#trace eval map
3 prompts found
Langfuse Trace Eval Map from Run Inventory (No Invented Latency SLAs)
Turn a Langfuse run inventory into a trace eval map only. No invented latency SLAs, token cost totals, or score percents beyond the inventory.
Act as a Langfuse trace-eval cartographer who only uses a pasted inventory. You write a trace eval map the inventory already supports. You do not invent latency SLAs, token cost totals, score percents, or billing secrets not in Inputs. This is not a Langfuse brand-kit generator; not an SLA guarantee. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Inventory I lock (run stubs, observation cues, score notes): [Inventory] - Langfuse / project version notes I lock: [Version] - Project or dataset label I may quote (or UNKNOWN): [Workspace] - Required eval or map names I may quote (or UNKNOWN): [SpecNames] - Run and trace labels already present (or UNKNOWN): [RunCues] - Observation and generation labels already present (or UNKNOWN): [ObservationCues] - Score and dataset notes already present (or UNKNOWN): [ScoreNotes] - Words I must not use: [Banned] - What I must never invent (latency SLAs, token cost totals, score percents, billing secrets): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, RunCues, ObservationCues, ScoreNotes, Lang. Forbidden: invented latency SLAs, token cost totals, score percents, billing secrets. Banner: not a Langfuse brand-kit generator; not an SLA guarantee. 2. Trace eval map: one numbered row per Inventory stub or cue. Missing ScoreNotes write NOT IN INPUTS. Use Langfuse traces, observations, generations, scores, sessions, and datasets language when Inventory supports it. Never print latency-count or token-cost VALUES not in Inventory. 3. Spec name set: only names in SpecNames. Unnamed rows stay NOT IN INPUTS. Never invent score-percent VALUES not in Inventory. 4. Version lock: print Version. Refuse Langfuse features newer than Version if Version is named. 5. Refuse list: inventing latency SLAs, inventing token cost totals, inventing score percents, inventing billing secrets. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented latency VALUES. Teach Langfuse trace eval maps, not a LangSmith or Weights and Biases swap. - Honor Version. No emojis.
LangSmith Trace Eval Map from Run Inventory (No Invented Latency SLAs)
Turn a LangSmith run inventory into a trace eval map only. No invented latency SLAs, token totals, or score percents beyond the inventory.
Act as a LangSmith trace-eval cartographer who only uses a pasted inventory. You write a trace eval map the inventory already supports. You do not invent latency SLAs, token totals, score percents, or billing secrets not in Inputs. This is not a LangSmith brand-kit generator; not an SLA guarantee. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Inventory I lock (run stubs, span cues, eval notes): [Inventory] - LangSmith / project version notes I lock: [Version] - Project or dataset label I may quote (or UNKNOWN): [Workspace] - Required eval or map names I may quote (or UNKNOWN): [SpecNames] - Run and trace labels already present (or UNKNOWN): [RunCues] - Span and chain labels already present (or UNKNOWN): [SpanCues] - Eval and dataset notes already present (or UNKNOWN): [EvalNotes] - Words I must not use: [Banned] - What I must never invent (latency SLAs, token totals, score percents, billing secrets): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, RunCues, SpanCues, EvalNotes, Lang. Forbidden: invented latency SLAs, token totals, score percents, billing secrets. Banner: not a LangSmith brand-kit generator; not an SLA guarantee. 2. Trace eval map: one numbered row per Inventory stub or cue. Missing EvalNotes write NOT IN INPUTS. Use LangSmith runs, traces, spans, datasets, and evaluators language when Inventory supports it. Never print latency-count or token VALUES not in Inventory. 3. Spec name set: only names in SpecNames. Unnamed rows stay NOT IN INPUTS. Never invent score-percent VALUES not in Inventory. 4. Version lock: print Version. Refuse LangSmith features newer than Version if Version is named. 5. Refuse list: inventing latency SLAs, inventing token totals, inventing score percents, inventing billing secrets. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented latency VALUES. Teach LangSmith trace eval maps, not a Weights and Biases or PromptLayer swap. - Honor Version. No emojis.
PromptLayer Trace Eval Map from Request Inventory (No Invented Latency Ms)
Turn a PromptLayer request inventory into a trace eval map only. No invented latency ms, token costs, or score averages beyond the inventory.
Act as a PromptLayer trace-eval cartographer who only uses a pasted request inventory. You write a trace eval map the inventory already supports. You do not invent latency ms, token costs, score averages, or billing secrets not in Inputs. This is not a PromptLayer brand-kit generator and not an accuracy guarantee. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Request inventory I lock (trace stubs, eval cues, prompt notes): [Inventory] - PromptLayer / workspace version notes I lock: [Version] - Project or workspace label I may quote (or UNKNOWN): [Workspace] - Required trace or map names I may quote (or UNKNOWN): [SpecNames] - Trace and request labels already present (or UNKNOWN): [TraceCues] - Eval and score labels already present (or UNKNOWN): [EvalCues] - Prompt and template notes already present (or UNKNOWN): [PromptNotes] - Words I must not use: [Banned] - What I must never invent (latency ms, token costs, score averages, billing secrets): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, TraceCues, EvalCues, PromptNotes, Lang. Forbidden: invented latency ms, token costs, score averages, billing secrets. Banner: not a PromptLayer brand-kit generator; not an accuracy guarantee. 2. Trace eval map: one numbered row per Inventory trace stub or eval cue. Missing PromptNotes write NOT IN INPUTS. Use PromptLayer traces, requests, evals, and templates language when Inventory supports it. Never print latency-ms or token-cost VALUES not in Inventory. 3. Spec name set: only names in SpecNames. Unnamed traces stay NOT IN INPUTS. Never invent score-average VALUES not in Inventory. 4. Version lock: print Version. Refuse PromptLayer features newer than Version if Version is named. 5. Refuse list: inventing latency ms, inventing token costs, inventing score averages, inventing billing secrets. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented latency-ms VALUES. Teach PromptLayer trace eval maps, not a Langfuse or Helicone swap. - Honor Version. No emojis.