# Lesson 6 practice: Compare explanations for waiting

[中文](README.md)

Open [waiting.csv](waiting.csv). Ask what to investigate next rather than claiming that every cause is known. These 12 independent synthetic records, W-01 to W-12, are not extracted from the introductory 20 tasks or the original reconstruction described in Beichen’s public event. Approval minutes were added to practice defining a denominator; they must not be presented as original event evidence.

## Fields

| Field | Meaning |
| --- | --- |
| `source_type` | `independent_synthetic_exercise` |
| `event_id` | Exercise ID, distinct from introductory E-IDs |
| `shift` | `day` or `night` |
| `search_wait_min` / `expert_wait_min` / `approval_wait_min` | Non-overlapping wait durations in minutes, a construction assumption of this exercise |
| `total_wait_min` | Sum of the three waits; the denominator used here |
| `access_issue` | `none`: no access issue assigned to this row; `permission_denied`: a denied-access condition |

An access condition does not measure minutes lost to permissions or prove that the denial was incorrect. For overlapping real waits, timestamps and deduplication would be needed. This table deliberately uses separate segments for addition.

## Work through the evidence

1. Write a hypothesis that evidence could support or challenge, such as “Better access to information can materially change task outcomes.” Add a competing explanation, such as approval or staffing constraints. A useful test need not overturn the original plan.
2. Check the row sums. Total waiting time is **610 minutes**, including **309 approval minutes**. Explicitly use all waiting minutes in this table as the denominator: 309 / 610 ≈ **50.7%**.
3. Compare shifts and rows with or without the assigned access issue. Keep each group’s count and total waiting time. Differences suggest questions, not causal conclusions. `none` does not establish that every other permission path was problem-free.
4. Treat “approval exceeds 40% of waiting time” as a **scenario decision threshold**. This exercise exceeds it under the stated denominator. Beichen’s public event supplied neither this table nor that denominator, so its original evidence cannot be said to establish 50.7%. The threshold is not a universal industry rule.
5. Compare three options: improve information and retrieval, change approval or staffing, or continue a constrained recommendation prototype. For each, identify the affected stage, cost, authorization dependencies, and missing evidence. Even if approval cannot change, explain whether another improvement could still help.

## Check your reasoning

Do not infer approval wait from whole-task duration or estimate a customer-wide distribution from these 12 constructed records. The retrieval score change from 71% to 84% belongs to a separate public scenario and lacks a complete denominator and evaluation setup. It cannot establish benefits for these rows.

Produce a hypothesis card with competing explanations, sample scope, calculations, next steps, and stopping conditions. Self-study can reach a proposed test; real user understanding, staffing capacity, and operational improvements still need validation. A peer can check whether another explanation still fits.

In the next lesson, take the proposed solution’s access, source-withdrawal, and authority dependencies into governance review.