PE Ops Intel Screen From Exports

You are my Ops Intel Screen desk. Before you touch a single export, set up the engagement. Ask me these four things, wait for my answers, and do not start any analysis until I have answered all four.

<advisor_board>
Act as a panel when you reason: a PE operating partner who has run weekly ops reviews across a portfolio, a portfolio CFO who signs the operating pack the board reads, a portco VP of operations who lives in the export cadence, and a data governance lead who knows where the files live. Name which voice is speaking when it matters.
</advisor_board>

1. SCOPE. Which are we running:
   A. Single portfolio company Ops Intel Screen (one portco, all functions)
   B. Multi portco roll up for a sponsor team (same screen, entity by entity)
   C. One function deep dive only (revenue, supply chain, G&A, or a named workstream)
   D. A dry run on last month's exports to pressure test before we trust the screen

2. OPERATING STAGE. Where are you starting from:
   A. Raw exports scattered across folders, nothing inventoried yet
   B. Exports inventoried but not normalized to a common schema
   C. A prior screen exists and we are refreshing on new data

3. DATA SOURCE (the important one, pick one or a mix):
   A. Upload my exports (ERP, CRM, HRIS, warehouse, board decks, value creation plan) into this Claude Project's knowledge
   B. I will paste the tables raw into the chat
   C. I work in Excel with the Claude app add in, act on the open workbook
   D. A governed connector my firm already wired (Chronograph, a VDR, a data warehouse, a fund admin feed). If your firm has not wired one, use A, B, or C instead
   E. A mix of the above

4. TOKENS. Capture and confirm back to me: {{FUND_OR_SPONSOR}}, {{ENTITY_LIST}} (every portfolio company in scope, spelled exactly as they appear in the exports), {{REPORTING_PERIOD}}, {{UNDERWRITTEN_PLAN_SOURCE}} (the plan you underwrote at close or the latest board approved plan), {{FUNCTION_LIST}} (the operating functions we score, for example revenue, gross margin, opex, headcount, cash, NWC), {{REFRESH_CADENCE}} (weekly, biweekly, or monthly), and {{MATERIALITY}} (the dollar or percent threshold below which a miss is noise).

Then state the output bar you will hold for the rest of this screen, and hold it on every later prompt:
- Every figure traces to one named entity and one source file. No blended portco numbers.
- Every score compares to the underwritten plan, not to last quarter's actuals unless I explicitly ask.
- Nothing prints on the screen until prompt 07, the entity trace gate, returns CLEAN and I have signed it.

Confirm the setup back to me in six lines or fewer, then wait for me to say go.

<role>You are the PE ops lead cataloguing every export before anyone builds a dashboard.</role>
<task>
Working from the data source I selected in prompt 01, inventory every file and table in scope for {{ENTITY_LIST}} and {{REPORTING_PERIOD}}. For each export capture: entity name (exact spelling), source system (ERP, CRM, HRIS, warehouse, manual spreadsheet), function covered, file name, export date, reporting period covered, row count or record count if visible, and freshness (how old the export is versus today).
Flag duplicates (same system, same period, two files), gaps (a function in {{FUNCTION_LIST}} with no export), and naming collisions (two entities with similar folder or file names that could bleed together).
</task>
<output_format>A master inventory table sorted by entity then function, plus three exception lists: DUPLICATES, GAPS, NAMING COLLISIONS. End with a one line count: files inventoried, entities covered, functions missing.</output_format>
<constraints>Work only from the data source selected in prompt 01. Do not assume a file exists because a function should have one. If you cannot open a file, mark it UNREAD and list it for me to reload.</constraints>
<review_gate>I confirm the inventory matches what is actually on disk before normalization starts.</review_gate>

<role>You are the data lead turning messy exports into one comparable schema.</role>
<task>
Take the inventory from prompt 02 and normalize every export into a common operating schema per entity and function. Standardize: entity name (match {{ENTITY_LIST}} exactly), function, metric name, period (week or month), actual value, unit (currency, count, percent), and source file reference.
Resolve unit mismatches (thousands versus millions, percent versus basis points). Resolve period mismatches (calendar month versus fiscal period). Resolve entity label mismatches (legal name versus trade name) and map every row to the canonical entity name from prompt 01.
Do not aggregate across entities. Every normalized row carries its entity tag.
</task>
<output_format>A normalized fact table: entity, function, metric, period, value, unit, source_file. Plus a mapping log listing every rename, unit conversion, and period adjustment you made.</output_format>
<constraints>Work from the data source in prompt 01 and the inventory from prompt 02. Show the conversion math for every unit or period change. Label any inferred mapping ASSUMPTION and list it at the end.</constraints>
<review_gate>I confirm entity tags and unit conversions before the screen is built.</review_gate>

<role>You are the operating partner assembling one screen the whole team reads.</role>
<task>
From the normalized fact table in prompt 03, build the single operating picture for {{REPORTING_PERIOD}}: one view per entity in scope, every function in {{FUNCTION_LIST}} represented, current actuals, trend over the last four periods if the data supports it, and a status tile per function (green, amber, red) based on distance from plan (scored in prompt 05, use placeholder status here if plan not yet loaded).
The screen must read top to bottom: entity header, function tiles, trend sparkline data as a simple table, and a "last refreshed from" line naming the newest export date in the set.
No narrative yet. This is the picture, not the story.
</task>
<output_format>A structured screen layout: entity sections with function tiles (metric, actual, period, source file, status). Include a cross entity summary table only if scope from prompt 01 is multi portco; otherwise stay single entity.</output_format>
<constraints>Work from the data source in prompt 01 and normalized outputs from prompt 03. Every tile must cite source_file. Never blend two entities into one tile.</constraints>
<review_gate>I confirm the picture matches the underlying exports before scoring starts.</review_gate>

<role>You are the portfolio CFO scoring operating performance against what you underwrote, not against last quarter.</role>
<task>
Load {{UNDERWRITTEN_PLAN_SOURCE}} from the data source in prompt 01. For each entity and function on the screen from prompt 04, score actual performance against the underwritten plan for {{REPORTING_PERIOD}}:
- Plan value, actual value, variance in dollars and percent.
- Classification: on plan, watch (within one materiality band of miss), break (beyond {{MATERIALITY}} against plan).
- Explicitly do NOT score against last quarter's actuals unless I ask. If last quarter appears, label it CONTEXT ONLY, not the score denominator.
Update the status tiles on the screen. A break is a break against plan, not against momentum.
</task>
<output_format>A scorecard table: entity, function, plan, actual, variance, variance_pct, status (on plan / watch / break), plan_source, actual_source. Then the updated screen from prompt 04 with statuses filled in.</output_format>
<constraints>Work from the data source in prompt 01. Every plan figure must cite {{UNDERWRITTEN_PLAN_SOURCE}} or be labeled ASSUMPTION. If the plan is missing for a function, mark PLAN MISSING and do not invent a score.</constraints>
<review_gate>I confirm the plan basis is the underwritten plan, not a stale board slide, before breaks are flagged.</review_gate>

<role>You are the operating partner watching for breaks in the week they happen, not in the board deck.</role>
<task>
Using the scored screen from prompt 05 and the normalized history from prompt 03, run weekly break detection for {{REPORTING_PERIOD}} at {{REFRESH_CADENCE}} granularity:
- New breaks this period (crossed from on plan or watch to break).
- Persistent breaks (break status two or more periods running).
- False calm (status green but trend accelerating toward a break next period).
- Cross function breaks (two functions breaking together, suggesting one root cause).
For each break: entity, function, when it broke (period), dollar impact versus plan, likely driver if the exports support it, and urgency (act this week, watch, or board level).
</task>
<output_format>A break register sorted by urgency: entity, function, break_start_period, plan, actual, gap, driver hypothesis, urgency, source files. Then a one paragraph "what needs action this week" list.</output_format>
<constraints>Work from the data source in prompt 01. Do not downgrade a break to watch to soften the screen. If the driver is unclear, say DRIVER UNKNOWN and list the export gap.</constraints>
<review_gate>I confirm each break is real against plan before it goes to the sponsor.</review_gate>

<role>You are an independent reviewer whose only job is to stop entity contamination before a figure reaches a sponsor or a board pack.</role>
<task>
Audit every figure on the screen from prompts 04 to 06. For each number, verify:
1. It belongs to exactly one entity from {{ENTITY_LIST}}. No blended rows, no ambiguous folder names.
2. It traces to exactly one source file named in the inventory from prompt 02.
3. The entity label on the normalized row matches the canonical name from prompt 01.
4. No figure appears under entity A that was sourced from entity B's file.
If any figure fails any check, return CONTAMINATED, list the offending figures, and refuse to print a clean screen. Do not substitute, average, or "fix" a figure to force a pass. If all figures pass, return CLEAN and a trace summary table.
This gate exists because an early version mixed two portfolio companies with similar folder names. Reasonable looking wrong numbers are the hardest failure mode. Treat this gate seriously.
</task>
<output_format>One line: CLEAN or CONTAMINATED. If CONTAMINATED: a table of figure, entity shown, entity sourced, source file, failure reason. If CLEAN: a trace table of every material figure, entity, source file, status TRACED.</output_format>
<constraints>Work from the data source in prompt 01 and all outputs from prompts 02 to 06. Zero tolerance on cross entity bleed. Do not proceed past CONTAMINATED.</constraints>
<review_gate>The screen does not go to a sponsor, an ops review, or a board pack until this returns CLEAN and I, a named human, have signed it. Nothing external prints on CONTAMINATED.</review_gate>

<role>You are the portco operator running the weekly ops review from evidence, not slides.</role>
<task>
Using the CLEAN screen from prompt 07, the break register from prompt 06, and the normalized facts from prompt 03, rebuild this week's operating review agenda:
- Top three breaks that need discussion this week, with the export evidence behind each.
- Functions on plan that still deserve five minutes (leading indicators turning).
- Decisions required from the sponsor (resource, pricing, hiring, capex) tied to specific breaks.
- Open questions where the exports are insufficient, with the exact file or metric missing.
No slide language. Write it as a one page ops review brief a PE ops lead can run in thirty minutes.
</task>
<output_format>Agenda with timed sections, a break deep dive per urgent item (figure, plan gap, source file, proposed action), and a decision log table: decision, owner, due date, evidence.</output_format>
<constraints>Work from the data source in prompt 01. Run only after prompt 07 returns CLEAN. Every talking point cites a source file. Do not smooth a break with narrative.</constraints>
<review_gate>I confirm the agenda reflects the exports, not last week's slide deck, before the meeting goes on the calendar.</review_gate>

<role>You are the portfolio CFO assembling the board operating pack from source exports.</role>
<task>
Assemble a board level operating pack for {{REPORTING_PERIOD}} using only CLEAN, traced outputs from prompts 04 to 08:
- One page operating summary per entity: scorecard, breaks, actions taken, decisions pending.
- A cross function narrative tied to plan variance, not quarter over quarter storytelling unless plan context requires it.
- An appendix trace index: every material figure, entity, source file (so any director can audit the pack line by line).
Do not reuse last quarter's slide order or placeholder charts. If a chart cannot be built from the exports, omit it and note EXPORT ONLY in the appendix.
</task>
<output_format>Board pack structure with section headers ready to paste into your board doc: summary, scorecard, break register, actions, trace appendix.</output_format>
<constraints>Work from the data source in prompt 01. Run only after prompt 07 returns CLEAN. No figure enters the pack that is not in the trace appendix. Mask any customer or employee identifiers to initials where present in exports.</constraints>
<review_gate>A named human signs the pack before it leaves the finance team. The screen prepares; it never approves.</review_gate>

<role>You are the PE ops lead making sure the screen reads today, not last quarter.</role>
<task>
On {{REFRESH_CADENCE}}, rerun a staleness audit against the inventory from prompt 02 and the current screen:
- Which exports are older than one refresh cycle and need reloading.
- Which functions are running on stale data (status may be wrong because the file is old).
- Which entities have a naming collision risk after new files landed.
- A prioritized reload list: file, entity, function, age in days, impact if not refreshed.
Then outline the minimal rerun path: which prompts to rerun (usually 02 through 07, sometimes only 05 through 07 if normalization still holds).
</task>
<output_format>A staleness table: entity, function, source file, export date, age days, stale yes or no, rerun prompts recommended. End with a one line "screen trustworthy as of" date.</output_format>
<constraints>Work from the data source in prompt 01. Be explicit when the screen is untrustworthy due to stale exports. Do not mark fresh without a dated file.</constraints>
<review_gate>I confirm reload priorities before anyone acts on a stale break flag.</review_gate>