Acceptance Test Plan — NDA Review v1.0.1

Skill summary

Reviews mutual or unilateral non-disclosure agreements for unusual provisions, perspective-calibrated risks, and recommended position. Supports discloser, recipient, and mutual perspective inputs; calibrated to common in-house counsel review.

Test corpus requirements

Source 6–10 anonymized NDAs covering:

• At least 2 mutual NDAs (cross-disclosure scenarios — JV exploration, partnership discussions, technology evaluations).
• At least 2 unilateral NDAs from the discloser perspective (the user's organization is the disclosing party — e.g., a customer NDA the user's company sends to a vendor before a deal discussion).
• At least 2 unilateral NDAs from the recipient perspective (the user's organization is receiving confidential information — e.g., the counterparty has imposed their NDA before the deal can advance).
• At least 1 unusual-structure NDA that includes one or more of: trailing residuals carveout for M&A diligence, employee non-solicitation, return-or-destroy with proof, multi-jurisdiction governing law selection, or a notably long (5+ year) survival period.
• At least 1 routine vendor NDA (a clean, market-standard template — to confirm the skill doesn't over-flag on a baseline document).

For perspective-branching tests, the same NDA can be run twice with different perspective inputs to verify perspective-aware calibration.

Source documents from: prior matters in the user's practice, anonymized samples from the operator's organization, or — if none are available — public templates from sources like the SEC EDGAR filings (which include NDAs as exhibits to public filings).

Test scenarios

Scenario 1: Routine mutual NDA (baseline calibration)

Inputs: A clean, market-standard mutual NDA with no unusual provisions. Perspective: mutual.

Expected output structure:

• Markdown report with sections: "Bottom line", "Findings", "Recommended position summary", "What this skill does not do".
• Severity tags follow the rubric (**[Critical]**, **[Material]**, **[Minor]**).
• Citations reference specific clause numbers or page locations in the source.
• Output starts with "Bottom line" — two-three sentences leading with recommendation.

Expected calibration:

• 0 critical findings (this is a baseline document).
• 0–2 material findings.
• 2–6 minor findings (typically: routine clause-language preferences, optional carveouts not present, governing-law observations).
• "Bottom line" reads as "recommend executing as-drafted with optional minor edits."
• No invented authorities; any citation to law is cleanly referenced.

Edge cases to verify:

• Skill does not flag standard provisions (e.g., a 2-year survival period for a generic business-discussion NDA) as critical.
• Skill identifies the document as mutual (not unilateral) and applies mutual-perspective calibration.

Pass criteria:

• Structural pass: All required sections present; severity tags formatted correctly; citations resolve.
• Calibration pass: Reviewing attorney confirms the finding density and severity distribution is consistent with how they would review a routine mutual NDA.

Scenario 2: Unilateral NDA, discloser perspective

Inputs: A unilateral NDA where the user's organization is the disclosing party. Perspective: discloser.

Expected output structure:

• Same structure as Scenario 1.
• Findings calibrated to discloser concerns: scope of permitted use, return/destruction obligations, survival, remedies for breach, residuals carveouts (if any) carefully evaluated.

Expected calibration:

• 0–2 critical findings (typically only if the recipient's residuals carveout is overbroad, the survival is too short, or the remedies are inadequate).
• 1–4 material findings (typically: permitted-use scope, return/destruction proof, third-party recipient handling, breach notification).
• 3–8 minor findings.
• "Bottom line" leads with discloser-side recommendation (negotiate, accept-with-edits, or accept-as-is depending on severity distribution).

Edge cases to verify:

• Skill identifies provisions favorable to the recipient and flags them at appropriate severity from the discloser's perspective.
• Skill does not lapse into recipient-favorable analysis mid-report.
• If the document includes a residuals carveout for M&A or technology-evaluation contexts, the skill calibrates the carveout against the deal type rather than reflexively flagging.

Pass criteria:

• Structural pass: As above.
• Calibration pass: Reviewing attorney confirms perspective-aware findings — specifically that asymmetric provisions favoring the recipient are surfaced at appropriate severity.

Scenario 3: Unilateral NDA, recipient perspective

Inputs: A unilateral NDA where the user's organization is the recipient. Perspective: recipient.

Expected output structure:

• Same structure as Scenario 1.
• Findings calibrated to recipient concerns: scope of confidential information definition (over-broad?), permitted-use scope (too narrow?), survival period (commensurate with risk?), residuals (allowed?), remedies (proportionate?), employee/contractor exposure.

Expected calibration:

• 0–2 critical findings (typically only if the confidentiality definition is unbounded, residuals are not allowed for routine business operations, or remedies include unilateral injunction without standard).
• 1–4 material findings.
• 3–8 minor findings.
• "Bottom line" leads with recipient-side recommendation.

Edge cases to verify:

• Skill flags overly broad "confidential information" definitions (e.g., "all information disclosed in any form") at appropriate severity from recipient perspective.
• Skill addresses residuals carveouts as material from the recipient's side.
• Skill does not lapse into discloser-favorable analysis mid-report.

Pass criteria:

• Structural pass: As above.
• Calibration pass: Reviewing attorney confirms perspective-aware findings — specifically that asymmetric provisions favoring the discloser are surfaced at appropriate severity.

Scenario 4: M&A diligence NDA

Inputs: An NDA scoped to M&A diligence (residuals carveout, longer survival on commercially sensitive information, broader permitted purposes). Perspective: mutual or whichever perspective applies.

Expected output structure:

• Same structure as Scenario 1.
• The skill recognizes the M&A diligence context (either via the document's framing or via an optional deal_type input) and calibrates accordingly.

Expected calibration:

• 0–1 critical findings (M&A NDAs are typically negotiated to a market position; substantive concerns are usually around residuals and post-deal-collapse handling).
• 1–3 material findings around residuals scope, post-collapse return/destruction, non-solicitation if present.
• 2–6 minor findings.

Edge cases to verify:

• Skill does not flag the residuals carveout reflexively (it is expected in M&A diligence).
• Skill identifies non-solicitation provisions if present and addresses them at appropriate severity.
• Skill addresses standstill provisions if present.

Pass criteria:

• Structural pass: As above.
• Calibration pass: Reviewing attorney confirms M&A-context-aware calibration.

Scenario 5: Vendor NDA with data-handling complications

Inputs: A vendor NDA where the recipient (vendor) will handle some sensitive customer information as part of the engagement. Perspective: discloser (the user is the customer providing information).

Expected output structure:

• Same structure as Scenario 1.
• Findings calibrated to discloser-as-customer with vendor handling sensitive data: data-handling provisions, security requirements, breach notification, return/destruction, third-party processor handling, sub-contractor approval.

Expected calibration:

• 0–2 critical findings if security requirements or breach notification are missing.
• 1–4 material findings around data-handling specifics.
• 2–6 minor findings.
• "Bottom line" notes whether the NDA alone is sufficient or whether a separate Data Processing Agreement (DPA) should be required.

Edge cases to verify:

• Skill recognizes that a vendor handling sensitive data may require a DPA in addition to the NDA, and surfaces this in the recommendations.
• Skill does not duplicate DPA Checklist Review's analysis; the cross-pointer is the appropriate response.

Pass criteria:

• Structural pass: As above.
• Calibration pass: Reviewing attorney confirms appropriate cross-pointer to DPA Checklist Review when data handling is in scope.

Scenario 6: Notably unusual NDA (intentional edge cases)

Inputs: An NDA with one or more unusual structural elements: 5+ year survival period, broad employee non-solicitation, jurisdiction selection in an unusual venue, or an unusually narrow permitted purpose. Perspective: as appropriate.

Expected output structure:

• Same structure as Scenario 1.
• Each unusual element surfaces as a finding with explicit severity calibration.

Expected calibration:

• The unusual elements drive the severity distribution: 1–3 critical findings on the unusual elements; 1–3 material findings on supporting concerns.
• The "Bottom line" reflects the unusual posture (typically "negotiate or escalate" rather than "execute as-drafted").

Edge cases to verify:

• Skill explicitly identifies the unusual structural elements rather than treating them as routine.
• Skill provides recommended language for each unusual element (where applicable).
• Skill defers enforceability questions ("the 7-year survival period is unusual; enforceability of overbroad survival periods is jurisdiction-specific and warrants escalation") rather than asserting enforceability outcomes.

Pass criteria:

• Structural pass: As above.
• Calibration pass: Reviewing attorney confirms each unusual element is surfaced at appropriate severity and with appropriate enforceability deferral.

Refusal scenarios

Refusal 1: Document is not an NDA

Input: An MSA, employment agreement, or other contract type misidentified as an NDA.

Expected behavior:

• Skill identifies that the document is not (primarily) an NDA.
• Skill declines to apply NDA-specific analysis.
• Skill recommends the appropriate alternative skill (e.g., MSA Review — SaaS for an MSA; "no specific skill applies" for an employment agreement).
• Skill does not produce findings as if the document were an NDA.

Pass criteria: Skill explicitly refuses with a clear explanation.

Refusal 2: Document is not in English

Input: A non-English NDA (e.g., a French or Spanish NDA).

Expected behavior:

• Skill identifies that the document is not in English.
• Skill notes that the skill is calibrated to English-language NDAs and US-law / common-law conventions.
• Skill asks the user to confirm before proceeding (rather than silently producing output that may be miscalibrated for the document's actual jurisdiction).

Pass criteria: Skill explicitly notes the language and seeks confirmation.

Cross-cutting verification

These checks apply across all scenarios:

• No invented authorities. Every citation to law (case law, statute, regulation) is real and accurately characterized. The skill does not cite cases or statutes that do not exist.
• No enforceability opinions. The skill does not assert "this clause is unenforceable" or "this clause will not hold up." Enforceability deferred to "this is unusual" or "this raises an enforceability question."
• "What this skill does not do" enumeration is present. The output lists the out-of-scope items (typically: jurisdiction-specific enforceability analysis, transactional structuring, employment-law-specific implications of non-solicitation provisions, etc.).
• Recommended language is operationally usable. Recommended replacement language is clean, ready to drop in with minor party-name edits.
• Citations resolve. Every citation in the report points at a clause or page in the source document and the verbatim quote (where present) appears verbatim in the source.
• Conservative posture maintained. No findings overclaim outside the skill's documented scope.

Pass / fail decision

NDA Review v1.0.1 passes acceptance testing when:

1. All 6 test scenarios pass structural checks.
2. All 6 test scenarios pass calibration evaluation by a reviewing attorney.
3. Both refusal scenarios trigger the documented refusal behavior.
4. Cross-cutting verification passes on every scenario.

Any failure category is filed as a GitHub Issue with the acceptance-test-fail label. Issues block the v1.0.1 release until resolved. Resolved issues are closed with a link to the fix PR; the skill is re-tested against the affected scenarios after the fix.

Reviewer notes

The reviewing attorney for NDA Review acceptance testing should be familiar with both customer-side and vendor-side NDA negotiation. Specific competencies:

• Distinguishing M&A diligence NDAs from vendor procurement NDAs (different calibrations apply).
• Calibrating residuals carveouts (acceptable in M&A; problematic in vendor relationships).
• Recognizing standard vs. unusual survival periods in different deal contexts.
• Identifying data-handling provisions that require cross-reference to DPA Checklist Review.

Reviewer documents calibration assessment in test-results/nda-review-v1.0.1/calibration-assessment.md following the template in acceptance-testing-framework.md.