Reconcile Two Spreadsheets Through a Deterministic-First Prompt Chain

ROLE Act as a data reconciliation workflow designer. OBJECTIVE Create a prompt chain to compare [DATASET A DESCRIPTION] with [DATASET B DESCRIPTION]. CONTEXT Candidate matching keys: [MATCHING KEYS]. Business comparison rules: [COMPARISON RULES]. Available execution environment: [EXECUTION ENVIRONMENT]. INSTRUCTIONS Design stages for schema inspection, normalization planning, deterministic matching, discrepancy classification, and reviewer explanation. Write complete prompts for model-assisted stages and specify executable logic for exact comparisons. Do not ask a language model to perform large-scale arithmetic or exhaustive row matching from memory. Preserve raw values, row identifiers, duplicates, and unmatched records. Separate exact matches from suggested fuzzy matches; never silently accept the latter. Define treatment of dates, currencies, nulls, rounding, and many-to-one relationships. Produce a reviewable discrepancy report rather than modifying source files. OUTPUT FORMAT Provide the chain map, stage prompts, matching-rule specification, discrepancy schema, and a small synthetic test dataset with expected results. If execution is unavailable, provide reproducible instructions without claiming the comparison ran. QUALITY CHECK Flag assumptions and avoid fabricated statistics. Confirm that every input row receives a traceable disposition.