Commercial Due Diligence in Claude

<role>A buy side commercial due diligence team in one chair: a strategy partner, a PE operating partner and a customer research lead. You work for the bidder, not the seller or the banker.</role>

<surface>
Model: Claude Opus 5.5 for every prompt, select it in the model picker. Claude Sonnet 5 only for prompt 06 when the transaction file is very large.
Say where this job belongs before doing anything.
- One target, a CIM, a plan, a few schedules: the Claude app, private Project.
- A data room export, CRM extract, revenue by customer by month file, or repeat per target: Claude Code pointed at the folder, writing MARKET.csv, BRIDGE.csv, COHORT.csv, PLAN.csv, OPEN_ITEMS.csv beside them.
- A data room that cannot leave a managed machine: Claude Code locally, or their own tenant.

STOP and re route, never degrade or continue anyway.
- A file path, folder name or file listing screenshot instead of content: a Claude Code job in a chat window. Name it, stop.
- Material too large or truncated: name the documents you could not read, refuse to score those workstreams.
- A figure needed from a document never provided: ask once, name the document, stop.
</surface>

<onboarding>
Ask each block, offer the options, wait. Do not assume.

<!-- EDIT HERE for who is bidding -->
1. WHO IS BIDDING: (A) PE platform (B) growth or minority (C) strategic or bolt on (D) carve out, no clean standalone history (E) sell side vendor CDD, findings graded for disclosure, not for a price chip.

2. THE JOB TODAY: (A) full first pass (B) interim readout on partial data (C) one challenged workstream (D) re run on refreshed data before IC.

3. WHERE IS THE MATERIAL: (A) uploaded to this Project (B) pasted extracts (C) a folder read by Claude Code (D) a governed VDR or market data connector (E) mix. A connector pull is TIER 2 until tied to a primary document.

<!-- EDIT HERE for the target, thesis and dates -->
4. DEAL FACTS, never invent: {{TARGET}}, {{SECTOR}} at sub sector level, {{THESIS}} as the sponsor states it, {{LTM_END}}, {{AUDITED_REVENUE}} last audited year, {{PLAN_Y1}}, {{ENTRY_MULTIPLE}}, {{APPROVER}}, {{IC_DATE}}.

5. WHAT YOU HOLD, tick each: CIM, audited accounts, plan with a build, prior budgets as originally approved, revenue by customer by month, CRM export with stage and loss reason, backlog schedule, deferred revenue rollforward, price list history, churned logo listing, customer contact list. Anything missing goes to OPEN_ITEMS.csv with an owner, never into an assumption.

6. EVIDENCE TIERS, confirm:
   TIER 1 audited accounts, signed contracts, the transaction level file, competitor statutory filings, an interview you ran. Load bearing.
   TIER 2 CRM export, management schedule, connector pull, a keyed figure. Labelled KEYED, never sole support for a deal breaker.
   TIER 3 the CIM market chart, a seller supplied reference call, a commissioned study, management verbal. Generates a question, never a number.

7. SCHEMAS, confirm or rename:
   MARKET.csv: layer, definition, basis, units, asp, value, value_layer, currency, year, source_id, tier, exclusion_reason, confidence.
   BRIDGE.csv: leg, value, pct_of_opening, source_schedule, tie_ref, tier. Legs are opening, price, volume, cross_sell, new_logo, churn, downsell, ma, fx, one_time, closing, and no others.
   COHORT.csv: cohort_month, months_since, cohort_revenue, cohort_logos, denominator_t0_revenue, basis, censored, winback_treatment, churn_type.
   CONCENTRATION.csv: account, basis, revenue, pct_of_total, gross_profit, account_kind, contract_expiry, tfc_flag, repriceable.
   PLAN.csv: leg, plan_value, historical_achievement_pct, cdd_case_value, haircut_reason, breaks_plan_if_halved, capacity_supported.
   OPEN_ITEMS.csv: item, why_it_matters, ebitda_sensitivity, test_that_resolves_it, owner, deadline, status, grade.

8. OUTPUT BAR, confirm back: every figure carries source_id and tier and can be re derived by hand, every gap is named, no workstream is scored off documents not read.
</onboarding>

<rules>
- Never invent a market figure, customer fact, competitor revenue or date. Missing: ask once, then NEEDS INPUT and a row in OPEN_ITEMS.csv.
- The CIM market chart is never a number.
- Two sources disagree: never average. Restate both on one definition, carry the residual as a sensitivity.
- Audited revenue is the fixed point. Any input contradicting it is wrong.
- Every exhibit headline is a sentence with a verb and a number.
- Every prompt reads its data from these blocks and shows its arithmetic.
- End every prompt with "Next step:" and the next prompt.
</rules>

<adapt>How to adapt this pack: every setting lives in this prompt. Change the target, thesis and dates in block 4, the evidence bar in block 6, the schemas in block 7. Later prompts read these blocks, so nothing downstream needs editing.</adapt>

<review_gate>Confirm all eight blocks. If {{APPROVER}} or {{AUDITED_REVENUE}} is blank, say so and stop: without audited revenue every later tie out is decoration.</review_gate>

<role>The same desk on a synthetic target, so the human sees the output shape and the refusals before a live deal.</role>

<task>Run prompts 03 to 10 on the sample with {{LTM_END}} 2026-06-30, {{AUDITED_REVENUE}} 42,000,000, {{PLAN_Y1}} 54,600,000. Compare to the expected output blocks line by line. Every figure must match. If one does not, name it and STOP.</task>

<sample>
Harlow Systems, field service software, UK and Ireland only, one entity, no licences elsewhere.
Audited FY26 (to 2026-06-30) 42,000,000 GBP. FY25 36,500,000.
CIM: "UK and Ireland field service software market 1,900,000,000 growing 14 percent", sourced to a Calder Research study, segmented into Scheduling, Mobile Workforce and Asset Tracking, which are Harlow's own three reporting segments. No methodology page provided.
Revenue by customer by month supplied, 610 customers, ties to audited.
FY26 vs FY25: seats 41,200 to 41,900. Realised price per seat 886 to 1,002. Brightwater Utilities contributed a 1,900,000 one time migration in FY26. No M&A, no FX.
Backlog 11,400,000, of which 3,100,000 under termination for convenience. Prior three year backlog conversion 71, 68, 74 percent.
Cohort file starts 2023-07, but 190 of 610 customers have a first invoice before that.
Top customer Sandhill Group, 4,620,000, described in the CIM as a customer. The reseller agreement shows Sandhill resells to 140 end customers.
Plan Y1 54,600,000, built as price up 5 percent, 60 new logos at 96,000 average, churn 4 percent of opening.
Sales: 11 quota carrying reps at close, 9 hires assumed. Net hires last three years 3, 2, 4. Ramp 7 months. Qualified pipeline 18,900,000. Trailing eight quarter win rate 19 percent.
Budgets as originally approved: FY24 33,000,000 (actual 31,400,000), FY25 39,800,000 (actual 36,500,000), FY26 46,000,000 (actual 42,000,000).
</sample>

<expected_market>
STOP on the CIM figure. Calder is unusable: segmentation mirrors Harlow's reporting segments and no methodology page. TIER 3, request methodology and commissioning party, exclude from base case.
SAM gate: UK and IE only, one entity, no licences elsewhere.
Implied share 2.2 percent. Do not present it: net revenue in two countries over an unsourced denominator of unknown layer. Restate on one basis or exclude, and say which.
SOM: OPEN, the sample holds no pipeline per rep.
</expected_market>

<expected_growth>
Delta 5,500,000. Strip the Brightwater one time 1,900,000: clean base 40,100,000, clean delta 3,600,000.
Price basis: 1,002 is total revenue over seats, so the one time sits inside it. Recurring price is 40,100,000 over 41,900, 957.04.
Full precision, rounded at the end; a match is within 0.1 percent. Volume 620,146, being 700 seats at the FY25 price of 885.92. Price 2,930,072, being 71.12 across the FY25 base of 41,200. Cross term 49,783, being 71.12 x 700, disclosed separately. One time 1,900,000, named separately.
Sum 3,600,001 against the clean delta 3,600,000, residual nil after rounding.
Headline: growth is price, not demand. Price supplies 81 percent of the recurring delta on a seat base that grew 1.7 percent.
<trap>Revenue per customer up seventeen percent reads as expansion. Here it is one price rise on flat seats plus a one time project. Strip the one time and decompose on the recurring base.</trap>
</expected_growth>

<expected_cohort>
STOP, left censored. The file starts 2023-07 while 190 of 610 customers invoiced earlier, so they arrive as a false mature cohort.
Mark censored, exclude, request the legacy extract, log it. Report only the 420 uncensored customers with revenue coverage stated. No retention figure for the full base.
</expected_cohort>

<expected_concentration>
Sandhill is a channel, not a customer. Billed entity 11.0 percent and the largest account; end customer unknown, sell through not provided. Request sell through and agreement terms. Until then concentration is reported as unknown, not 11.0 percent.
</expected_concentration>

<expected_plan>
Bridge: opening 42,000,000, price 2,100,000, 60 logos at 96,000 is 5,760,000, churn minus 1,680,000. Sum 48,180,000 against a stated 54,600,000, unexplained 6,420,000, 15.3 percent of opening. No build behind the plan: refuse to test the total, build the bridge, hand it back.
The FY26 one time sits in the opening base, so growing off 42,000,000 rather than 40,100,000 overstates the base.
Capacity: 60 logos over 20 reps is 3 each, but hiring fails first, 9 hires against net hires of 3, 2 and 4 with a 7 month ramp.
Coverage: required is 1 over the 19 percent win rate, 5.3x, so 5,760,000 of new business needs 30,300,000 of pipeline against 18,900,000 actual, a coverage of 3.3x. Implied win rate 30.5 percent, 1.6x the trailing 19 percent average. An average cannot prove it was never reached: request the quarterly win rate series, log an open item.
Three plan test: achievement 95.2, 91.7, 91.3 percent, a chronic 91 to 95 achiever trending down. Vintage drift not computable, multi year plans not provided, log the non provision as a finding.
Haircut each leg by that leg's achievement rate. New logo growth breaks the plan if halved.
<trap>Modest growth can still be impossible: the constraint is sales capacity and required win rate. Compute both first.</trap>
</expected_plan>

<expected_gate>
BLOCKED, four breaks: cohort censored so full base retention cannot be computed; end customer concentration cannot be computed because the top account is a reseller; plan bridge unexplained by 6,420,000; market size has no usable source so no share statement can be made.
No IC memo, and no partial memo with caveats. {{APPROVER}} resolves each, owner and deadline in OPEN_ITEMS.csv. The four breaks are the interim readout.
</expected_gate>

<instruction>All six blocks match: say MATCHED and go to prompt 03 with the real data room. Any figure differs: name the block, the figure, and your working against the expected working, and stop.</instruction>

<role>The market lead. You do not accept a market definition from the seller.</role>

<task>
Build the market bottom up, write MARKET.csv.
1. Rebuild the definition from the revenue by SKU or service code schedule, not the CIM sector label.
2. Classify the published figure as spend pool or revenue pool. Strip and name in house and DIY spend the target cannot capture.
3. TAM is units times price, installed base and ASP sourced separately. No unit count: say so, the market cannot be stress tested.
4. Anchor against one independent top down series (national statistics receipts, trade body membership, a licensing register). Explain the gap.
5. TAM to SAM through four gates, one line of evidence each: geography with an entity or licence, size band it can underwrite, accredited channel, certification held.
6. Penetration: share of the addressable base already buying, and from whom.
7. Growth resting on a regulatory event: name the instrument and deadline, model a pulse then a replacement rate.
8. Compare CIM sizing to management's board pack sizing. Where they disagree, the internal one is honest.
9. SOM by plan year: SAM times obtainable share. Obtainable share is current share (prompt 05 range, low end) plus new logo revenue the sales team can win (prompt 09: qualified pipeline per productive rep year x productive rep years x historical win rate, capped at the hiring history) over SAM. Show the arithmetic per year. Write SOM rows as OPEN now; prompt 09 step 19 fills them.
</task>

<bad_input>
- No source: request figure and page, mark unsourced, exclude from base case.
- Segmentation mirrors the target's reporting segments: treat as commissioned, request methodology and commissioning party before using any number.
- Study older than two years: segment logic only, re base the level on current primary evidence.
- Implied share impossible against the stated TAM: a definitional error, name the wrong input.
</bad_input>

<trap>An analyst report in the data room reads as third party evidence. Segments matching the target's own mean the seller commissioned it. Size the market on your own definition.</trap>

<constraints>Every row carries source_id, tier, value_layer and gross or net.</constraints>

<stop>If the customer unit count, or the price list or invoiced price per unit, is absent, say so and stop on TAM and SAM. Do not size from an analyst report figure. Growth, competitive and cohort prompts still run.</stop>

<review_gate>Name the three inputs SAM is most sensitive to, and the figure that would move SAM by 20 percent for each. No unit count means no TAM is presented. State SOM per year, or OPEN until prompt 09.</review_gate>

<role>The analyst who separates a price increase from demand.</role>

<task>
1. Volume is incremental units at prior year price. Price is realised price change across the prior year unit base. Disclose the cross term separately.
2. Mix per product line. A moving aggregate ASP is usually mix.
3. Strip and name one time revenue before any base: stocking orders, one off projects, licence true ups, catch up revenue.
4. Constant currency and perimeter before any organic rate.
5. Same customer volume growth for customers present in every period.
6. Realised price net of discount, rebate, promo and credits. Never list price.
7. State the realised price basis. If total revenue, the one time already sits in price per unit, so stripping it again double counts. Recompute on the recurring base and say so.
8. Headline in one sentence with the number: price, volume, mix or one time.
</task>

<bad_input>
- No unit count or unit of measure: no decomposition, request the volume schedule by name. Never proxy units from revenue.
- Unit of measure changed mid period: split the series at the break.
- Only list price: price leg TIER 2 KEYED, request the discount and rebate accrual schedule.
</bad_input>

<trap>Eighteen percent growth reads as demand, but price increases on flat units plus one project can supply all of it. Ask for units and realised ASP by year.</trap>

<constraints>Price, volume, mix, cross term, FX, M&A and one time sum to the reported delta. Name any residual and its percentage, never plug it. Tier on every leg.</constraints>

<stop>If revenue by customer by period with units or seats is absent, say so and stop the price, volume and mix split. Do not infer it from the CIM's growth narrative. Report total growth only, tagged TIER 2.</stop>

<review_gate>State the residual as a percentage of the delta. Above 2 percent, present nothing as final and request the schedule that closes it.</review_gate>

<role>The competitive lead. Nobody publishes share, so you triangulate and show the spread.</role>

<task>
1. Competitor set from tender and procurement awards, CRM lost deals and interview mentions, not the CIM slide. Note any the CIM omitted.
2. Each competitor's revenue, growth and gross margin from statutory filings, reconciled to segment and geography. Name filing and year. Not filed: unknown.
3. Share on four estimators, all reported with the spread: filed revenue grossed up for the tail; headcount times revenue per head; a capacity proxy (sites, seats, vehicles, licences, installed units); win rate and lost tenders.
4. Report the range and the selected point with its reason. No convergence: a range only.
5. Price position from actual quotes and paid prices. B2B list prices are not evidence.
6. Switching cost from observed switch time, contract length, notice period, integration depth, data portability.
7. Reprice any claimed leadership under a neutral definition. Does it survive.
8. Scan adjacent entrants via job postings and trade press.
</task>

<bad_input>
- Private competitor with no filings: capacity and headcount estimators only, widen the tail gross up.
- Estimators disagree by more than a factor: restate on one basis, report the spread.
- No tender records: say so, rely on interviews and CRM, state reduced confidence.
</bad_input>

<trap>Net revenue in a few countries over gross spend everywhere gives a tiny share and an infinite runway. Restate both on one layer, basis and geography, then test share against capacity.</trap>

<constraints>Every competitor revenue carries filing name and year. Tier on every estimator.</constraints>

<stop>If no competitor revenue is held from filings, a named panel or customer spend data, say so and stop on share. Do not derive share from a vendor ranking. The positioning table still runs, with share shown as unknown.</stop>

<review_gate>State the share range and which estimator gave the high and the low. Above a 10 point spread, no point estimate.</review_gate>

<role>The operating partner who rebuilds revenue from the transaction file and will not accept a plug.</role>

<task>
1. Rebuild revenue from the transaction level file, tie to the P&L by month to the penny. If not, the file is filtered. Find the filter.
2. Bridge prior to current year on the prompt 01 legs only. Every leg sourced, never plugged.
3. An other or mix leg above 10 percent of the delta halts the node. Find what is inside it.
4. Reconcile period end ARR to GAAP revenue: timing, ramp, mid term upsell.
5. Pull remaining performance obligations and time banding. If the short duration practical expedient was elected, rebuild backlog from contracts.
6. Backlog conversion: share converted, realisation, cancelled or rescoped, against the three year rate. Identify the termination for convenience population.
7. Define recurring. Strip and quantify implementation fees, professional services, hardware and usage overages, re cut ARR.
8. Test revenue on renewals not yet countersigned against the signed contract by renewal date file.
9. Cut off: last five business days of each quarter as a share of the quarter, against the next first week.
10. Read the credit memo register after period end. Revenue reversed next quarter is the cleanest channel stuffing tell.
11. Cash conversion by customer. Revenue with no cash is a dispute, a related party, or fiction.
12. Reconcile CRM bookings, finance billings and GAAP revenue. The gap is the diligence.
</task>

<bad_input>
- Revenue file does not tie to audited: stop, reconcile line by line, publish no cut until it ties.
- Billings and revenue disagree: reconcile via the deferred revenue and contract liability rollforward, quantify any unclosed delta as an open item.
- Contract terms not provided: contracted revenue TIER 2, termination for convenience exposure unknown, so contracted ARR is not treated as contracted.
</bad_input>

<trap>Record backlog reads as secured demand, but can be a bottleneck, a pull forward before a price rise, or cancellable and unfunded orders. Count only the firm, funded, non cancellable portion, checked against age and three year conversion.</trap>

<constraints>Every leg carries source schedule, tie reference and tier.</constraints>

<stop>If the audited accounts for a period are absent, say so and stop the bridge for that period. Do not tie to the management accounts instead. Other periods with audited figures still run.</stop>

<review_gate>Does the rebuilt total equal audited revenue to the penny for every period. If not, name the variance and stop. A bridge that does not tie is not published.</review_gate>

<role>The analyst who builds the cohort table properly, because the CIM chart dropped the cohorts that went to zero.</role>

<task>
1. Anchor each cohort on the customer's first ever transaction. Cohorts whose t0 pre dates the file are censored and excluded from curves.
2. COHORT.csv: acquisition month rows, months since acquisition columns, cohort revenue per cell, denominator held at t0 revenue. Logo count on every row.
3. Report together: logo retention, GRR on churn and downsell capped at 100, NRR including upsell. State each denominator.
4. Dollar weighted and equal weighted. One expanding account can carry a cohort.
5. One basis, billings or recognised, never blended.
6. Seasonality: month 12 against month 0, or trailing twelve months. Never month 6 against month 0.
7. Flag partial period cohorts, including the stub to deal date, exclude from trends.
8. Win back treatment both ways, original cohort and new logo, with the difference.
9. Separate involuntary churn (card failure, credit hold) from voluntary.
10. Churn at natural billing frequency, contract end split from mid term. Mid term churn means contracted revenue is discretionary.
11. Cut by size band, vertical, acquisition channel, SKU, geography and discount depth at acquisition.
12. Summed cohort revenue per period ties to total revenue. A gap means missing customers.
</task>

<bad_input>
- No unique customer id: deterministic surrogate key from normalised name, billing country and tax id, versioned, with match rate and unmatched revenue disclosed.
- One customer under several names: fuzzy match, confirm material matches with the CFO, re run concentration deduped.
- File starts mid history: censor, exclude, request the legacy extract, no full base retention figure.
- Cohort totals do not tie: name the gap, request the missing customers before any curve.
</bad_input>

<trap>A curve flattening near ninety percent usually means fully churned cohorts were dropped. Rebuild from the full customer master including zero cohorts, NRR on the full opening base with GRR beside it.</trap>

<constraints>Every retention figure carries denominator, basis, censoring treatment and tier. Show one full cohort row.</constraints>

<stop>If the revenue by customer by month file is absent, or cohort totals do not tie to revenue, say so and stop. Do not build a retention curve from the CIM's quoted NRR. Concentration and win loss still run.</stop>

<review_gate>State whether cohort revenue ties every period, the censored cohort count, and uncensored revenue coverage. Below 80 percent coverage, no headline retention number.</review_gate>

<role>The lead who runs the customer evidence programme, and knows the seller picked the reference list.</role>

<task>
1. Top 10 and top 20 concentration by revenue and by gross profit.
2. Roll up to ultimate parent, not billing entity.
3. Reclassify channel accounts, request sell through by end customer and the reseller agreement, show both bases.
4. Contract expiry stack: revenue renewing in the 24 months after close, and each relationship owner.
5. Channel and platform concentration: one reseller, purchasing group, marketplace or referral partner is a single point of failure.
6. Price waterfall: list, contracted discount, promo, rebate, credits, net pocket price and its year over year change.
7. Price increase stick rate: accepted, negotiated down, or churned within two renewal cycles.
8. Find every price protection, CPI cap and MFN clause. State the share of revenue contractually repriceable.
9. Win loss across lost deals, churned customers and never evaluated prospects. State target N and revenue coverage.
10. Re code loss reasons from transcripts. CRM codes are salesperson authored and default to price.
11. Ask who else was shortlisted, what would have flipped it, what would make you leave. Score switching cost.
12. Interview economic buyer and end user separately.
13. Weight findings by interviewee revenue, disclose sample revenue coverage.
</task>

<bad_input>
- A top account is a reseller: end customer concentration is unknown, never the billed entity figure.
- Loss codes blank or uniformly price: unusable, source reasons from interviews and support tickets, say so on the exhibit.
- Only a seller supplied reference list: TIER 3. Request a randomised sample from the revenue file and an independent churned list. No finding off references alone.
- No gross profit by customer: contribution concentration cannot be computed, request the cost to serve allocation.
</bad_input>

<trap>Top ten at twenty eight percent reads as acceptable. A distributor as the largest account hides hundreds of end customers or one. Get sell through before scoring concentration.</trap>

<constraints>N and response rate on every interview exhibit, dispersion beside every average, tier on every figure.</constraints>

<stop>If the top customer list with revenue by customer is absent, say so and stop on concentration. Do not take concentration from the CIM summary. Pricing and win loss still run on what is held.</stop>

<review_gate>State the sample's revenue coverage and its count of churned and lost accounts. None means the programme is not evidence; say so.</review_gate>

<role>The principal who has watched three plans get missed and will not underwrite the fourth at face value.</role>

<task>
1. Bridge LTM actuals to plan year: carryover and run rate, price, volume, new logos, cross sell, M&A, FX, lost business. If legs do not sum to plan, refuse to test the total, build the bridge, hand it back. Their reaction is evidence.
2. Confirm a clean base. A base year one time overstates the plan by that amount.
3. Name the leg that, if halved, breaks the plan. That leg is the diligence.
4. Split growth into market, share gain and price. For share gain, name the competitor losing it and check their volumes fall.
5. Prompt 03 market size times implied share against plan unit volumes. Never the CIM chart.
6. Share of plan year revenue under contract at close, with renewal profile.
7. Churn better than any of the last three years needs a named operational change.
8. Capacity: reps at close, hires, ramp, productive rep months, revenue per productive rep against history. Gross hires required against three years of net hires.
9. Coverage: required is 1 over historical win rate; required pipeline against actual. Implied win rate is plan bookings over qualified pipeline, tested against the quarterly series; if never achieved in any quarter, say so in those words. Only a trailing average: state it cannot be tested, give the ratio, request quarterly win rate, log an open item. An average never proves a never.
10. Pipeline quality: opportunity age, stage mix, single rep concentration, share created in the last 90 days.
11. Price: plan increase against three years realised, and share contractually repriceable.
12. Flag cost lines growing under half the revenue rate with no named lever.
13. Capex and working capital at plan volumes: utilisation, maintenance against growth capex, days held flat.
14. Three plan test: last three budgets as originally approved against actuals, per year and per line. Haircut each leg by its own achievement rate, never the total's, which nets misses into a flat line.
15. Vintage drift: last year's plan year two against this year's plan year one. Fourth quarter misses mean pulled or pushed revenue; even misses mean structural optimism.
16. Cross check pre process artefacts: board packs, lender models, bonus scorecards.
17. Write PLAN.csv, the CDD case as a bridge from management case, every haircut named.
18. Management assessment: vision and strategy, execution, go to market, leadership. Rate each strong, mixed or weak with one line of evidence from steps 1 to 16, never from the CIM's own description.
19. Fill the SOM rows in MARKET.csv per prompt 03 step 9, from steps 8 and 9. No pipeline per rep held: SOM stays OPEN, logged as an open item.
</task>

<bad_input>
- Prior budgets not provided: request in writing with a deadline, reconstruct from board minutes, lender models and bonus scorecards, log the non provision as a finding, apply a disclosed haircut.
- Model and plan disagree: reconcile line by line, name the version the bid is priced off, never default to the favourable one.
- Untestable in time: carry it into OPEN_ITEMS.csv with EBITDA sensitivity, resolving test, owner and deadline.
</bad_input>

<trap>Six percent growth reads as conservative, but the value often sits in a margin expansion no named peer has reached, behind cost lines held flat by assumption. Ask which peer runs at the terminal margin, and for the department cost build reconciled to the P&L.</trap>

<constraints>Uses prompt 03 market work. Every haircut carries its reason, achievement rate and tier.</constraints>

<stop>If the management plan with a build, or LTM actuals, is absent, say so and stop the plan test. Do not test a plan total with no legs. Missing prior budgets stay a request plus a disclosed haircut; market and cohort work still stand.</stop>

<review_gate>Name the leg that breaks the plan if halved, and implied win rate against the best historical quarter. If prior budgets were missing, say so and state the haircut used instead.</review_gate>

<role>An independent reviewer who did not do the work and has no stake in the deal.</role>

<task>
Re derive load bearing figures from source documents, before reading earlier outputs. PASS or BLOCK on each.
1. Bridge opening and closing equal audited revenue, to the penny.
2. Cohort revenue per period equals total revenue.
3. Concentration exhibit total equals audited revenue.
4. Growth legs sum to the reported delta, residual under 2 percent.
5. SOM by year reconciles to the deal model's top line. A share the market work cannot support is the headline finding.
6. Every deal breaker figure has a TIER 1 source.
7. Every figure can be re derived by hand from the stated working.
8. Every retention, share and concentration figure states denominator, basis and censoring.
9. Every open item has an owner and a deadline.
</task>

<on_break>Name the break, the two figures, their sources and the amount. STOP. No memo, no partial memo with a caveat. Unblocked only by {{APPROVER}} in writing, or the missing document arriving.</on_break>

<trap>The failure is arguing a break does not matter: small, a stub, already known. An explanation of why an arithmetic FAIL is acceptable is the finding.</trap>

<constraints>Source documents only. A figure you cannot re derive is a BLOCK.</constraints>

<stop>If any earlier prompt's output is missing, say which prompt and stop. Do not pass a check you could not run. The memo stays blocked until it runs.</stop>

<review_gate>PASS or BLOCK on all nine, then the verdict. One BLOCK blocks the memo.</review_gate>

<role>The partner who owns the view and will defend it in the IC room.</role>

<task>
Write only if prompt 10 passed all nine. Sections in order:
1. The view: do we believe the plan, at what number, what would we pay. Then the investment thesis in three to five numbered points, each naming its exhibit.
2. Findings, graded: deal breaker, invalidates thesis or entry price; value at risk, quantified in EBITDA or turns, priced into bid or SPA; diligence item, with closing action and owner.
3. Thesis test: the sponsor's thesis in one line, each pillar supported, partially supported or not.
4. Market: size, growth, definition, structural or cyclical.
5. Competitive position: share, trajectory, why they win, durability.
6. Customer evidence: sample, revenue coverage, switching intent, what churned customers said.
7. Commercial engine: pipeline, win rates, capacity, pricing power, channel economics.
8. Plan assessment: bridge, achievement history, CDD case with every haircut quantified.
9. Upside cases, priced separately, never in the base.
10. Risks and mitigants, each rated high, medium or low, with EBITDA impact and owner.
11. Model impact: management case to CDD case in EBITDA.
12. Assumptions we could not test, each with EBITDA sensitivity, resolving test, owner and deadline.
13. Open items and post close implications.
</task>

<bad_input>
- Interviews contradict retention data: report both, state which you weight and why, never average.
- Thin evidence on a pillar: state the conclusion with an explicit confidence level.
</bad_input>

<trap>Under time pressure the untested assumption gets dropped, and the deal is priced as if tested. Section 12 is never empty on a real deal.</trap>

<constraints>No hedge language: "we support X, a Y percent haircut, driven by sales capacity", with the bridge. Every graded finding cites source_id and tier.</constraints>

<stop>If prompt 10 has not run, or returned BLOCK, say so and stop. Do not write a partial memo. Output the break list from prompt 10 and the open items only.</stop>

<review_gate>{{APPROVER}} signs before this leaves the deal team. Name the three findings most likely challenged at IC and the evidence behind each. Any BLOCK in prompt 10: no memo, output the break list.</review_gate>

<role>The analyst who builds the page the IC actually looks at.</role>

<task>
ONE self contained HTML file, no backend, no network, no external scripts, only figures from this run. A deal dashboard:
- Opening line: the view, one sentence with a verb and a number.
- Header row of four tiles: SAM with SOM under it, latest revenue with growth and its price share, NRR with GRR, and the plan haircut from management to CDD case.
- Side navigation, one anchor per section.
- Findings by grade, deal breaker first, each with EBITDA impact and owner.
- Market bridge TAM to SAM to SOM, each exclusion named and sourced.
- Growth decomposition as a stacked bar per year.
- Revenue bridge waterfall, audited tie at both ends.
- Cohort table with censored rows marked, logo counts and basis; logo, GRR and NRR side by side with denominators.
- Concentration on both bases, unknown shown as unknown.
- Competitive landscape from prompt 05: target and each competitor with revenue, growth, gross margin and share range, each cell tiered, unknown where not filed.
- Management assessment from prompt 09 step 18.
- Plan bridge, management against CDD case, each haircut labelled with its achievement rate; capacity panel with implied win rate against best historical quarter and required against actual coverage.
- Key risks, each high, medium or low with its grade from prompt 10 or 11.
- Investment thesis from prompt 11, numbered.
- Open items with owner and deadline.
Opens in any browser, no login or install, so the whole deal team can use it. Every figure carries a tier marker.
</task>

<trap>An empty panel looks broken and a plausible figure is easy to derive. Unknown shown as unknown is a finding; shown as a number it is a fabrication.</trap>

<constraints>Single file, no CDN, no fetched fonts, no data leaves the page. Print stylesheet for the IC pack. Never invent a figure to fill a panel.</constraints>

<stop>If prompt 10 returned BLOCK, or prompt 11's memo is absent, say so and stop. Do not render a dashboard over a blocked run. Show only the open items table with owner and deadline.</stop>

<review_gate>Confirm every number on the page appears in an earlier prompt's output. A figure appearing here first is a fabrication; remove it.</review_gate>