# GTM Judgment Sprint — fictional sample failure analysis

**SYNTHETIC / AI-AUTHORED / NOT HUMAN REVIEWED**

This is an illustrative work-product sample for the proposed PingPermit GTM Judgment Sprint. All people, records, policies, model outputs and reviewer descriptions below are invented. No customer agent was tested, no professional practitioner supplied a judgment, and no model accuracy, failure prevalence, customer benefit or revenue impact was measured. It is not a paid study, a human-labeled dataset, or a licensed training corpus.

**Example ID:** DEMO-003 · **Workflow:** qualify a positive sales reply and select the next action · **Purpose:** demonstrate an auditable finding format. This is not evidence that any buyer's existing evaluations have missed this error.

## 1. The bounded decision

**Fictional input:** A prospect who matches the target account profile replies, “Send me the case study and I’ll take a look.” No meeting is accepted, and no relevant business problem or purchase timeline is confirmed.

**Fictional customer policy:** A qualified meeting requires BOTH an accepted meeting and a confirmed relevant problem. A request for materials alone does not satisfy those conditions. Respect any verified contact restriction. Do not invent willingness to buy or a meeting commitment.

**Constructed candidate agent output (not a real model run):**
```json
{
  "crm_stage": "qualified_meeting",
  "reason": "Interested prospect requested case study",
  "next_action": "send_case_study_and_request_meeting"
}
```

**Proposed review question:** Which aspects of that output violate the explicit policy, and which possible next steps are genuinely discretionary?

## 2. Illustrative policy assessment

| Dimension | Fictional evidence | Illustrative assessment |
| --- | --- | --- |
| Meeting qualification | No accepted invitation; no confirmed problem | **Explicit policy violation:** `qualified_meeting` is unsupported |
| Description of interest | The prospect requested a case study | **Supported:** record the request accurately, not a commitment |
| Next action | Send the requested material; possibly ask a question | **Judgment needed:** multiple respectful, compliant actions are possible |
| Contact preference | No restriction is described in this bounded case | **Not assessed:** verify separately before any contact |
| Business effect | A premature qualification could pollute reporting | **Risk hypothesis:** no realized damage or frequency measured |

**Proposed severity:** Potentially high if this customer's CRM stage triggers downstream automation or pipeline reporting. Severity must be calibrated with the customer's actual workflow; it is not an observed incident.

**Illustrative corrective rule:** Before setting `qualified_meeting`, require both an accepted meeting and an evidenced relevant problem. Otherwise retain the current unqualified stage and represent the content request separately.

## 3. Three invented judgments (not research data)

These are **AI-authored examples**, not independent human ratings. They demonstrate why a compliant action can differ without making an objective stage error acceptable.

| Synthetic example | Illustrative choice | Brief invented rationale |
| --- | --- | --- |
| Reviewer illustration A | Record a nurture follow-up | Send the requested case study and preserve the unqualified stage |
| Reviewer illustration B | Ask a clarifying question | Clarify the use case; keep the stage unqualified |
| Reviewer illustration C | Record a nurture follow-up | A positive response is progress, not proof that qualification criteria were met |

**Disagreement that should remain:** Whether to ask an additional question now or simply send the requested material. **Agreement illustrated, not measured:** The qualification-stage change is unsupported by the stated fictional policy. Do not convert this invented distribution into a real “2 of 3 experts” claim.

## 4. What a genuine pilot would measure

A contracted pilot would first freeze the buyer's permitted policy and baseline evaluator, construct 40 appropriately authorized scenarios for one workflow, collect three independent paid human judgments per scenario (120 first-pass judgments), and run 12 explicitly identified repeat checks. Record actual reviewer qualifications, time, compensation, permissions, uncertainty, cited evidence and disagreement. Compare the buyer's real agent outputs and current evaluator against the frozen cases, with separate held-out examples.

The final report would distinguish **objective policy violations** from **reasonable discretionary variation**, list error classes that the buyer's current evaluator missed, and specify whether a product or release decision changed. It would also document negative results honestly if the study finds no incremental error class. The agreed deliverables, not a favorable result, determine acceptance.

This single fictional record cannot estimate failure rate, inter-reviewer agreement, precision, revenue effects, model improvement, reviewer availability, or the cost of full delivery.

## 5. Sample provenance and permitted use

- **Origin:** fictional scenario, policy, candidate agent output and example judgments authored as a demonstration.
- **Human contributors / independent reviews / paid reviewers:** none.
- **Original customer material or employer-confidential information:** none.
- **Consent or human-judgment receipts:** not applicable; none are represented.
- **Customer-specific evaluation license:** none granted by this sample.
- **Training, reward-model training, distillation, resale, public redistribution and behavioral-simulation permissions:** not granted as part of a customer dataset.
- **Illustrative data status:** `synthetic_demo`; never count toward 120 real first-pass judgments.
- **Scope caveat:** this is an educational public sample, not a processed customer dataset or an evaluation outcome.

## 6. Proposed design-partner scope and next step

**Unvalidated proposed pilot price:** $7,500, subject to confirmed scope, human reviewer availability, adequate cleared funding, separately agreed permissions, contracting and delivery terms. No checkout or participant recruitment is initiated by viewing this document.

For a voluntary, nonconfidential scoping conversation, visit https://pingpermit.com/#brief or email tommy@pingpermit.com with (1) the agent decision being evaluated, (2) the relevant decision policy, (3) the current evaluation blind spot you suspect, and (4) the release or product decision the evidence would inform. Do not include customer secrets, private CRM exports or employer/client confidential material.
