OpenPipe Fine-Tune Dataset Row Checklist from Project Notes (No Invented Eval Scores)

Prompt

Act as a OpenPipe fine-tune dataset-row checklist engineer who only uses pasted notes. You compile a fine-tune dataset-row checklist the notes already support. You do not invent eval scores, loss ranks, training scoreboards, accuracy promises. This is not a live OpenPipe sync, not LangSmith merge, and not ML production readiness advice.
You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs.

Inputs:
- Notes I lock (dataset stubs, row cues, split fragments): [ProjectNotes]
- OpenPipe version or project notes I lock: [Version]
- Project or dataset label I may quote (or UNKNOWN): [ProjectLabel]
- Dataset names already present (or UNKNOWN): [DatasetNames]
- Row cues already present (or UNKNOWN): [RowCues]
- Split cues already present (or UNKNOWN): [SplitCues]
- Model cues already present (or UNKNOWN): [ModelCues]
- Words I must not use: [Banned]
- What I must never invent (eval scores, loss ranks, training scoreboards, accuracy promises): [Never]
- Output format: [Format]
- Language: [Lang]

Generate:
1. Honesty ledger: ProjectNotes nouns, Version, ProjectLabel, DatasetNames, RowCues, SplitCues, ModelCues, Lang. Banner: not ML production readiness advice; not a live OpenPipe sync. Forbidden: invented eval scores, loss ranks, training scoreboards, accuracy promises.
2. Fine-tune dataset-row checklist: one checkbox row per DatasetNames entry. Attach only RowCues named beside that entry in ProjectNotes. Missing row write NOT IN INPUTS.
3. Split sketch: for each SplitCues entry, list rows that name it. Do not invent a 0.91 eval score claim if absent.
4. Model caution block: quote ModelCues only. Extra packs not in ProjectNotes stay NOT IN INPUTS.
5. Refuse list: inventing 0.91 eval score values, inventing loss ranks, inventing training scoreboards, inventing accuracy promises.
6. Compliance pass: quote Banned and Never hits. Cut them. Print counts from ProjectNotes only. Format as Format.

Constraints:
- Fine-tune dataset-row checklist from ProjectNotes only. No invented eval scores.
- Honor Version. No emojis. Not a live OpenPipe dashboard. Not ml production readiness advice.

Instructions

Replace every [bracket] with your details before running. Works on ChatGPT, Claude, and Gemini.

Generated Output

This image was generated using the prompt above.

OpenPipe Fine-Tune Dataset Row Checklist from Project Notes (No Invented Eval Scores) - Result

Examples

Example Input

ProjectNotes: dataset name Harbor Quay Support FT Set as pasted row cue Row Pier Refund Pair as pasted split cue Split Train Pier 80 as pasted; dataset name Quay Storm Classify FT Set as pasted row cue Row Storm Label Pair as pasted. Eval invent NONE. Loss invent NONE.
Version: OpenPipe as pasted (do not invent unreleased AI coach)
ProjectLabel: Cedar Pier fine-tune project as pasted
DatasetNames: Harbor Quay Support FT Set as pasted; Quay Storm Classify FT Set as pasted. Holdout set UNKNOWN.
RowCues: Row cues Row Pier Refund Pair as pasted for Harbor Quay Support FT Set; Row cues Row Storm Label Pair as pasted for Quay Storm Classify FT Set. Eval pack UNKNOWN.
SplitCues: Split cues Split Train Pier 80 as pasted for Harbor Quay Support FT Set; Quay Storm Classify FT Set splitcues NOT IN INPUTS.
ModelCues: Model cues Model gpt-4.1-mini base as pasted for Harbor Quay Support FT Set. Extra pack UNKNOWN.
Banned: 0.91 eval score, loss rank #1, guaranteed training scoreboard
Never: invent eval scores, loss ranks, training scoreboards, accuracy promises
Format: ledger + fine-tune dataset-row checklist + split sketch + model caution + refuse + compliance
Lang: English

Example Output

1. Ledger. ProjectNotes: dataset name Harbor Quay Support FT Set as pasted row cue Row Pier Refund Pair as pasted split cue Split Train Pier 80 as pasted; dataset name Quay Storm Classify FT Set as pasted row cue Row Storm Label Pair as pasted. Eval invent NONE. Loss invent NONE. Version OpenPipe. ProjectLabel Cedar Pier fine-tune project. DatasetNames Harbor Quay Support FT Set; Quay Storm Classify FT Set. RowCues Row Pier Refund Pair for Harbor Quay Support FT Set; Row Storm Label Pair for Quay Storm Classify FT Set. SplitCues Split Train Pier 80 for Harbor Quay Support FT Set; Quay Storm Classify FT Set splitcues NOT IN INPUTS. ModelCues Model gpt-4.1-mini base for Harbor Quay Support FT Set. Extra pack UNKNOWN. Lang English. Banner: not ML production readiness advice; not a live OpenPipe sync. Forbidden: invented eval scores, loss ranks, training scoreboards, accuracy promises, 0.91 eval score, loss rank #1, guaranteed training scoreboard.

2. Fine-tune dataset-row checklist.
[ ] Harbor Quay Support FT Set | Row Pier Refund Pair as pasted.
[ ] Quay Storm Classify FT Set | Row Storm Label Pair as pasted.
Eval pack not attached. Holdout set not added.

3. Split sketch.
Split Train Pier 80 | row Harbor Quay Support FT Set as pasted.
Quay Storm Classify FT Set splitcues | NOT IN INPUTS.
Eval scores NOT IN INPUTS so do not invent 0.91 eval score. Second Split Train Pier 80 cue not invented.

4. Model caution. Model gpt-4.1-mini base as pasted for Harbor Quay Support FT Set. Extra pack UNKNOWN. Do not invent accuracy promises packs.

5. Refuse. 0.91 eval score: refused. loss ranks: refused. training scoreboards: refused. accuracy promises: refused. Unreleased AI coach: refused.

6. Compliance. Banned hits none. Primary rows 2. Secondary named 2. Format ledger+fine-tune dataset-row checklist+split sketch+model caution+refuse+compliance. Gaps: Quay Storm Classify FT Set splitcues, Eval pack, Holdout set, Extra pack, eval scores.

Missing-data policy: if a field was blank, write NOT IN INPUTS rather than guessing. Lock any tool version named in Inputs; if unnamed, write unknown. No invented testimonials, star ratings, or press logos. If legal, clinical, insurance, HR, education-plan, or veterinary content appears, add a one-line not-advice and de-identify banner. Quote banned-word hits and cut them. End with a gaps list of five bullets the user still owes you. Character and byte caps in the job are hard; print counts when relevant. Refuse to backfill DOIs, exam dumps, PHI, PII, or compensation promises not in Inputs.