Ma Deal Workbench Review

M&A Deal Workbench Structured Review

Use this skill when a task asks you to act as buyer- or seller-side counsel on an M&A deal, gather deal records from the deal workbench, compare the current draft terms against a playbook or committee policy, and return only a JSON object that conforms to a provided answer_template.json.

The five recurring deliverable shapes this skill covers:

• Issue register — list of issues vs a playbook, with priority order and summary metrics.
• Closing/economics package — holder-level consideration allocation, indemnity/escrow/survival/NWC, consents, regulatory, covenants, D&O tail/expenses, closing readiness.
• Committee escalation memo — only the draft terms that are out-of-policy or restricted for committee approval (distractors excluded).
• Carveout transition review — IP/domain transition, TSA scope/fees, §1060 allocation, transfer-tax split, employee continuity, outside-date protection, governing law/forum.
• SPA deviation matrix — buyer positions on indemnity/survival/scrape/escrow/consents/HSR/material contracts, plus closing blockers and risk totals.

The method is the same for all five. Only the standard (playbook vs policy), the sub-resources needed, and the output schema change — and the answer template tells you the last two.

Prerequisites

• environment_access.md is staged in the working directory. It is the only source for network access: it gives the workbench base URL, the POST /api/query token, and the allowlist of endpoints. Do not use any other network source.
• A task prompt naming a deal_id, a side (buyer/seller), and a deliverable shape.
• An answer_template.json defining the exact output contract.

Procedure

1. Read environment_access.md first

Capture the base URL, the /api/query token, and the allowed-endpoint list. Use only endpoints on that list. If the base URL is given as <TASK_ENV_BASE_URL>, substitute the real value from environment_access.md.

2. Load answer_template.json before fetching anything

The template is the contract. Build a checklist from it:

• Required top-level fields and per-object fields.
• allowed_enums (risk_rating, recommended_action, issue_status, business_outcome, etc.) — every emitted enum value must match exactly, including case.
• units — the precision for currency, percent points, and months. Precision varies by task (whole percent, 1 dp, 2 dp, 4 dp for holder percentages); read it, do not assume.
• Any stable_issue_ids / stable_redline_ids / possible_issue_ids — use these verbatim as IDs; do not invent new ones.
• Ordering rules for arrays (by issue_id ascending, by priority, etc.).
• The summary-metrics / risk-totals / aggregate-summary block — these are the exact metrics to compute; do not invent extras.

3. Fetch the deal record

GET /api/deals/<deal_id> returns the deal object plus a links map to every sub-resource. From the deal object capture:

• client_side (buyer/seller), transaction_type (APA / SPA / merger / carveout) — these set the analysis posture.
• The value basis fields used to quantify dollars: headline_value, upfront_cash, stock_value, milestone_value.
• The governing standard: playbook_id and/or policy_id (one or both may be null).

4. Identify and fetch the governing standard (scoped)

• Negotiation-position reviews (issue register, transition review, deviation matrix, closing package) → use the playbook named in the prompt/deal record: GET /api/playbooks/<playbook_id>/rules. Each rule carries preferred_position, fallback_position, limit_value, limit_unit, basis, required_action, risk_default.
• Committee escalation → use the policy: GET /api/policies/<policy_id>/thresholds. Each threshold carries threshold_value, threshold_unit, restricted_flag, approval_required, policy_standard, basis.

Always use the path-scoped endpoint (with the id in the URL). If you use POST /api/query for cross-table checks instead, you must filter playbook_rules/policy_thresholds by the deal's specific playbook_id/policy_id. Those tables hold rows for all playbooks/policies; an unscoped join produces cross-standard contamination (e.g., one category appearing with two conflicting thresholds from different policies).

5. Gather the sub-resources the task needs

Pull them from the links map (equivalently /api/deals/<deal_id>/<resource>). Typical set: terms, consents, employees, regulatory, benchmarks, risk-estimates, diligence-findings, notes, material-contracts. Add cap-table for SPA holder allocation; documents for version tracking. Key fields per resource are in references/workbench_api.md. Do not fetch resources the deliverable does not use.

6. Filter distractors before comparing

The workbench intentionally contains noise. Clean it first:

• Drop draft terms where staleness_flag != "current".
• De-duplicate categories that repeat from an earlier draft.
• Match on the exact deal_id; ignore similarly-named projects (e.g., two projects whose names differ by one letter are different deals with different terms).
• For committee tasks specifically: escalate only restricted_flag="yes" terms that breach their threshold. Exclude in-policy terms, non-restricted terms, and terms whose approval_required is below committee level — these are deliberate distractors.
• When a term's source conflicts, prefer the latest document version.

7. Compare each current draft term to its matching standard by category

For each draft term, find the standard rule/threshold with the same category and classify issue_status:

• in_policy — draft is at or inside the preferred/standard position.
• draft_exceeds_playbook — draft is harsher than the fallback allows (higher cap/escrow, longer survival, larger fee than the side's fallback).
• draft_below_playbook — draft is weaker than preferred (or below a buyer-preferred minimum) but within fallback.
• out_of_policy — draft breaches a policy threshold (committee tasks).
• missing_required_term — no current draft term exists for a category the side's position requires. Use source_term_ids: []. Treat draft silence as an issue when the standard demands an affirmative provision and the surrounding deal data (consents, employees, regulatory, diligence) shows the term is needed.

Assign risk_rating (LOW/MEDIUM/HIGH) from the rule's risk_default, the term severity, and the quantified dollar exposure. Pick recommended_action from the template enum, guided by the rule's required_action. The full classification table is in references/comparison_method.md.

8. Quantify dollars from the correct base

Read each term's/rule's basis field; do not default to headline_value:

• purchase price / equity value → headline_value
• upfront cash → upfront_cash
• enterprise value → headline_value unless the deal record gives a distinct EV
• identified findings / a specific finding or consent → that record's amount / amount_at_risk

amount = percent_points × base, rounded to integer USD. If a source text states an explicit dollar amount on a different basis, use the stated amount. Compute delta_to_fallback (and delta_to_preferred / shortfall_to_fallback where the template asks). Quantification detail in references/comparison_method.md.

9. Position against benchmarks (when the template asks)

Use /api/deals/<id>/benchmarks: median_value, upper_quartile, sample_size set the benchmark position enum (at_or_below_median / between_median_and_upper_quartile / at_upper_quartile / above_upper_quartile / not_applicable).

10. Aggregate summary metrics

Compute exactly the metrics the template's summary block lists (e.g., issue_count, high_risk_count, total_quantified_exposure_low/high, total_negotiation_delta, required_closing_consent_count, total_employee_count, total_pto_liability, closing_blocker_count). For exposure totals, sum the risk_estimates categories the task says to include and exclude the ones it says to exclude (e.g., a task may include closing-certainty + indemnity but exclude transition-disruption). Do not invent metrics.

11. Emit only JSON

Return a single JSON object conforming to the template. No prose, no markdown fences, no commentary. Every enum value spelled exactly as in allowed_enums. Nulls only where the template permits null. Stable IDs copied verbatim from the workbench (term_id, consent_id, contract_id, finding_id, employee_id, holder name). Arrays sorted per the template's ordering instructions.

Pitfalls (confirmed against the workbench)

• Unscoped SQL joins contaminate. playbook_rules and policy_thresholds hold all standards. A join without WHERE playbook_id=<deal's> / policy_id=<deal's> returns conflicting thresholds for the same category. Prefer the path-scoped REST endpoints; if you must use SQL, filter by the deal's standard id.
• Stale/duplicate rows. Always filter staleness_flag="current". Committee deals carry stale in-policy rows as distractors.
• Wrong value base. "equity value" usually means headline_value, not stock_value. Read basis.
• Percent precision varies. Whole / 1 dp / 2 dp / 4 dp — read the template's units.
• Committee distractors. In-policy and non-restricted terms must be excluded, not escalated.
• Missing-term issues. Need source_term_ids: [] and issue_status: missing_required_term — do not skip a required category just because the draft is silent.
• "Return only JSON" is literal. Any prose outside the object fails the task.

References

• references/workbench_api.md — full endpoint catalog, auth, and the 14-table data model with key fields per resource.
• references/comparison_method.md — issue-status classification table, quantification basis map, benchmark positioning, and aggregation rules.
• references/formatting.md — per-type units/precision, enum conformance, stable-ID and ordering rules, and JSON-only output discipline.