# Mediation Problem Validator > Evaluate completed mediation-competition problems consisting of general information and confidential packets for both sides. Use when asked to validate, audit, score, quality-control, or assess competition readiness; identify factual, numerical, chronological, authority, commercial-logic, or confidentiality defects; test mediation usefulness, BATNAs, settlement corridors, information asymmetry, and opportunities for firmness, cooperation, imagination, and clarification; or produce an author-only validation report. Perform a qualified partial review when packets are missing. Diagnose by default; do not generate a new problem or edit source packets unless repairs are expressly requested. Requires access to all supplied files and a host capable of reading their formats; otherwise disclose the unreviewed material and resulting limits. - Skill: `gabrielmoreira/mediation-problem-validator` (Agent Skill, multi-file: 6 files) - Install (CLI): `npx skillmds@latest add gabrielmoreira/mediation-problem-validator` - Raw SKILL.md: https://api.skillmd.com/api/skills/gabrielmoreira/mediation-problem-validator/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Security - Author: gabrielmoreira (https://skillmd.com/u/gabrielmoreira) - Updated: 2026-09-09 - Page: https://skillmd.com/skills/gabrielmoreira/mediation-problem-validator --- # Mediation Problem Validator Evaluate the three packets as views of one case. Treat internal consistency as a mandatory gate, then assess mediation usefulness and pedagogical quality. Never allow a high score to offset a contradiction, broken calculation, confidentiality leak, impossible authority, or unavailable settlement. ## Load the Required Guidance Read `resources/validation-rubric.md` completely before evaluating any packet. Read `resources/report-template.md` before drafting the final report. ## Establish Scope For a complete validation, require: 1. general information available to both parties; 2. Party A confidential information; and 3. Party B confidential information. If a packet is missing, perform only a partial audit. State exactly which gates and simulations cannot be completed. Do not infer the missing party's information or issue an unqualified competition-ready conclusion. Before promising a complete validation, confirm that the host can read every supplied file and inspect every page. Use whatever document-reading capability the host provides. If a format cannot be read reliably, ask the user for extracted text or a supported conversion; identify the missing content and issue only a qualified partial review. Record exact filenames and versions. Cite packet sections and page numbers where available. Treat the validation report, reconstructed model, candidate packages, and cross-packet findings as author-only. Do not merge confidential packets into a participant-facing artifact. ## Reconstruct the Case Before Judging It Build compact working tables for: - entities, names, roles, defined terms, assets, and jurisdictions; - objective chronology and deadlines; - every material number, its type, owner, visibility, timing, and calculation; - each representative's authority, resources, approvals, and protected interests; - each material fact's status and visibility; and - claims, interests, BATNAs, WATNAs, desired information, and suspected facts. Classify propositions as `objective`, `allegation`, `belief`, `legal_uncertainty`, `deliberate_ambiguity`, or `author_only`. Classify visibility as public, Party A only, Party B only, shared private, or author only. Do not silently reconcile conflicting source text. Log the conflict and explain its consequence. ## Apply Mandatory Gates Check: - identity and role consistency; - factual and chronological consistency; - numerical and accounting integrity; - commercial logic and causal plausibility; - authority, ownership, and third-party dependencies; - epistemic consistency; and - the information firewall. Recalculate every material total, percentage, installment, valuation, loss, ownership share, offset, and settlement range. Test timing, not merely nominal totals. Distinguish cash from accounting value, gross from net, revenue from profit, and purchase price from settlement proceeds. Test for double counting. Assign severity: - **High:** objective contradiction, broken math, confidentiality leak, impossible authority, missing commercial premise, or a defect making sound settlement impossible or obvious. The packet fails until repaired. - **Medium:** material confusion, imbalance, unrealistic conduct, weak implementation, or avoidable clarification request. Ordinarily repair before release. - **Low:** nonmaterial wording, redundancy, or typography. Repair if practicable. Any unresolved high defect blocks a `competition_ready` result. ## Test Mediation Usefulness Determine whether: - the public dispute is understandable without specialized outside law; - legal uncertainty creates risk but does not consume the exercise; - the case is sufficiently defined for productive bargaining; - at least three genuine bargaining dimensions exist; - each dimension connects something one party values to something the other controls; - each BATNA is realistic rather than artificially catastrophic or dominant; - neither party dominates legal rights, money, time, information, and alternatives simultaneously; and - participants can improve outcomes without first resolving who would win adjudication. Distinguish: - accidental ambiguity requiring repair; - deliberate legal uncertainty; - factual uncertainty that can be investigated or priced; - private information that must be elicited; and - uncertainty best managed through a contingency. ## Test the Settlement Corridor Classify the corridor: - `impossible`: no package satisfies both sides and outperforms both BATNAs; - `brittle`: only one hidden package or number works; - `narrow_sound`: two to four materially different packages require linked concessions and information exchange; - `too_easy`: obvious matches, mirrored menus, or generous overlap make agreement nearly automatic. Require `narrow_sound` for a presumptive pass. Construct two to four author-only candidate packages and at least one tempting but unsound package. Test every candidate for: - authority and approvals; - financeability and payment timing; - legal and operational availability; - protected interests; - third-party rights; - at least two linked concessions; - implementation, monitoring, cure, and exit terms; - downside scenarios; and - superiority to both BATNAs. Do not assume an attractive term is available merely because it would solve the case. ## Test Opportunities to Demonstrate Skills Assess Party A and Party B separately. ### Firmness Require a principled protected interest, authority boundary, or BATNA-supported reason to reject an attractive proposal while offering a workable alternative. ### Cooperation Require consequential opportunities for selective disclosure, acknowledgement, agenda design, reciprocal exchange, assistance, joint testing, monitoring, or implementation. ### Imagination Require raw materials for packages beyond a cash midpoint: contingencies, options, staging, licenses, substitute performance, benchmark pricing, revised risk allocation, future business, or implementation design. The packets must supply ingredients, not matching answer menus. ### Clarification For each side, identify at least five questions a skilled representative should ask. At least two answers must materially change which package is sound, and the other side's packet must contain responsive information that may properly be disclosed. Complete a skill-opportunity matrix showing the opportunity, packet evidence, and material consequence for each party and each skill. ## Run Role-Limited Audits When independent agents are available, use: - Party A reviewer: general packet plus Party A packet only; - Party B reviewer: general packet plus Party B packet only; - mediator reviewer: general packet only; and - full-package reviewer: all packets and author-only working tables. Give role-limited reviewers no diagnosis or expected answer. Instruct them not to inspect unprovided files. If true isolation is unavailable, treat the exercise as an independence aid rather than a security boundary. Ask party reviewers to identify interests, constraints, desired information, BATNA, at least two plausible bargaining directions, one tempting but unsound agreement, and clarification questions. Ask the mediator reviewer whether the public case is coherent, balanced, adequately defined, and productive without disclosing the solution. Ask the full reviewer to test contradictions, leaks, feasibility, balance, candidate packages, and the single defect most threatening the exercise. If independent agents are unavailable, perform the readings sequentially with fresh notes and strict information partitions. ## Forecast Clarifications Simulate at least ten likely participant or judge questions. Classify each: - `answer_in_packet`; - `repair_before_release`; - `facts_sufficient`; - `information_to_elicit`; or - `contingency_candidate`. Do not use `facts_sufficient` to conceal broken math, conflicting facts, or a missing commercial premise. ## Score After the Gates Use the 30-point rubric in `resources/validation-rubric.md`. Scoring is diagnostic and never overrides a failed gate. A presumptive `competition_ready` result requires: - all mandatory gates passed; - no unresolved high defect; - no scored dimension below 2; - at least 24/30; - a `narrow_sound` corridor; - two to four sound packages; - at least one tempting but unsound package; - both party-limited audits identifying at least two plausible directions; and - accidental clarification risks repaired. Use `revise_and_reaudit` when repairable defects remain. Use `unsuitable` when the economic or pedagogical engine cannot be repaired without materially redesigning the problem. ## Iterate Only When Authorized By default, diagnose and report; do not edit source packets. If the user requests repairs: 1. revise the underlying source-of-truth model; 2. propagate every affected change across all packets; 3. rerun calculations, firewall checks, corridor tests, and role audits; and 4. require two consecutive rounds with no high or newly discovered high defect. Do not repair one sentence in isolation when the change affects economics, authority, chronology, or knowledge elsewhere. ## Deliver Produce an author-only validation report following `resources/report-template.md`. Lead with one conclusion: - `competition_ready`; - `revise_and_reaudit`; - `partial_review_only`; or - `unsuitable`. List high defects first, with exact source locations, consequences, and repairs. Separate proven contradictions from judgment calls. Explain every score with packet evidence. Keep proposed candidate packages author-only. When the user requests an editable artifact, deliver Markdown by default. Create DOCX or PDF only when the host provides a document-generation capability with visual verification. Otherwise deliver Markdown and state that the requested conversion could not be completed in the current host. ## Jurisdiction and professional use The validation method is jurisdiction-neutral because it primarily tests internal consistency, commercial soundness, information design, mediation usefulness, and teaching opportunities. It does not determine whether a packet's substantive law is correct in a particular jurisdiction. When doctrine, procedure, ethics rules, or enforceability matters to the exercise, require authoritative jurisdiction-specific sources or flag the legal premise for independent review. ## Bundled resources - `resources/validation-rubric.md` — reconstruction tables, mandatory gates, settlement-corridor tests, role-limited audits, 30-point scoring rubric, and exit checklist; read before evaluation. - `resources/report-template.md` — required structure for the author-only validation report; read before drafting the deliverable. ## Limitations and risks - A complete conclusion depends on complete, accurately extracted versions of the public packet and both confidential packets. Missing pages, OCR errors, or superseded versions can invalidate findings. - The score and corridor classification require expert judgment. They organize review but do not prove that judges, students, mediators, or competition administrators will reach the same result. - Role-limited reviews are independence aids, not security boundaries. A host with shared context may retain information from other packets even when instructed not to use it. - Candidate packages, reconstructed economics, and cross-packet findings are author-only. Their accidental disclosure can compromise the exercise. - Do not place privileged, confidential, personal, or client information into an AI host unless authorized and consistent with the host's privacy, retention, and security terms. - The report is pedagogical quality control, not legal advice, an enforceability opinion, or a prediction of an actual dispute. Human review remains necessary. This package contains no executable code. It performs no network calls, credential access, subprocess execution, or dynamic evaluation.