[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-braintrust-braintrust-define-eval-release-gate":3,"mdc--r8f71v-key":38,"related-repo-braintrust-braintrust-define-eval-release-gate":339,"related-org-braintrust-braintrust-define-eval-release-gate":436},{"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-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},"braintrust","Braintrust","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fbraintrust.png","braintrustdata",[13,17,20,23,26],{"name":14,"slug":15,"type":16},"LLM","llm","tag",{"name":18,"slug":19,"type":16},"Evals","evals",{"name":21,"slug":22,"type":16},"Agents","agents",{"name":24,"slug":25,"type":16},"CI\u002FCD","ci-cd",{"name":9,"slug":8,"type":16},7,"https:\u002F\u002Fgithub.com\u002Fbraintrustdata\u002Feval-library","2026-08-20T03:53:15.686158",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-define-eval-release-gate","---\nname: braintrust-define-eval-release-gate\ndescription: >-\n  Create, edit, audit, or apply release gates for LLM applications and agents. Use to combine\n  minimum meaningful improvement, statistical significance, regression rate, subgroup consistency,\n  worst-run stability, all-attempts reliability, safety upper bounds, latency, and cost into an\n  explicit ship-or-hold policy, to turn metrics into a CI gate, or to explain why a candidate\n  failed a gate and what evidence would justify reconsideration. Do not use for general result\n  analysis without a deployment decision.\n---\n\n# Define or apply a release gate\n\nContract: `references\u002Finteraction-contract.md`. Calibration, templates, provenance: `references\u002Fgate-rows.md`.\n\n## Trigger\n\n- \"Create release gates.\" \u002F \"Should this ship?\" \u002F \"Why did this candidate fail?\"\n- A ship decision being made from a single average.\n- A gate that exists but has no owner and blocks nothing.\n\n## Do\n\n1. Identify the shipping claim and the failures that cannot be traded away. Draft the table before\n   pinning thresholds needing product-owner input — mark those `Needs decision`.\n2. Build a row per property the claim implicitly makes, using the row set in `reference.md`:\n   magnitude, significance, consistency, stability, reliability, safety, latency\u002Fcost. A gate\n   should test **every** property, not just the average.\n3. Keep improvement metrics and guardrails structurally separate: guardrails are constraints,\n   never terms in a weighted sum.\n4. Require the run-integrity precondition — effective N and error accounting per arm. A gate on a\n   run with unexplained missing items is a gate-shaped artifact of whichever items survived.\n5. Enter red-team findings as **existence constraints**, never averaged scores, re-tested every\n   round rather than retired once passed.\n6. Emit both forms: machine-readable rules for CI, and a human-readable rationale per row. On\n   failure, name the failed rows, what each failure **means**, and the evidence that would justify\n   reconsideration.\n\n## Avoid\n\n- Do not let an average improvement buy down a safety or reliability constraint.\n- Do not gate on point estimates for near-zero rates; size n to the tolerance.\n- Do not gate on a scorer not validated for gating, or on the set the candidate was tuned against.\n- Do not write thresholds nobody owns; an unowned gate becomes a dashboard.\n- Do not compute the statistics here.\n\n## Check\n\n- Every row has a rule, a statistical basis, and a threshold with a stated source and owner.\n- Guardrails expressed as constraints; reliability and safety rows use bounds and worst cases.\n- Effective-N precondition explicit.\n- Machine-readable rules match the human-readable rationale exactly.\n- Failure interpretation written per row — each failure means something different.\n\n## Risk\n\n- Thresholds without uncertainty, or without operational ownership, turn gates into decoration.\n- A gate that passes says nothing about inputs the set does not cover; completeness is\n  unreachable, so gates pair with monitoring rather than replacing it.\n- Gates tuned until the current candidate passes are not gates.\n\n## Braintrust\n\n**Encode each row as a scorer-plus-threshold check on the confirmatory experiment and block\npromotion on any failing row.** A gate producing a chart instead of a block is not a gate. Reuse\nscorers by name and version (`references\u002Fplatform-mechanics.md` §5); a gate wired to a\ndifferently-named scorer silently stops testing what it was written for. Check each scorer's\nfitness statement first — where a scorer is prohibited from gating, **do not attach it at all**,\nsince absence beats a warning nobody reads.\n\nRows needing a derived value rather than a native score: the paired difference with clustered CI;\nper-category win rate and regression rate; the leave-one-out recomputation; `pass^k` across the K\nruns; the Wilson or rule-of-three **upper bound** for safety; p95 per run then the **worst** run.\n§8 lists the wiring failures these invite — the last two attract them most.\n\nGate against the experiment marked **confirmatory**, pinned to the held-out dataset version, on a\ncached pull.\n",{"data":39,"body":40},{"name":4,"description":6},{"type":41,"children":42},"root",[43,52,75,82,102,108,179,185,213,219,247,253,271,275,300,327],{"type":44,"tag":45,"props":46,"children":48},"element","h1",{"id":47},"define-or-apply-a-release-gate",[49],{"type":50,"value":51},"text","Define or apply a release gate",{"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\u002Fgate-rows.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},"\"Create release gates.\" \u002F \"Should this ship?\" \u002F \"Why did this candidate fail?\"",{"type":44,"tag":87,"props":93,"children":94},{},[95],{"type":50,"value":96},"A ship decision being made from a single average.",{"type":44,"tag":87,"props":98,"children":99},{},[100],{"type":50,"value":101},"A gate that exists but has no owner and blocks nothing.",{"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",{},[112,124,145,150,155,167],{"type":44,"tag":87,"props":113,"children":114},{},[115,117,123],{"type":50,"value":116},"Identify the shipping claim and the failures that cannot be traded away. Draft the table before\npinning thresholds needing product-owner input — mark those ",{"type":44,"tag":59,"props":118,"children":120},{"className":119},[],[121],{"type":50,"value":122},"Needs decision",{"type":50,"value":74},{"type":44,"tag":87,"props":125,"children":126},{},[127,129,135,137,143],{"type":50,"value":128},"Build a row per property the claim implicitly makes, using the row set in ",{"type":44,"tag":59,"props":130,"children":132},{"className":131},[],[133],{"type":50,"value":134},"reference.md",{"type":50,"value":136},":\nmagnitude, significance, consistency, stability, reliability, safety, latency\u002Fcost. A gate\nshould test ",{"type":44,"tag":138,"props":139,"children":140},"strong",{},[141],{"type":50,"value":142},"every",{"type":50,"value":144}," property, not just the average.",{"type":44,"tag":87,"props":146,"children":147},{},[148],{"type":50,"value":149},"Keep improvement metrics and guardrails structurally separate: guardrails are constraints,\nnever terms in a weighted sum.",{"type":44,"tag":87,"props":151,"children":152},{},[153],{"type":50,"value":154},"Require the run-integrity precondition — effective N and error accounting per arm. A gate on a\nrun with unexplained missing items is a gate-shaped artifact of whichever items survived.",{"type":44,"tag":87,"props":156,"children":157},{},[158,160,165],{"type":50,"value":159},"Enter red-team findings as ",{"type":44,"tag":138,"props":161,"children":162},{},[163],{"type":50,"value":164},"existence constraints",{"type":50,"value":166},", never averaged scores, re-tested every\nround rather than retired once passed.",{"type":44,"tag":87,"props":168,"children":169},{},[170,172,177],{"type":50,"value":171},"Emit both forms: machine-readable rules for CI, and a human-readable rationale per row. On\nfailure, name the failed rows, what each failure ",{"type":44,"tag":138,"props":173,"children":174},{},[175],{"type":50,"value":176},"means",{"type":50,"value":178},", and the evidence that would justify\nreconsideration.",{"type":44,"tag":76,"props":180,"children":182},{"id":181},"avoid",[183],{"type":50,"value":184},"Avoid",{"type":44,"tag":83,"props":186,"children":187},{},[188,193,198,203,208],{"type":44,"tag":87,"props":189,"children":190},{},[191],{"type":50,"value":192},"Do not let an average improvement buy down a safety or reliability constraint.",{"type":44,"tag":87,"props":194,"children":195},{},[196],{"type":50,"value":197},"Do not gate on point estimates for near-zero rates; size n to the tolerance.",{"type":44,"tag":87,"props":199,"children":200},{},[201],{"type":50,"value":202},"Do not gate on a scorer not validated for gating, or on the set the candidate was tuned against.",{"type":44,"tag":87,"props":204,"children":205},{},[206],{"type":50,"value":207},"Do not write thresholds nobody owns; an unowned gate becomes a dashboard.",{"type":44,"tag":87,"props":209,"children":210},{},[211],{"type":50,"value":212},"Do not compute the statistics here.",{"type":44,"tag":76,"props":214,"children":216},{"id":215},"check",[217],{"type":50,"value":218},"Check",{"type":44,"tag":83,"props":220,"children":221},{},[222,227,232,237,242],{"type":44,"tag":87,"props":223,"children":224},{},[225],{"type":50,"value":226},"Every row has a rule, a statistical basis, and a threshold with a stated source and owner.",{"type":44,"tag":87,"props":228,"children":229},{},[230],{"type":50,"value":231},"Guardrails expressed as constraints; reliability and safety rows use bounds and worst cases.",{"type":44,"tag":87,"props":233,"children":234},{},[235],{"type":50,"value":236},"Effective-N precondition explicit.",{"type":44,"tag":87,"props":238,"children":239},{},[240],{"type":50,"value":241},"Machine-readable rules match the human-readable rationale exactly.",{"type":44,"tag":87,"props":243,"children":244},{},[245],{"type":50,"value":246},"Failure interpretation written per row — each failure means something different.",{"type":44,"tag":76,"props":248,"children":250},{"id":249},"risk",[251],{"type":50,"value":252},"Risk",{"type":44,"tag":83,"props":254,"children":255},{},[256,261,266],{"type":44,"tag":87,"props":257,"children":258},{},[259],{"type":50,"value":260},"Thresholds without uncertainty, or without operational ownership, turn gates into decoration.",{"type":44,"tag":87,"props":262,"children":263},{},[264],{"type":50,"value":265},"A gate that passes says nothing about inputs the set does not cover; completeness is\nunreachable, so gates pair with monitoring rather than replacing it.",{"type":44,"tag":87,"props":267,"children":268},{},[269],{"type":50,"value":270},"Gates tuned until the current candidate passes are not gates.",{"type":44,"tag":76,"props":272,"children":273},{"id":8},[274],{"type":50,"value":9},{"type":44,"tag":53,"props":276,"children":277},{},[278,283,285,291,293,298],{"type":44,"tag":138,"props":279,"children":280},{},[281],{"type":50,"value":282},"Encode each row as a scorer-plus-threshold check on the confirmatory experiment and block\npromotion on any failing row.",{"type":50,"value":284}," A gate producing a chart instead of a block is not a gate. Reuse\nscorers by name and version (",{"type":44,"tag":59,"props":286,"children":288},{"className":287},[],[289],{"type":50,"value":290},"references\u002Fplatform-mechanics.md",{"type":50,"value":292}," §5); a gate wired to a\ndifferently-named scorer silently stops testing what it was written for. Check each scorer's\nfitness statement first — where a scorer is prohibited from gating, ",{"type":44,"tag":138,"props":294,"children":295},{},[296],{"type":50,"value":297},"do not attach it at all",{"type":50,"value":299},",\nsince absence beats a warning nobody reads.",{"type":44,"tag":53,"props":301,"children":302},{},[303,305,311,313,318,320,325],{"type":50,"value":304},"Rows needing a derived value rather than a native score: the paired difference with clustered CI;\nper-category win rate and regression rate; the leave-one-out recomputation; ",{"type":44,"tag":59,"props":306,"children":308},{"className":307},[],[309],{"type":50,"value":310},"pass^k",{"type":50,"value":312}," across the K\nruns; the Wilson or rule-of-three ",{"type":44,"tag":138,"props":314,"children":315},{},[316],{"type":50,"value":317},"upper bound",{"type":50,"value":319}," for safety; p95 per run then the ",{"type":44,"tag":138,"props":321,"children":322},{},[323],{"type":50,"value":324},"worst",{"type":50,"value":326}," run.\n§8 lists the wiring failures these invite — the last two attract them most.",{"type":44,"tag":53,"props":328,"children":329},{},[330,332,337],{"type":50,"value":331},"Gate against the experiment marked ",{"type":44,"tag":138,"props":333,"children":334},{},[335],{"type":50,"value":336},"confirmatory",{"type":50,"value":338},", pinned to the held-out dataset version, on a\ncached pull.",{"items":340,"total":435},[341,357,371,385,401,409,421],{"slug":342,"name":342,"fn":343,"description":344,"org":345,"tags":346,"stars":27,"repoUrl":28,"updatedAt":356},"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},[347,350,351,352,353],{"name":348,"slug":349,"type":16},"Analysis","analysis",{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":14,"slug":15,"type":16},{"name":354,"slug":355,"type":16},"Statistics","statistics","2026-08-20T03:53:01.13806",{"slug":358,"name":358,"fn":359,"description":360,"org":361,"tags":362,"stars":27,"repoUrl":28,"updatedAt":370},"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},[363,364,365,366,367],{"name":348,"slug":349,"type":16},{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":14,"slug":15,"type":16},{"name":368,"slug":369,"type":16},"Performance","performance","2026-08-20T03:53:40.036077",{"slug":372,"name":372,"fn":373,"description":374,"org":375,"tags":376,"stars":27,"repoUrl":28,"updatedAt":384},"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},[377,378,379,382,383],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":380,"slug":381,"type":16},"Datasets","datasets",{"name":18,"slug":19,"type":16},{"name":14,"slug":15,"type":16},"2026-08-20T03:53:33.304815",{"slug":386,"name":386,"fn":387,"description":388,"org":389,"tags":390,"stars":27,"repoUrl":28,"updatedAt":400},"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},[391,392,393,394,397],{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":14,"slug":15,"type":16},{"name":395,"slug":396,"type":16},"Product Management","product-management",{"name":398,"slug":399,"type":16},"Strategy","strategy","2026-08-20T03:53:00.07097",{"slug":4,"name":4,"fn":5,"description":6,"org":402,"tags":403,"stars":27,"repoUrl":28,"updatedAt":29},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[404,405,406,407,408],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":24,"slug":25,"type":16},{"name":18,"slug":19,"type":16},{"name":14,"slug":15,"type":16},{"slug":410,"name":410,"fn":411,"description":412,"org":413,"tags":414,"stars":27,"repoUrl":28,"updatedAt":420},"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},[415,416,419],{"name":9,"slug":8,"type":16},{"name":417,"slug":418,"type":16},"Deployment","deployment",{"name":18,"slug":19,"type":16},"2026-08-20T03:53:32.558937",{"slug":422,"name":422,"fn":423,"description":424,"org":425,"tags":426,"stars":27,"repoUrl":28,"updatedAt":434},"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},[427,428,429,432,433],{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":430,"slug":431,"type":16},"Experiments","experiments",{"name":14,"slug":15,"type":16},{"name":398,"slug":399,"type":16},"2026-08-20T03:53:36.534554",24,{"items":437,"total":562},[438,454,462,470,478,486,494,500,508,524,536,548],{"slug":439,"name":439,"fn":440,"description":441,"org":442,"tags":443,"stars":451,"repoUrl":452,"updatedAt":453},"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},[444,445,448],{"name":9,"slug":8,"type":16},{"name":446,"slug":447,"type":16},"Debugging","debugging",{"name":449,"slug":450,"type":16},"MCP","mcp",18,"https:\u002F\u002Fgithub.com\u002Fbraintrustdata\u002Fbraintrust-claude-plugin","2026-07-12T08:36:13.889274",{"slug":342,"name":342,"fn":343,"description":344,"org":455,"tags":456,"stars":27,"repoUrl":28,"updatedAt":356},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[457,458,459,460,461],{"name":348,"slug":349,"type":16},{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":14,"slug":15,"type":16},{"name":354,"slug":355,"type":16},{"slug":358,"name":358,"fn":359,"description":360,"org":463,"tags":464,"stars":27,"repoUrl":28,"updatedAt":370},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[465,466,467,468,469],{"name":348,"slug":349,"type":16},{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":14,"slug":15,"type":16},{"name":368,"slug":369,"type":16},{"slug":372,"name":372,"fn":373,"description":374,"org":471,"tags":472,"stars":27,"repoUrl":28,"updatedAt":384},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[473,474,475,476,477],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":380,"slug":381,"type":16},{"name":18,"slug":19,"type":16},{"name":14,"slug":15,"type":16},{"slug":386,"name":386,"fn":387,"description":388,"org":479,"tags":480,"stars":27,"repoUrl":28,"updatedAt":400},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[481,482,483,484,485],{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":14,"slug":15,"type":16},{"name":395,"slug":396,"type":16},{"name":398,"slug":399,"type":16},{"slug":4,"name":4,"fn":5,"description":6,"org":487,"tags":488,"stars":27,"repoUrl":28,"updatedAt":29},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[489,490,491,492,493],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":24,"slug":25,"type":16},{"name":18,"slug":19,"type":16},{"name":14,"slug":15,"type":16},{"slug":410,"name":410,"fn":411,"description":412,"org":495,"tags":496,"stars":27,"repoUrl":28,"updatedAt":420},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[497,498,499],{"name":9,"slug":8,"type":16},{"name":417,"slug":418,"type":16},{"name":18,"slug":19,"type":16},{"slug":422,"name":422,"fn":423,"description":424,"org":501,"tags":502,"stars":27,"repoUrl":28,"updatedAt":434},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[503,504,505,506,507],{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":430,"slug":431,"type":16},{"name":14,"slug":15,"type":16},{"name":398,"slug":399,"type":16},{"slug":509,"name":509,"fn":510,"description":511,"org":512,"tags":513,"stars":27,"repoUrl":28,"updatedAt":523},"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},[514,515,516,517,520],{"name":9,"slug":8,"type":16},{"name":380,"slug":381,"type":16},{"name":18,"slug":19,"type":16},{"name":518,"slug":519,"type":16},"Observability","observability",{"name":521,"slug":522,"type":16},"Tracing","tracing","2026-08-20T03:53:37.274703",{"slug":525,"name":525,"fn":526,"description":527,"org":528,"tags":529,"stars":27,"repoUrl":28,"updatedAt":535},"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},[530,531,532,533,534],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":18,"slug":19,"type":16},{"name":14,"slug":15,"type":16},{"name":368,"slug":369,"type":16},"2026-08-20T03:53:16.078609",{"slug":537,"name":537,"fn":538,"description":539,"org":540,"tags":541,"stars":27,"repoUrl":28,"updatedAt":547},"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},[542,543,544,545,546],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":380,"slug":381,"type":16},{"name":18,"slug":19,"type":16},{"name":14,"slug":15,"type":16},"2026-08-20T03:53:36.900129",{"slug":549,"name":549,"fn":550,"description":551,"org":552,"tags":553,"stars":27,"repoUrl":28,"updatedAt":561},"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},[554,555,556,557,558],{"name":21,"slug":22,"type":16},{"name":9,"slug":8,"type":16},{"name":446,"slug":447,"type":16},{"name":18,"slug":19,"type":16},{"name":559,"slug":560,"type":16},"Triage","triage","2026-08-20T03:53:00.423941",27]