2026-04-03-yomiya-content-intake-benchmark

Yomiya Content Intake Flow Implementation Plan

For Claude: REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.

Goal: Replace the heavier intake benchmark with a simpler research-stage flow where humans preselect resources and the system only handles transcript acquisition, classification, and failure fallback.

Architecture: Keep value judgment outside the system intake flow. Treat human selection as the upstream gate, then make the docs describe only the downstream system responsibilities.

Tech Stack: Markdown docs in product/YomiyaContentSystem/archive/plans/ and the historical Yomiya documentation tree

Task 1: Update the design summary

Files:

Create: product/YomiyaContentSystem/archive/plans/2026-04-03-yomiya-content-intake-benchmark-design.md

Step 1: Write the simplified design

Cover:

why the old benchmark still over-modeled system-side gatekeeping
why human preselection should be treated as a prerequisite
why the system now focuses on transcript and classification

Step 2: Verify the file exists

Run: test -f product/YomiyaContentSystem/archive/plans/2026-04-03-yomiya-content-intake-benchmark-design.md Expected: success

Task 2: Rewrite the benchmark document as a lightweight flow

Files:

Modify: product/YomiyaContentSystem/30-Research-and-Evidence/foundations/yomiya-content-intake-benchmark.md

Step 1: Remove value-gating language

Delete or replace:

research_pool
formal_candidate
multi-stage intake value judgment

Step 2: Replace with the lightweight processing flow

Document:

minimal registration
transcript path selection
classification outputs
failure fallback

Step 3: Add the simplified Mermaid flowchart

Keep the diagram centered on:

human preselection
transcript path
classify
fallback

Task 3: Rewrite the workflow doc

Files:

Modify: product/YomiyaContentSystem/30-Research-and-Evidence/foundations/yomiya-content-research-workflow.md

Step 1: Change the recommended execution order

Replace the old benchmark-first wording with:

human picks a batch
system tries transcript
classify successes
record failures

Step 2: Replace the workflow diagram

The workflow should no longer imply that the system is rating resource value.

Task 4: Align the sample-list usage note

Files:

Modify: product/YomiyaContentSystem/30-Research-and-Evidence/samples/播客与视频竞品统一样本清单.md

Step 1: Update the usage guidance

Make future editors follow:

human preselection first
transcript first
classification second
do not overfill rows before text is available

Step 2: Verify the note is readable

Run: sed -n '170,188p' product/YomiyaContentSystem/30-Research-and-Evidence/samples/播客与视频竞品统一样本清单.md Expected: the note reflects the simplified flow