Crescent Finance Ops Reporting

Crescent Finance Ops Reporting

Core workflow

1. Read the task prompt, payloads/request_memo.json, payloads/answer_template.json, and payloads/environment_access.json.
2. Treat answer_template.json as the contract: return exactly one JSON object with the requested keys, field names, list ordering, and numeric rounding. Do not add narrative fields just because the memo asks for background context unless the template includes them.
3. Use the base_url from environment access and query the listed remote API endpoints directly. Do not assume a local task server or local cached data.
4. Inspect each endpoint's actual shape before calculating. Some endpoints return arrays, while rate books and scenarios are keyed objects.
5. Use a small script or structured calculation rather than manual arithmetic. Terminal output can truncate large /records responses, so fetch into memory before aggregating.
6. Return JSON numbers, not formatted strings. Currency fields round to 2 decimals. Percent, ratio, margin, and growth fields round to 4 decimals as decimal rates, e.g. 0.1234, not 12.34.

General conventions

• Stable IDs are meaningful. Sort ID lists ascending unless the template says a rank or top/bottom fact requires another order.
• For descending ranks, sort by the target metric high-to-low and break ties by stable ID ascending.
• Growth and variance percentages use (current - comparison) / comparison.
• If a denominator is zero, do not guess; use null only if the template permits it, otherwise recheck the data and memo.
• Names in memos may mention draft workbooks or notes. Reconcile against active API data, not memo prose.
• For objects with fixed category keys, include all template-required categories with 0 when absent. For per-person category maps that explicitly ask for nonzero categories, omit zero categories.

Finance branch and regional reports

Use:

• /api/finance/branches
• /api/finance/period-map
• /api/finance/accounts
• /api/finance/records

Build fiscal-year period sets from /period-map instead of hardcoding. In the staged data, M1-M12 map to FY2024 and M13-M24 map to FY2025, but future tasks should still read the map.

Account formulas:

• revenue = all records whose account category is revenue.
• cogs = category cogs.
• gross_margin = revenue - cogs.
• sga = category sga.
• allocations = category allocations.
• ebitda = gross_margin - sga - allocations.
• ebitda_margin = ebitda / revenue.
• sales_per_labor_headcount = revenue over the same period set divided by aggregated labor_headcount, unless the prompt explicitly asks for average headcount.
• arpu = revenue over the same period set divided by revenue_units, unless the prompt explicitly defines a different user/customer denominator.

For a close-period income statement, aggregate only the requested period, e.g. M24. For current fiscal-year views, aggregate all periods in that fiscal year. For regional reporting, filter branches by region_id, then sum branch metrics.

Ranking facts:

• Branch sales or revenue growth ranks compare all branches over the requested fiscal-year pair.
• EBITDA ranks inside regional context usually compare the branches in that target region only.
• top_*_branch_id and bottom_*_branch_id should be selected after applying the same fiscal-year or period window used for the metric.
• region_reconciliation_variance should tie regional totals back to the sum of included branch-level totals; when both are derived from the same records this should round to 0.00.

Period labels should come from /period-map when requested. Include the period code and fiscal year when useful, but preserve any template-provided field names such as current_month and prior_month.

Compensation summaries

Use:

• /api/compensation/rate-book
• /api/compensation/rosters

Filter rosters by ensemble_id. Use rate_book.pay_types for pay-type ordering and exact labels.

Per employee, per quarter:

1. Use employee.weeks_by_quarter[quarter] when present. This matters for partial-quarter players; do not blindly use 13 weeks.
2. Minimum Weekly Scale = minimum_weekly_scale * weeks.
3. Titled Position Premium = minimum_weekly_scale * title_premium_pct[title] * weeks, only when the title has a premium and combined_overscale_includes_title is false.
4. Seniority = seniority band weekly amount for years_of_service times weeks. Bands are inclusive of min_years and max_years; a null max means no upper bound.
5. Overscale = overscale_weekly * weeks.

Then sum by quarter, pay type, and year. annual_total is the sum of all pay-type totals and should also equal the sum of quarter totals after rounding tolerance.

Roster treatment counts:

• roster_count = number of filtered roster rows.
• combined_overscale_employee_count = count of employees where combined_overscale_includes_title is true.
• partial_quarter_employee_count = count of employees with any quarter week count different from rate_book.quarter_weeks.

largest_pay_type is the pay type with the largest annual total; use rate-book pay-type order to break ties.

Compensation forecasts

Use:

• /api/compensation/rate-book
• /api/compensation/rosters
• /api/compensation/scenarios

Start with the current-year compensation method above. Select the requested scenario_id.

Forecast handling:

• Year + 1 uses one added year of service for seniority-band selection; Year + 2 uses two added years.
• Apply annual growth rates year over year for dollar rates. For example, Year + 2 minimum weekly scale should build on the Year + 1 scaled rate, then apply the Year + 2 mws_growth.
• Apply overscale_growth to employee overscale weekly amounts and seniority_growth to seniority weekly amounts after selecting the aged service band.
• Apply each forecast year's title_pct_multiplier to the base title premium percentage for that forecast year.
• Continue to skip separate title premium for employees whose combined overscale includes title.

Forecast growth_rates use annual totals:

• year_plus_1_vs_current = (year_plus_1 - current) / current.
• year_plus_2_vs_year_plus_1 = (year_plus_2 - year_plus_1) / year_plus_1.

For largest_growth_pay_type, use the greatest dollar increase from current to Year + 2 unless the memo names a different interval. If the memo asks for a driver explanation but the template only provides the enum field, return only the enum.

Weekly payroll reviews

Use:

• /api/payroll/rate-book
• /api/payroll/productions

Select the requested production_id. Build a schedule map by service_id, then calculate each roster musician from their assigned_service_ids.

Service and category calculations:

• service_counts counts schedule rows by exact service_type.
• Performance and Audit are per-service flat rates from service_rates.
• Rehearsal is hourly at the rehearsal rate with a 3-hour minimum call: max(duration_hours, 3) * rate.
• Sound checks are per-service flat rates by exact service type, such as 1hr Sound Check or 2hr Sound Check, and roll into the sound_check category.
• Base service pay is the sum of performance, audit, rehearsal, and sound-check base amounts.
• Role/electronic/quartet premiums are applied to base service pay using premium_pct. Apply principal_or_lead once if either principal or lead is true. Apply electronic, quartet, or concertmaster only when the roster exposes the corresponding flag or role.
• Doubles are a separate category: 25% of base service pay for the first extra instrument plus 10% for each additional extra instrument.
• Vacation is 4% of base service pay plus premiums, including doubles, when vacation_eligible is true.
• Weekly guarantee adjustment applies only to non-substitute regular players when base service pay is below weekly_guarantee.
• Do not invent a substitute_adjustment; include it only if the rate book, roster, or template makes it applicable. Substitutes do not receive weekly guarantee unless the rate book says otherwise.

Per musician:

• Sum all category amounts to total.
• Include only nonzero categories in each musician's categories map when the template asks for nonzero categories.
• Sort per_musician by musician_id.
• top_paid_musician_id is the highest total, with musician_id ascending as a tie-breaker.

Conflict flags:

• REHEARSAL_EARLY_START: rehearsal starts before conflict_thresholds.rehearsal_earliest_start.
• REHEARSAL_LATE_END: rehearsal ends after conflict_thresholds.rehearsal_latest_end.
• SERVICE_OVER_TIME_LIMIT: duration exceeds service_time_limits[service_type].
• SOUND_CHECK_DURATION_MISMATCH: a sound-check duration does not match the hour count implied by its service type.

Return conflict_flags sorted alphabetically.

Final answer checklist

• The response is a single JSON object with no Markdown fence.
• Top-level keys match required_top_level_keys.
• Currency is rounded to 2 decimals; ratios and growth rates to 4 decimals.
• ID lists and per-person lists follow the template's ordering rule.
• Category keys use the template's exact spelling and casing.
• No train-only IDs, memo notes, or explanatory comments are included unless the current task template asks for them.