name":14,"slug":15,"type":16},{"name":304,"slug":305,"type":16},{"slug":339,"name":339,"fn":340,"description":341,"org":438,"tags":439,"stars":21,"repoUrl":22,"updatedAt":353},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[440,441,442,443,444],{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":304,"slug":305,"type":16},{"name":348,"slug":349,"type":16},{"name":351,"slug":352,"type":16},{"slug":355,"name":355,"fn":356,"description":357,"org":446,"tags":447,"stars":21,"repoUrl":22,"updatedAt":367},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[448,449,450,451,452],{"name":329,"slug":330,"type":16},{"name":9,"slug":8,"type":16},{"name":363,"slug":364,"type":16},{"name":14,"slug":15,"type":16},{"name":304,"slug":305,"type":16},{"slug":369,"name":369,"fn":370,"description":371,"org":454,"tags":455,"stars":21,"repoUrl":22,"updatedAt":379},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[456,457,458],{"name":9,"slug":8,"type":16},{"name":376,"slug":377,"type":16},{"name":14,"slug":15,"type":16},{"slug":381,"name":381,"fn":382,"description":383,"org":460,"tags":461,"stars":21,"repoUrl":22,"updatedAt":393},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[462,463,464,465,466],{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":389,"slug":390,"type":16},{"name":304,"slug":305,"type":16},{"name":351,"slug":352,"type":16},{"slug":468,"name":468,"fn":469,"description":470,"org":471,"tags":472,"stars":21,"repoUrl":22,"updatedAt":482},"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},[473,474,475,476,479],{"name":9,"slug":8,"type":16},{"name":333,"slug":334,"type":16},{"name":14,"slug":15,"type":16},{"name":477,"slug":478,"type":16},"Observability","observability",{"name":480,"slug":481,"type":16},"Tracing","tracing","2026-08-20T03:53:37.274703",{"slug":484,"name":484,"fn":485,"description":486,"org":487,"tags":488,"stars":21,"repoUrl":22,"updatedAt":494},"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},[489,490,491,492,493],{"name":329,"slug":330,"type":16},{"name":9,"slug":8,"type":16},{"name":14,"slug":15,"type":16},{"name":304,"slug":305,"type":16},{"name":319,"slug":320,"type":16},"2026-08-20T03:53:16.078609",{"slug":496,"name":496,"fn":497,"description":498,"org":499,"tags":500,"stars":21,"repoUrl":22,"updatedAt":506},"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},[501,502,503,504,505],{"name":329,"slug":330,"type":16},{"name":9,"slug":8,"type":16},{"name":333,"slug":334,"type":16},{"name":14,"slug":15,"type":16},{"name":304,"slug":305,"type":16},"2026-08-20T03:53:36.900129",{"slug":508,"name":508,"fn":509,"description":510,"org":511,"tags":512,"stars":21,"repoUrl":22,"updatedAt":520},"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},[513,514,515,516,517],{"name":329,"slug":330,"type":16},{"name":9,"slug":8,"type":16},{"name":405,"slug":406,"type":16},{"name":14,"slug":15,"type":16},{"name":518,"slug":519,"type":16},"Triage","triage","2026-08-20T03:53:00.423941",27]