[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-braintrust-braintrust-map-eval-evidence":3,"mdc-avysph-key":32,"related-repo-braintrust-braintrust-map-eval-evidence":312,"related-org-braintrust-braintrust-map-eval-evidence":417},{"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-map-eval-evidence","create and audit evaluation evidence maps","Create or audit an evidence map that connects eval constructs or behavior specifications to observable success signals, failure signals, proxy limitations, and possible gaming paths. Use when a quality such as helpfulness, safety, trust, correctness, or task success is not directly measurable, when asking \"how would we observe this,\" \"what evidence would show this behavior happened,\" or when a BEHAVIOR.md must be translated into trace-review signals. Do not use to define storage or schema fields, choose product metrics, or write scorers.",{"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},"Analysis","analysis","tag",{"name":18,"slug":19,"type":16},"Evals","evals",{"name":9,"slug":8,"type":16},7,"https:\u002F\u002Fgithub.com\u002Fbraintrustdata\u002Feval-library","2026-08-20T03:52:56.298223",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-map-eval-evidence","---\nname: braintrust-map-eval-evidence\ndescription: >-\n  Create or audit an evidence map that connects eval constructs or behavior specifications to\n  observable success signals, failure signals, proxy limitations, and possible gaming paths.\n  Use when a quality such as helpfulness, safety, trust, correctness, or task success is not\n  directly measurable, when asking \"how would we observe this,\" \"what evidence would show this\n  behavior happened,\" or when a BEHAVIOR.md must be translated into trace-review signals. Do\n  not use to define storage or schema fields, choose product metrics, or write scorers.\n---\n\n# Map constructs to observable evidence\n\nContract: `references\u002Finteraction-contract.md`. Calibration, templates, provenance: `references\u002Fevidence-map.md`.\n\n## Trigger\n\n- A construct appearing in no trace field: helpfulness, trustworthiness, safety.\n- A behavior spec's review questions needing observable signals.\n- Scorers that exist but nobody can say what they are evidence *of*.\n\n## Do\n\n1. For each construct, fill four columns before asking anything: **success evidence**,\n   **failure evidence**, **limitation** (what this signal misses), **gaming path** (how to\n   raise it without improving the outcome).\n2. Require more than one proxy for any consequential outcome — several independent proxies make\n   it harder to win the eval without improving the product.\n3. Apply the usefulness test: closely connected to the outcome, covers its most important\n   dimensions, hard to improve **without** improving real behavior.\n4. Surface the most consequential blind spot explicitly and label it `Not yet measurable`\n   rather than substituting a convenient proxy.\n5. Mark evidence the system cannot currently produce. That list is what the trace schema has to\n   be extended to cover.\n\n## Avoid\n\n- Do not invent field names or schemas; that is the next skill.\n- Do not write scoring logic or thresholds.\n- Do not treat chain-of-thought or a self-reported rationale as behavioral evidence — it is a\n  claim about behavior, not a record of it.\n- Do not stop at success evidence. Failure evidence is what makes a scorer discriminate.\n\n## Check\n\n- Every construct has success evidence, failure evidence, ≥1 limitation, ≥1 gaming path.\n- Consequential constructs carry multiple independent signals.\n- Blind spots stated rather than papered over; uncapturable evidence flagged.\n\n## Risk\n\n- Weak proxies look precise. Latency, token count, and polish are easy to measure and routinely\n  stand in for outcomes they do not track.\n- Process-compliance evidence (\"called the right tool\") can diverge entirely from outcome\n  evidence (\"solved the user's problem\"); mapping only one produces a confident, wrong\n  conclusion.\n- Proxies chosen from what is already logged encode last year's instrumentation as this year's\n  construct.\n\n## Braintrust\n\nEach signal must be locatable in a span. Signals about **what the agent did** come from\ntool-call spans, so they exist only if the agent is instrumented at that granularity. Signals\nabout the output come from the top-level span. Signals you want to **slice by** are `metadata`,\nnot evidence — do not conflate them: evidence is what you score, metadata is how you group\nscores. **One signal, one scorer** (`references\u002Fplatform-mechanics.md` §5), so the multi-proxy\nrequirement shows up as multiple columns; two signals in one scorer breaks the gaming-path\nanalysis because you can no longer see which proxy moved. A `Not yet measurable` blind spot vanishes once numbers start\nappearing — record it in the dataset or experiment description, and where it matters to a gate,\nprefer a documented unimplemented placeholder over omitting the dimension and letting its\nabsence read as a pass.\n",{"data":33,"body":34},{"name":4,"description":6},{"type":35,"children":36},"root",[37,46,69,76,103,109,182,188,211,217,235,241,259,263],{"type":38,"tag":39,"props":40,"children":42},"element","h1",{"id":41},"map-constructs-to-observable-evidence",[43],{"type":44,"value":45},"text","Map constructs to observable evidence",{"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\u002Fevidence-map.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},"A construct appearing in no trace field: helpfulness, trustworthiness, safety.",{"type":38,"tag":81,"props":87,"children":88},{},[89],{"type":44,"value":90},"A behavior spec's review questions needing observable signals.",{"type":38,"tag":81,"props":92,"children":93},{},[94,96,102],{"type":44,"value":95},"Scorers that exist but nobody can say what they are evidence ",{"type":38,"tag":97,"props":98,"children":99},"em",{},[100],{"type":44,"value":101},"of",{"type":44,"value":68},{"type":38,"tag":70,"props":104,"children":106},{"id":105},"do",[107],{"type":44,"value":108},"Do",{"type":38,"tag":110,"props":111,"children":112},"ol",{},[113,147,152,164,177],{"type":38,"tag":81,"props":114,"children":115},{},[116,118,124,126,131,133,138,140,145],{"type":44,"value":117},"For each construct, fill four columns before asking anything: ",{"type":38,"tag":119,"props":120,"children":121},"strong",{},[122],{"type":44,"value":123},"success evidence",{"type":44,"value":125},",\n",{"type":38,"tag":119,"props":127,"children":128},{},[129],{"type":44,"value":130},"failure evidence",{"type":44,"value":132},", ",{"type":38,"tag":119,"props":134,"children":135},{},[136],{"type":44,"value":137},"limitation",{"type":44,"value":139}," (what this signal misses), ",{"type":38,"tag":119,"props":141,"children":142},{},[143],{"type":44,"value":144},"gaming path",{"type":44,"value":146}," (how to\nraise it without improving the outcome).",{"type":38,"tag":81,"props":148,"children":149},{},[150],{"type":44,"value":151},"Require more than one proxy for any consequential outcome — several independent proxies make\nit harder to win the eval without improving the product.",{"type":38,"tag":81,"props":153,"children":154},{},[155,157,162],{"type":44,"value":156},"Apply the usefulness test: closely connected to the outcome, covers its most important\ndimensions, hard to improve ",{"type":38,"tag":119,"props":158,"children":159},{},[160],{"type":44,"value":161},"without",{"type":44,"value":163}," improving real behavior.",{"type":38,"tag":81,"props":165,"children":166},{},[167,169,175],{"type":44,"value":168},"Surface the most consequential blind spot explicitly and label it ",{"type":38,"tag":53,"props":170,"children":172},{"className":171},[],[173],{"type":44,"value":174},"Not yet measurable",{"type":44,"value":176},"\nrather than substituting a convenient proxy.",{"type":38,"tag":81,"props":178,"children":179},{},[180],{"type":44,"value":181},"Mark evidence the system cannot currently produce. That list is what the trace schema has to\nbe extended to cover.",{"type":38,"tag":70,"props":183,"children":185},{"id":184},"avoid",[186],{"type":44,"value":187},"Avoid",{"type":38,"tag":77,"props":189,"children":190},{},[191,196,201,206],{"type":38,"tag":81,"props":192,"children":193},{},[194],{"type":44,"value":195},"Do not invent field names or schemas; that is the next skill.",{"type":38,"tag":81,"props":197,"children":198},{},[199],{"type":44,"value":200},"Do not write scoring logic or thresholds.",{"type":38,"tag":81,"props":202,"children":203},{},[204],{"type":44,"value":205},"Do not treat chain-of-thought or a self-reported rationale as behavioral evidence — it is a\nclaim about behavior, not a record of it.",{"type":38,"tag":81,"props":207,"children":208},{},[209],{"type":44,"value":210},"Do not stop at success evidence. Failure evidence is what makes a scorer discriminate.",{"type":38,"tag":70,"props":212,"children":214},{"id":213},"check",[215],{"type":44,"value":216},"Check",{"type":38,"tag":77,"props":218,"children":219},{},[220,225,230],{"type":38,"tag":81,"props":221,"children":222},{},[223],{"type":44,"value":224},"Every construct has success evidence, failure evidence, ≥1 limitation, ≥1 gaming path.",{"type":38,"tag":81,"props":226,"children":227},{},[228],{"type":44,"value":229},"Consequential constructs carry multiple independent signals.",{"type":38,"tag":81,"props":231,"children":232},{},[233],{"type":44,"value":234},"Blind spots stated rather than papered over; uncapturable evidence flagged.",{"type":38,"tag":70,"props":236,"children":238},{"id":237},"risk",[239],{"type":44,"value":240},"Risk",{"type":38,"tag":77,"props":242,"children":243},{},[244,249,254],{"type":38,"tag":81,"props":245,"children":246},{},[247],{"type":44,"value":248},"Weak proxies look precise. Latency, token count, and polish are easy to measure and routinely\nstand in for outcomes they do not track.",{"type":38,"tag":81,"props":250,"children":251},{},[252],{"type":44,"value":253},"Process-compliance evidence (\"called the right tool\") can diverge entirely from outcome\nevidence (\"solved the user's problem\"); mapping only one produces a confident, wrong\nconclusion.",{"type":38,"tag":81,"props":255,"children":256},{},[257],{"type":44,"value":258},"Proxies chosen from what is already logged encode last year's instrumentation as this year's\nconstruct.",{"type":38,"tag":70,"props":260,"children":261},{"id":8},[262],{"type":44,"value":9},{"type":38,"tag":47,"props":264,"children":265},{},[266,268,273,275,280,282,288,290,295,297,303,305,310],{"type":44,"value":267},"Each signal must be locatable in a span. Signals about ",{"type":38,"tag":119,"props":269,"children":270},{},[271],{"type":44,"value":272},"what the agent did",{"type":44,"value":274}," come from\ntool-call spans, so they exist only if the agent is instrumented at that granularity. Signals\nabout the output come from the top-level span. Signals you want to ",{"type":38,"tag":119,"props":276,"children":277},{},[278],{"type":44,"value":279},"slice by",{"type":44,"value":281}," are ",{"type":38,"tag":53,"props":283,"children":285},{"className":284},[],[286],{"type":44,"value":287},"metadata",{"type":44,"value":289},",\nnot evidence — do not conflate them: evidence is what you score, metadata is how you group\nscores. ",{"type":38,"tag":119,"props":291,"children":292},{},[293],{"type":44,"value":294},"One signal, one scorer",{"type":44,"value":296}," (",{"type":38,"tag":53,"props":298,"children":300},{"className":299},[],[301],{"type":44,"value":302},"references\u002Fplatform-mechanics.md",{"type":44,"value":304}," §5), so the multi-proxy\nrequirement shows up as multiple columns; two signals in one scorer breaks the gaming-path\nanalysis because you can no longer see which proxy moved. A ",{"type":38,"tag":53,"props":306,"children":308},{"className":307},[],[309],{"type":44,"value":174},{"type":44,"value":311}," blind spot vanishes once numbers start\nappearing — record it in the dataset or experiment description, and where it matters to a gate,\nprefer a documented unimplemented placeholder over omitting the dimension and letting its\nabsence read as a pass.",{"items":313,"total":416},[314,330,344,360,376,390,402],{"slug":315,"name":315,"fn":316,"description":317,"org":318,"tags":319,"stars":21,"repoUrl":22,"updatedAt":329},"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},[320,321,322,323,326],{"name":14,"slug":15,"type":16},{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":324,"slug":325,"type":16},"LLM","llm",{"name":327,"slug":328,"type":16},"Statistics","statistics","2026-08-20T03:53:01.13806",{"slug":331,"name":331,"fn":332,"description":333,"org":334,"tags":335,"stars":21,"repoUrl":22,"updatedAt":343},"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},[336,337,338,339,340],{"name":14,"slug":15,"type":16},{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":324,"slug":325,"type":16},{"name":341,"slug":342,"type":16},"Performance","performance","2026-08-20T03:53:40.036077",{"slug":345,"name":345,"fn":346,"description":347,"org":348,"tags":349,"stars":21,"repoUrl":22,"updatedAt":359},"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},[350,353,354,357,358],{"name":351,"slug":352,"type":16},"Agents","agents",{"name":9,"slug":8,"type":16},{"name":355,"slug":356,"type":16},"Datasets","datasets",{"name":18,"slug":19,"type":16},{"name":324,"slug":325,"type":16},"2026-08-20T03:53:33.304815",{"slug":361,"name":361,"fn":362,"description":363,"org":364,"tags":365,"stars":21,"repoUrl":22,"updatedAt":375},"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},[366,367,368,369,372],{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":324,"slug":325,"type":16},{"name":370,"slug":371,"type":16},"Product Management","product-management",{"name":373,"slug":374,"type":16},"Strategy","strategy","2026-08-20T03:53:00.07097",{"slug":377,"name":377,"fn":378,"description":379,"org":380,"tags":381,"stars":21,"repoUrl":22,"updatedAt":389},"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},[382,383,384,387,388],{"name":351,"slug":352,"type":16},{"name":9,"slug":8,"type":16},{"name":385,"slug":386,"type":16},"CI\u002FCD","ci-cd",{"name":18,"slug":19,"type":16},{"name":324,"slug":325,"type":16},"2026-08-20T03:53:15.686158",{"slug":391,"name":391,"fn":392,"description":393,"org":394,"tags":395,"stars":21,"repoUrl":22,"updatedAt":401},"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},[396,397,400],{"name":9,"slug":8,"type":16},{"name":398,"slug":399,"type":16},"Deployment","deployment",{"name":18,"slug":19,"type":16},"2026-08-20T03:53:32.558937",{"slug":403,"name":403,"fn":404,"description":405,"org":406,"tags":407,"stars":21,"repoUrl":22,"updatedAt":415},"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},[408,409,410,413,414],{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":411,"slug":412,"type":16},"Experiments","experiments",{"name":324,"slug":325,"type":16},{"name":373,"slug":374,"type":16},"2026-08-20T03:53:36.534554",24,{"items":418,"total":543},[419,435,443,451,459,467,475,481,489,505,517,529],{"slug":420,"name":420,"fn":421,"description":422,"org":423,"tags":424,"stars":432,"repoUrl":433,"updatedAt":434},"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},[425,426,429],{"name":9,"slug":8,"type":16},{"name":427,"slug":428,"type":16},"Debugging","debugging",{"name":430,"slug":431,"type":16},"MCP","mcp",18,"https:\u002F\u002Fgithub.com\u002Fbraintrustdata\u002Fbraintrust-claude-plugin","2026-07-12T08:36:13.889274",{"slug":315,"name":315,"fn":316,"description":317,"org":436,"tags":437,"stars":21,"repoUrl":22,"updatedAt":329},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[438,439,440,441,442],{"name":14,"slug":15,"type":16},{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":324,"slug":325,"type":16},{"name":327,"slug":328,"type":16},{"slug":331,"name":331,"fn":332,"description":333,"org":444,"tags":445,"stars":21,"repoUrl":22,"updatedAt":343},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[446,447,448,449,450],{"name":14,"slug":15,"type":16},{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":324,"slug":325,"type":16},{"name":341,"slug":342,"type":16},{"slug":345,"name":345,"fn":346,"description":347,"org":452,"tags":453,"stars":21,"repoUrl":22,"updatedAt":359},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[454,455,456,457,458],{"name":351,"slug":352,"type":16},{"name":9,"slug":8,"type":16},{"name":355,"slug":356,"type":16},{"name":18,"slug":19,"type":16},{"name":324,"slug":325,"type":16},{"slug":361,"name":361,"fn":362,"description":363,"org":460,"tags":461,"stars":21,"repoUrl":22,"updatedAt":375},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[462,463,464,465,466],{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":324,"slug":325,"type":16},{"name":370,"slug":371,"type":16},{"name":373,"slug":374,"type":16},{"slug":377,"name":377,"fn":378,"description":379,"org":468,"tags":469,"stars":21,"repoUrl":22,"updatedAt":389},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[470,471,472,473,474],{"name":351,"slug":352,"type":16},{"name":9,"slug":8,"type":16},{"name":385,"slug":386,"type":16},{"name":18,"slug":19,"type":16},{"name":324,"slug":325,"type":16},{"slug":391,"name":391,"fn":392,"description":393,"org":476,"tags":477,"stars":21,"repoUrl":22,"updatedAt":401},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[478,479,480],{"name":9,"slug":8,"type":16},{"name":398,"slug":399,"type":16},{"name":18,"slug":19,"type":16},{"slug":403,"name":403,"fn":404,"description":405,"org":482,"tags":483,"stars":21,"repoUrl":22,"updatedAt":415},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[484,485,486,487,488],{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":411,"slug":412,"type":16},{"name":324,"slug":325,"type":16},{"name":373,"slug":374,"type":16},{"slug":490,"name":490,"fn":491,"description":492,"org":493,"tags":494,"stars":21,"repoUrl":22,"updatedAt":504},"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},[495,496,497,498,501],{"name":9,"slug":8,"type":16},{"name":355,"slug":356,"type":16},{"name":18,"slug":19,"type":16},{"name":499,"slug":500,"type":16},"Observability","observability",{"name":502,"slug":503,"type":16},"Tracing","tracing","2026-08-20T03:53:37.274703",{"slug":506,"name":506,"fn":507,"description":508,"org":509,"tags":510,"stars":21,"repoUrl":22,"updatedAt":516},"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},[511,512,513,514,515],{"name":351,"slug":352,"type":16},{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":324,"slug":325,"type":16},{"name":341,"slug":342,"type":16},"2026-08-20T03:53:16.078609",{"slug":518,"name":518,"fn":519,"description":520,"org":521,"tags":522,"stars":21,"repoUrl":22,"updatedAt":528},"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},[523,524,525,526,527],{"name":351,"slug":352,"type":16},{"name":9,"slug":8,"type":16},{"name":355,"slug":356,"type":16},{"name":18,"slug":19,"type":16},{"name":324,"slug":325,"type":16},"2026-08-20T03:53:36.900129",{"slug":530,"name":530,"fn":531,"description":532,"org":533,"tags":534,"stars":21,"repoUrl":22,"updatedAt":542},"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},[535,536,537,538,539],{"name":351,"slug":352,"type":16},{"name":9,"slug":8,"type":16},{"name":427,"slug":428,"type":16},{"name":18,"slug":19,"type":16},{"name":540,"slug":541,"type":16},"Triage","triage","2026-08-20T03:53:00.423941",27]