Self Attempt 02

Crescent Arts Collective — Finance Ops Reporting Skill

1. Environment Setup

• Always use the remote API specified in environment_access.md (e.g. GDPEVO_ENV_BASE_URL).
• Ignore any localhost / 127.0.0.1 URLs inside payloads/environment_access.json; the remote entrypoint overrides them.
• Do not read, enter, or run anything in an env/ source directory.

2. Input Files to Read

For every task, read these staged payloads before calling the API:

1. payloads/environment_access.json — note the listed endpoints only; override the base_url.
2. payloads/request_memo.json — extract target IDs, periods, years, focus lists, and any memo notes.
3. payloads/answer_template.json — this is the authoritative output schema. Copy its required keys and field types exactly.

3. API Endpoint Mapping

Call only the endpoints relevant to the task family. Query all records needed to compute every required output field.

4. Core Calculation Rules

• Currency fields → round to 2 decimals (round(x, 2)).
• Percent, ratio, and growth-rate fields → round to 4 decimals (round(x, 4)), expressed as decimal values (e.g. 0.1523 for 15.23%).
• Period convention mapping (period-map): M1–M12 → one fiscal year string, M13–M24 → the next fiscal year string. Use the map to resolve current_month / prior_month labels.
• Income statement line items: revenue, cogs, gross_margin, sga, allocations, ebitda. Compute gross_margin and ebitda from the underlying account/record data if not provided directly.
• Growth / variance: pct = (current - prior) / prior when prior ≠ 0; handle zero-prior explicitly (return 0.0 or null per template guidance).
• EBITDA margin = ebitda / revenue (revenue ≠ 0).
• Sales per labor headcount = revenue / labor_headcount (use active roster/headcount data).
• ARPU = revenue / user_count or defined metric from finance records.
• Rankings are descending (1 = highest value). Ties: preserve stable ascending ID order.
• Reconciliation variance = sum of branch values − region total (or similar cross-foot per task context), rounded to currency.

5. Output Formatting Conventions

• Return exactly one JSON object.
• Include all required_top_level_keys from answer_template.json; missing keys cause validation failure.
• List ordering:
  • Default: ascending by stable ID (e.g. branch_id, musician_id) unless a rank field or explicit template instruction states otherwise.
  • conflict_flags → alphabetically sorted.
  • per_musician → ordered by musician_id ascending.
• Enum values (e.g. pay types, conflict flags) must match the template strings exactly, including case and spaces.
• Do not include test answers, derivations, or extra commentary inside the JSON.

6. Compensation-Specific Conventions

• Pay types from the rate book are the authoritative ordered list.
• Annual totals = sum of quarters or direct annual aggregation.
• Quarter totals = sum of applicable weeks/services in that quarter.
• Overscale counts: employees receiving any overscale pay in the period.
• Partial-quarter counts: employees with service that does not cover the full quarter.
• Largest pay type / largest growth pay type: determined by absolute annual total or absolute growth amount; break ties by ascending pay-type name if needed.

7. Payroll-Specific Conventions

• Service types map to pay categories (performance, audit, rehearsal, sound_check, premium, doubles, vacation, guarantee_adjustment, substitute_adjustment).
• Weekly total = sum of all category totals.
• Per-musician totals = sum of categories for that musician; include only nonzero categories in the categories object.
• Top-paid musician = highest total; break ties by ascending musician_id.
• Conflict flags (CBA checks): derive from production schedule vs. rate-book rules (e.g. early/late rehearsal, overtime, sound-check duration mismatch). Return only flags that are triggered, sorted alphabetically.

8. Verification Checklist

• Remote base URL used for every request.
• All required top-level keys present in final JSON.
• Currency rounded to 2 decimals; percents/ratios to 4 decimals.
• Lists sorted per convention (ascending ID, alphabetical, or rank-ordered as specified).
• Zero-division cases handled gracefully.
• Derived fields (margins, growth, rankings, variance) recalculated from raw API data, not hard-coded.