Strong-Edit Report: Claude Beyond Code

Pipeline History

Stage Commit Words Change
Raw notes 9b042d2 727 —
/flesh-out 0667fe0 2279 3.1x expansion
Human edit 402be65 2093 -8%
More culling da57ea6 2007 -12% total from flesh-out
/review-steps ed9be10 2257 +12% (added METR, trust data, afterword)
/strong-edit working tree 1929 -15%

Provenance: Ideas and Arguments

All core ideas trace to the raw notes (9b042d2):

  • The leverage shift / fulcrum parallel
  • Blocker → enabler framing
  • The three skills and their properties (review/flesh-out/strong-edit)
  • kūchūsen metaphor
  • “Sensitivity to generated documents” caveat
  • “False sense of completeness” caveat (“An LLM will confidently proclaim…”)
  • Standout Step observation (“This is one of the most valuable stages I’ve found”)
  • Practitioner observations (scaffolding precision, term sensitivity, mode detection) — human experiential claims

Provenance: References

Reference Source
METR study Agent-surfaced in /review-steps, human already aware (HN)
Stack Overflow trust survey Agent (/review-steps)
Diataxis Agent (/review-steps)
Book Factory Agent (/review-steps)
Wispr Human (podcast, added via GitHub edit)
Bicameral AI Human (podcast, added via GitHub edit)

Provenance: Prose

Almost no sentences survive verbatim from the raw notes. Three agent passes (/flesh-out, /review-steps, /strong-edit) rewrote nearly everything. Specific attribution:

  • Opening (lines 32-34): Written by agent in /strong-edit session, replacing the original “surprising thing” hook. Human selected from three proposed options.
  • Leverage shift (lines 36-52): Human ideas in agent-constructed sentences, refined three times. The parallel structure (traditional vs agentic fulcrum) was in the raw notes. The “trap” framing and METR reference added by /review-steps. The enabler paragraph substantially rewritten in /strong-edit based on human direction about configuration management.
  • Skills introduction (lines 54-66): Agent-structured. The table format and Information/Noise/Refinement columns are agent-originated; the core skill descriptions are human. Docs-as-code prerequisite relocated here in /strong-edit.
  • Skill descriptions (lines 68-76): Detailed stage-by-stage breakdown agent-generated in /flesh-out. Division-of-labor framing agent-originated.
  • Caveats (lines 86-94): Caveats 1 and 3 human-originated from raw notes, agent-refined across passes. Caveat 2 (hallucinations/trust with survey data) agent-originated in /review-steps. Caveat 3 rewritten in /strong-edit per human direction.
  • Final Words (lines 96-100): kūchūsen metaphor human-originated. Agent expanded with full Japanese text and ground battle antonym. Closing sentence rewritten by agent in /strong-edit.
  • Afterword (lines 104-116): “This article” section written by agent in /strong-edit from verified git data. “In general” observations are human experiential claims, agent-structured prose, added in /review-steps.
  • See also (lines 134-139): Mix of agent-found (Diataxis, Book Factory) and human-sourced (Wispr, Bicameral AI) references.

Strong-Edit: Changes Made

Main Body

  1. Opening rewritten — soft “brings back the fun” hook replaced with thesis-first opening (“Productive developers have always resisted documentation…”)
  2. Leverage shift sentences tightened — overloaded sentences at lines 42-43 and 50 broken up into cleaner parallel structure
  3. Blocker/enabler hyperbole signaled — “Documentation was a blocker” changed from assertion to acknowledged rhetorical device
  4. False binary replaced — “Do you back and forth explain the idea in dribs and drabs…” reframed around configuration management vs ephemeral conversation
  5. Docs-as-code relocated — standalone section removed, folded into skills introduction as prerequisite
  6. Pipeline ordering corrected — “runs in order” changed to “typical pipeline… but skills can be applied in any order”
  7. Throat-clearing removed — “I’ve stated that I think you should be writing documents…” paragraphs cut
  8. “Key principle” label downgraded — kept the idea, removed the overclaim
  9. Caveat 3 sharpened — vague defensive framing replaced with direct statement about AI rejection
  10. Ending rewritten — generic “start of the era” proclamation replaced with callback to idea and the article’s actual scope
  11. Bold/italic consistency — code, idea, mental clarity, agentic clarity usage made consistent across leverage shift section and bookending opening/closing
  12. Writing transition improved — “Across all three skills, the division of labor is the same” bridges from pipeline to detail

Afterword

  1. AGENTS.md bullet merged with scaffolding precision to eliminate “clearer contract” repetition
  2. Context awareness bullet cut — vague, no actionable content
  3. Three tail sections collapsed into single “See also”
  4. Future Extension preserved as own section (human’s original idea, not a reference)
  5. Run-on at line 126 fixed — split into two sentences, “on tangent” corrected
  6. Diataxis “homework for me” admission cut
  7. “detecting requirement” typo fixed → “detecting requirement gaps”
  8. Subject-verb agreement fixed in Future Extension
  9. “This article” section rewritten with verified git history data

Rough Attribution

  Human Agent Co-created
Ideas and arguments ~95% ~5% —
Research and references ~50% ~50% —
Prose (the actual sentences) ~5% ~60% ~35%
Editorial decisions 100% 0% —