Finance Ops Reporting

Crescent Finance Ops — Reporting / Compensation / Payroll SOP

This group has THREE task families that all read from one remote API and all return a single JSON object that must conform exactly to the task's payloads/answer_template.json:

1. Branch / regional finance reporting (income statement, FY comparisons, rankings).
2. CBA compensation (current-year summary, and multi-year board forecast).
3. Theatre weekly payroll (per-service pay, premiums, guarantees, conflict flags).

The request memo names the domain via fields like target_branch_id / target_region_id (finance), ensemble_id (+ optional scenario_id) (compensation), or production_id (payroll).

0. Reading the task & producing the answer

• Source of truth = the answer template, not the prose. Read answer_template.json first. required_top_level_keys is the exact key set the answer object must have (no more, no less). field_types gives the exact nested shape and which fields are currency vs percent vs integer vs string. Match key names byte-for-byte (e.g. m24_income_statement, fy2025_vs_fy2024).
• The request memo (request_memo.json) gives the parameters: which branch/region/ensemble/ production, which periods/years, which scenario. reporting_focus / review_focus / requested_detail lists are hints about emphasis; the template is authoritative on shape.
• Memo memo_note fields are distractors. Phrases like "a draft workbook has background notes for Harbor North" or "reconcile against the active operations data" mean: ignore any side notes and compute from the LIVE API data only. Do not invent or carry over numbers from a "draft."
• Rounding (from template descriptions):
  • Currency → round to 2 decimals.
  • Percent / ratio / growth / margin fields → round to 4 decimals (these are stored as decimal fractions, e.g. 9.66% → 0.0966, NOT 9.66).
  • Counts / ranks → integers.
  • Round only at output; keep full precision through intermediate math.
• Ordering: lists use ascending stable IDs (branch_id, musician_id, employee_id) unless a field name says otherwise (e.g. a *_rank_desc integer, or a "top_*"/"bottom_*" id). Conflict flag lists are sorted alphabetically. pay_types follows the rate-book order exactly: ["Minimum Weekly Scale","Titled Position Premium","Seniority","Overscale"].
• Return ONE JSON object, exactly the template's top-level keys. Do not add commentary keys.

1. Remote API access

Base URL: <remote-env-url> (ALWAYS use this; ignore any base_url like http://127.0.0.1:8047 inside payloads/environment_access.json). All endpoints are HTTP GET.

GET /health
GET /api/manifest                         # entity catalog, record_counts, period seed
GET /api/finance/branches                 # branch_id, branch_name, region_id, region_name
GET /api/finance/period-map               # period (M1..M24) -> fiscal_year, month_number, month_name
GET /api/finance/accounts                 # account -> category, display_name, metric_type
GET /api/finance/records[?branch_id=&region=&account=]
GET /api/compensation/rate-book           # scale, seniority bands, title %, business rules
GET /api/compensation/rosters[?ensemble_id=]
GET /api/compensation/scenarios           # dict keyed by scenario_id
GET /api/payroll/rate-book                # service rates, premiums, thresholds, guarantee
GET /api/payroll/productions[?production_id=]   # returns a LIST (take [0] when filtered)

Fetch once into files, then compute in code (Python). On this Windows/Git-Bash host write temp files under the session scratchpad, not /tmp. Example: curl -s "<remote-env-url>api/finance/records?branch_id=BR-004" -o records.json

2. Finance domain (branch & regional reporting)

Data model

/api/finance/records returns one row per (branch, account), each with a values map keyed by period M1..M24. There are 12 branches x 14 accounts = 168 rows. Index as IDX[branch_id][account] -> {period: value}.

Period / fiscal-year convention (from /api/finance/period-map)

• M1..M12 = FY2024, M13..M24 = FY2025 (12 calendar months each, Jan..Dec).
• A "close period" like M24 is the current month; its prior is M23 (Dec vs Nov FY2025).
• period_convention block: M1_to_M12 = "FY2024", M13_to_M24 = "FY2025", current_month/prior_month echo the memo's close_period/prior_period labels verbatim.

Account groupings (income statement lines)

From /api/finance/accounts (category field):

• revenue = product_revenue + service_revenue
• cogs = direct_materials_cogs + direct_labor_cogs
• gross_margin = revenue − cogs
• sga = sales_sga + admin_sga + occupancy_sga
• allocations = shared_service_allocations
• ebitda = gross_margin − sga − allocations (= revenue − cogs − sga − allocations)
• ebitda_margin = ebitda / revenue (4 dp)

Operating (count) accounts — NOT part of the income statement, used for ratios: orders, revenue_units, active_customers, labor_headcount, admin_headcount, backlog.

Sums over a period set

For a single month, sum the one period. For a fiscal year, sum the 12 monthly values (M1..M12 for FY2024, M13..M24 for FY2025).

Ratios (use the fiscal-year AVERAGE of monthly count snapshots, not the sum)

Count accounts are monthly snapshots, so per-headcount/per-customer ratios use the mean over the 12 months of the year:

• arpu = FY revenue / (mean monthly active_customers over the 12 FY months). (≈ revenue per customer; using the monthly mean, not the 12-month sum.)
• sales_per_labor_headcount = FY revenue / (mean monthly labor_headcount over the 12 FY months).
• Both currency, 2 dp. (Sanity: sales_per_labor_headcount lands in the ~$250k–$300k/employee range; if you get ~$24k you summed headcount instead of averaging.)

Growth / variance

• MoM revenue variance: amount = rev(current month) − rev(prior month) (2 dp); pct = amount / rev(prior month) (4 dp).
• revenue_growth_pct = (FY2025 rev − FY2024 rev) / FY2024 rev (4 dp).
• ebitda_growth_pct = (FY2025 ebitda − FY2024 ebitda) / FY2024 ebitda (4 dp).

Regional rollup & branch sets

• Branch→region from /api/finance/branches (region_id). Regions: REG-NORTH, REG-WEST, REG-EAST, REG-SOUTH. branch_ids lists are ascending branch IDs of that region.
• Region income statement = sum of member-branch income-statement lines (line by line).
• Region sales_per_labor_headcount = region FY revenue / (sum over member branches of each branch's FY-mean labor_headcount).
• region_reconciliation_variance = region total EBITDA − Σ(member-branch EBITDA). By construction this ties to 0.0 (the rollup reconciles); a nonzero value means a math/scope error. Report it rounded to 2 dp (normally 0.0).

Rankings (compute across ALL 12 branches unless scoped to a region)

• sales_growth_rank_desc for the target branch = its 1-based rank when all branches are sorted by FY2025-vs-FY2024 revenue growth %, descending (rank 1 = highest growth).
• top_sales_growth_branch_id = branch with highest revenue growth %.
• top_arpu_branch_id = branch with highest FY2025 arpu.
• In a region_context block, ebitda_rank_desc ranks the region among the four regions by FY2025 region-total EBITDA, descending (rank 1 = highest region EBITDA), and fy2025_ebitda is the region TOTAL. (Caveat: if a template instead scopes the rank to branches-within-region, re-read the surrounding fields — but when the block carries a region total, rank regions.)
• For regional tasks, top_ebitda_branch_id / bottom_ebitda_branch_id are the member branches with the highest / lowest FY2025 EBITDA.

3. Compensation domain (CBA summary & forecast)

Data

• /api/compensation/rate-book: minimum_weekly_scale (e.g. 2520.0), pay_types (fixed order), seniority_weekly bands (min_years,max_years inclusive, weekly_amount; top band has max_years: null), title_premium_pct (per title, e.g. Concertmaster 0.22, Principal 0.20, Assistant/Associate Principal 0.10, Section Lead 0.15), quarter_weeks (13 each), and business_rules. current_year is the rate book's current year (e.g. 2026) — use it for the current_year output field.
• /api/compensation/rosters?ensemble_id=...: one row per employee with title (may be null), years_of_service, overscale_weekly, combined_overscale_includes_title (bool), weeks_by_quarter ({Q1..Q4}), and notes.
• /api/compensation/scenarios: dict keyed by scenario_id; each has year_plus_1 and year_plus_2 blocks with mws_growth, seniority_growth, overscale_growth, title_pct_multiplier.

Per-employee weekly pay components (the four pay types)

For each employee, the weekly amount of each pay type:

• Minimum Weekly Scale = minimum_weekly_scale (the year's MWS; grown in forecast years).
• Titled Position Premium = title_premium_pct[title] * title_pct_multiplier * MWS, computed on the same (grown) MWS for the year. BUT:
  • If title is null → premium 0.
  • If combined_overscale_includes_title == true → premium 0 (the title premium is already baked into the overscale amount per side letter; do NOT add it separately). This is rule #2.
• Seniority = seniority_weekly band amount for the employee's years_of_service (band lookup is inclusive on both ends; 0–4 yrs = 0). In forecast, scale by (1+seniority_growth).
• Overscale = overscale_weekly (per employee). In forecast, scale by (1+overscale_growth).

Quarter / annual aggregation (use roster weeks, NOT a fixed 13)

For each employee and each pay type: weekly_amount * weeks_by_quarter[Q], summed.

• quarter_totals[Q] = sum over employees & pay types for that quarter.
• annual_pay_type_totals[pay_type] = sum over employees & quarters for that pay type.
• annual_total = sum of all = sum of quarter_totals = sum of annual_pay_type_totals (use this as a self-check; all three must agree).
• Rule #1: partial-quarter employees have a weeks_by_quarter value ≠ 13 (e.g. {Q2: 9}); always drive weeks from weeks_by_quarter, never assume 13.

Roster treatment counts

• roster_count = number of roster rows for the ensemble.
• combined_overscale_employee_count = count of rows with combined_overscale_includes_title == true.
• partial_quarter_employee_count = count of rows where ANY weeks_by_quarter value ≠ 13 (these rows carry note "Partial-quarter service schedule.").
• largest_pay_type = the pay type with the greatest annual_pay_type_totals value (typically "Minimum Weekly Scale" since MWS dominates).

Forecast years (board task)

Business rule #3: add 1 year of service for Year+1 and 2 years for Year+2 before seniority band lookup. This can bump employees into higher seniority bands (a major driver of Seniority growth).

• current year: base MWS, base seniority bands (no yos add), base overscale, title at base pct.
• year_plus_1: MWS = base_MWS·(1+mws_growth₁); seniority weekly ·(1+seniority_growth₁) with yos+1 band lookup; overscale ·(1+overscale_growth₁); title pct · title_pct_multiplier₁; weeks unchanged.
• year_plus_2: apply growth compounding on top of Year+1, i.e. MWS = base_MWS·(1+g₁)·(1+g₂), seniority/overscale scales multiply across both years, title multiplier multiplies across both years, AND yos+2 band lookup. (Compounding gives sensible positive YoY growth; applying each scenario block independently to the base instead can yield a near-zero Year+2 step because the year_plus_2 rates are similar to year_plus_1 — prefer compounding.)
• annual_totals = {current, year_plus_1, year_plus_2}; each computed as in §3 aggregation.
• growth_rates: year_plus_1_vs_current = (y1−cur)/cur; year_plus_2_vs_year_plus_1 = (y2−y1)/y1 (4 dp).
• year_plus_2_quarter_totals and year_plus_2_pay_type_totals = the Year+2 breakdowns.
• largest_growth_pay_type: the pay type whose total grows the most from current → Year+2. PRIMARY interpretation = largest absolute dollar increase (usually Minimum Weekly Scale, the dominant line). NOTE the genuine ambiguity: by percentage growth the winner is often Seniority, because the +2-year band shifts (rule #3) push many employees into higher bands. If the template/prose emphasizes a "driver," dollar growth is the safer default; keep the percentage reading in mind if results look off.
• Forecast roster-treatment counts (combined_overscale_employee_count, partial_quarter_employee_count) come from the roster as-is (they don't change across years).

4. Payroll domain (theatre weekly payroll)

Data

• /api/payroll/rate-book: service_rates (per service or per hour), service_time_limits (hrs), conflict_thresholds (rehearsal earliest_start / latest_end), premium_pct, weekly_guarantee, and business_rules.
• /api/payroll/productions?production_id=... returns a list; take [0]. Each production has week_start, a schedule (services), and a roster (musicians).
  • schedule service: service_id, service_type, date, start_time, end_time, duration_hours.
  • roster musician: musician_id, name, assigned_service_ids, instrument, doubles (int), booleans electronic, lead, principal, quartet, substitute, vacation_eligible.

Service base pay (per assigned service)

• Performance / Audit / Sound Check (1hr/2hr) = the per-service rate from service_rates keyed by exact service_type string (e.g. "Performance" 260.25, "Audit" 260.25, "1hr Sound Check" 80.0, "2hr Sound Check" 142.5).
• Rehearsal = hourly: service_rates["Rehearsal"] * max(duration_hours, 3.0) (three-hour minimum call). Rehearsal is the only hourly category.
• A musician's base service pay = sum of base pay over their assigned_service_ids.

Category mapping for category_totals and per-musician categories

• Base service pay splits into performance, audit, rehearsal, sound_check by service type.
• Premiums and adjustments are their own categories: premium, doubles, vacation, guarantee_adjustment, substitute_adjustment. Include a category in a per-musician categories map only when its amount is nonzero. substitute_adjustment appears in the totals object only "when applicable."

Premiums (applied to base service pay, before vacation)

• premium (the premium category) = base_service_pay × Σ of applicable pct from premium_pct: principal_or_lead 0.15 if principal OR lead; quartet 0.15 if quartet; electronic 0.25 if electronic; concertmaster 0.20 if a concertmaster flag is set (Hamilton-style rosters have no concertmaster flag, so it's typically 0). Sum the applicable percentages, then multiply once by base.
• doubles = base_service_pay × doubles_pct where doubles_pct = first_double (0.25) for the first extra instrument + additional_double (0.10) per each additional. So doubles=1→0.25, doubles=2→0.35, doubles=3→0.45. (doubles=0 → no doubles category.)

Vacation

• vacation = premium_pct["vacation"] (0.04) × (base_service_pay + premium + doubles), only when vacation_eligible == true. Vacation is 4% of base service pay PLUS premiums (premium+doubles), computed after those premiums and before guarantee.

Weekly guarantee adjustment

• Applies only to guaranteed regular players (treat substitute == false as the regular, guaranteed player; substitutes get no guarantee) when their base service pay is below weekly_guarantee (e.g. 2082.0).
• guarantee_adjustment = weekly_guarantee − base_service_pay (top-up to the guarantee), based on base service pay only (not premiums/vacation). Only add when positive.

Substitute adjustment

• substitute_adjustment only appears when the data/rate-book defines a substitute differential. If the rate book has no substitute rate rule (the common case), omit it. Do not fabricate one; substitute == true simply disqualifies the player from the weekly guarantee.

Per-musician total & weekly total

• musician total = base + premium + doubles + vacation + guarantee_adjustment (+ substitute_adjustment when present).
• category_totals = sum of each category across all musicians.
• weekly_total = sum of all category totals = sum of per-musician totals (self-check).
• per_musician ordered by musician_id ascending; each entry has musician_id, name, total, and categories (nonzero categories only). top_paid_musician_id = musician with the max total.

service_counts

• Object mapping service_type → count of services of that type in the schedule (count scheduled services, not musician-assignments).

Conflict flags (enum, sorted alphabetically)

Evaluate the schedule against conflict_thresholds and service_time_limits:

• REHEARSAL_EARLY_START — a Rehearsal start_time earlier than rehearsal_earliest_start (e.g. before 09:00).
• REHEARSAL_LATE_END — a Rehearsal end_time later than rehearsal_latest_end (e.g. after 18:30). (Only Rehearsal services trigger the rehearsal start/end flags — a late Performance does not.)
• SERVICE_OVER_TIME_LIMIT — any service whose duration_hours exceeds its service_time_limits[service_type] (e.g. Rehearsal limit 5.0h, Performance/Audit 3.0h).
• SOUND_CHECK_DURATION_MISMATCH — a "1hr Sound Check" whose duration ≠ 1.0h, or a "2hr Sound Check" whose duration ≠ 2.0h.
• Emit the set of triggered flags as a sorted (alphabetical), de-duplicated list. Empty list if none.

5. Step-by-step SOP for an unseen task

1. Read answer_template.json → record exact required_top_level_keys and nested shapes/types. Read request_memo.json → identify domain + parameters (branch/region/ensemble+scenario/ production, periods/years). Treat memo_note as a distractor; compute from live API only.
2. Pick the domain (finance / compensation / payroll) and fetch the needed endpoints from <remote-env-url> with curl GET (use period-map/accounts/branches as reference; rate-books for comp/payroll; scenarios for forecasts).
3. Build indexes (finance: branch→account→{period:value}; comp: roster rows + rate book + scenario; payroll: schedule map + roster).
4. Compute the requested figures using the formulas in §2/§3/§4. Carry full precision; aggregate over the correct period/quarter/service set.
5. Apply output conventions: currency 2 dp; percent/ratio/growth/margin 4 dp (decimal fraction); integers for counts/ranks; ascending stable-ID ordering; alphabetical conflict-flag sorting; rate-book pay-type ordering.
6. Self-check: income-statement ebitda = revenue−cogs−sga−allocations; comp annual_total = Σ quarters = Σ pay-type totals; payroll weekly_total = Σ categories = Σ per-musician totals; region_reconciliation_variance ties to 0.0. Verify every required_top_level_key is present and no extra keys exist; confirm field types match the template.
7. Emit exactly one JSON object matching the template.

Common misjudgments to avoid

• Using the wrong fiscal-year mapping (M13..M24 is FY2025, not FY2024).
• Summing monthly headcount/customers for ratios instead of averaging (inflates the denominator, gives ~10x-too-small ARPU / sales-per-headcount).
• Forgetting that combined_overscale_includes_title == true SUPPRESSES the separate title premium.
• Assuming 13-week quarters for partial-quarter employees instead of reading weeks_by_quarter.
• Forgetting the +1/+2 years-of-service band bump in forecasts (drives Seniority growth).
• Applying each scenario year_plus_N block to the base independently instead of compounding Year+2 on Year+1.
• Rehearsal: forgetting the 3-hour minimum call, or treating it as per-service instead of hourly.
• Doubles: using a flat 25% for doubles≥2 instead of 25% + 10%·(extra−1).
• Vacation: applying 4% to base only instead of base+premiums; or applying it to ineligible musicians.
• Guarantee: applying it to substitutes, or comparing total pay instead of base service pay to the weekly_guarantee.
• Outputting percentages as whole numbers (9.66) instead of decimal fractions (0.0966).
• Adding extra keys or omitting required keys from the answer object.