The problem
The phase and flow files carry full narrative case studies inline. frameworks/delegation/phases/7-diagnose.md is a 205-line file, and roughly 35 of those lines are three case studies:
- "Real-world case: Nicole Wipp — the first-hire failure that was actually a clarity failure" (two paragraphs, plus a COVID-era anecdote about a law firm operating from Hawaii)
- "Real-world case: Sanchez — raised dependents, not leaders" (three paragraphs, multiple block quotes from YouTube shorts)
- "Source gap acknowledged: accountability diffusion in flat / co-founder teams" (three paragraphs explaining what is missing from the vault)
frameworks/delegation/phases/4-build-brief.md carries two more (the Kyle TPS report story, and Nicole Wipp again — the same case, re-told at similar length in a second file).
What happened in practice
I loaded Phase 7 to run a post-mortem. I read all three case studies. I used zero of them. Not because they are bad — the Wipp diagnosis is genuinely the right lens for a "they dropped the ball" situation that is actually a clarity failure — but because the operative content was already in the file above them: the five structural gaps, their signal phrases, the failure-mode table, and the corrective-move table. Those are what I applied. The case studies were confirmation of things the tables had already told me.
The same session also loaded Phase 4 and Phase 5, and hit the Wipp case a second time in Phase 4, told slightly differently. Duplicated narrative across files means the redundancy scales with how many phases a flow touches.
The cost is not just tokens. It is that the operative material — five gaps, one table, one corrective per gap, which is genuinely tight and genuinely good — is buried in the middle of a long file. The signal-to-length ratio of these files is much worse than the quality of their thinking deserves.
Why the "Going deeper" pattern is already the answer
Every phase file already ends with exactly the right idea:
Going deeper. Everything you need to coach this section is above. If the user asks where this comes from, or you need a more detailed take, load:
That is the correct architecture, stated explicitly, on the same page as the thing that violates it. The reference vault exists. The pointers exist. The case studies should be behind them, not in front.
What good looks like
- Move the inline case studies into the
references/ files they cite. sullivan_who-not-how.md already exists and is the natural home for the Wipp case. sanchez_main-street-millionaire.md already exists and is where the Sanchez material belongs. Several of these blocks already end with "Source: [file]" pointing at exactly the file they should live in.
- Leave a one-line hook in the phase file. For example: "When a delegation failure looks like a personnel problem, it is usually an externalization problem — see the Nicole Wipp case in
references/sullivan_who-not-how.md." One line preserves the pointer at 3% of the length.
- Move the "Source gap acknowledged" sections out entirely. Notes about what the vault is missing are maintainer notes. They belong in the reference file's "Known gaps" section — which the note itself says already exists — not in the coaching path.
- De-duplicate across phases. The Wipp case should exist once.
Estimated effect on Phase 7: 205 lines down to roughly 165, with nothing lost that a coach needs mid-session.
The problem
The phase and flow files carry full narrative case studies inline.
frameworks/delegation/phases/7-diagnose.mdis a 205-line file, and roughly 35 of those lines are three case studies:frameworks/delegation/phases/4-build-brief.mdcarries two more (the Kyle TPS report story, and Nicole Wipp again — the same case, re-told at similar length in a second file).What happened in practice
I loaded Phase 7 to run a post-mortem. I read all three case studies. I used zero of them. Not because they are bad — the Wipp diagnosis is genuinely the right lens for a "they dropped the ball" situation that is actually a clarity failure — but because the operative content was already in the file above them: the five structural gaps, their signal phrases, the failure-mode table, and the corrective-move table. Those are what I applied. The case studies were confirmation of things the tables had already told me.
The same session also loaded Phase 4 and Phase 5, and hit the Wipp case a second time in Phase 4, told slightly differently. Duplicated narrative across files means the redundancy scales with how many phases a flow touches.
The cost is not just tokens. It is that the operative material — five gaps, one table, one corrective per gap, which is genuinely tight and genuinely good — is buried in the middle of a long file. The signal-to-length ratio of these files is much worse than the quality of their thinking deserves.
Why the "Going deeper" pattern is already the answer
Every phase file already ends with exactly the right idea:
That is the correct architecture, stated explicitly, on the same page as the thing that violates it. The reference vault exists. The pointers exist. The case studies should be behind them, not in front.
What good looks like
references/files they cite.sullivan_who-not-how.mdalready exists and is the natural home for the Wipp case.sanchez_main-street-millionaire.mdalready exists and is where the Sanchez material belongs. Several of these blocks already end with "Source: [file]" pointing at exactly the file they should live in.references/sullivan_who-not-how.md." One line preserves the pointer at 3% of the length.Estimated effect on Phase 7: 205 lines down to roughly 165, with nothing lost that a coach needs mid-session.