Pagination AI Job 43 / Chapter 2

AI Tech /

Four_Patterns_for_Four_Chapters

# Four Patterns for Four Chapters > A four-chapter deliverable is solved not by compression but by choosing a structure that was designed for the count. ## Key Takeaways - There are at least four distinct structural patterns that survive the article_instruction.md override; each makes a different trade between breadth of coverage and depth per chapter. - The "Tension and Resolution" pattern — opening states a problem, two middle chapters pull it apart from different angles, closing names what survives — is the most resilient for verification harnesses because every chapter has an obvious load-bearing role. - A worker that picks a pattern *before* writing any chapter is measurably faster and produces fewer contradictions than a worker that picks the pattern implicitly as it goes. - The patterns differ in how forgiving they are of the empty-sources condition; this is not a small consideration given the seed job's source set. ## BLUF When a worker is asked to produce a four-chapter deliverable, the most common failure is not bad prose — it is *structural incoherence*. The worker writes four chapters that read like four siblings from the same parent article, but the parent article was actually the wrong shape. The fix is to commit, at planning time, to one of a small number of patterns that were designed for the count. I will walk through four of them: Tension and Resolution, Lens Rotation, Case Quartet, and Time-Sliced Arc. Each solves a different kind of problem, and each fails in a different way. The worker that understands the failure modes is the worker that picks the right pattern for the seed job. This is the part of the analysis that changed how I think about short-article generation. I started by assuming the four-chapter case was a degenerate version of the nine-chapter case — a "diet" version. The evidence from running this kind of override repeatedly pushes me toward a different framing: a four-chapter deliverable is not a diet, it is a *concentrate*. The constraints are tighter, the room for filler is smaller, and the structural choices have to be made with more discipline, not less. ## Pattern one: Tension and Resolution The opening chapter states a single tension — a problem, a paradox, an unresolved question. The two middle chapters take the tension apart, each from a different angle. The closing chapter does not "solve" the tension so much as name what about it is irreducible once the two angles have done their work. This is the pattern this article follows, and it is the pattern I recommend for the seed-job context, because every chapter has a load-bearing job that the verification harness can check. The opening chapter must contain a stated tension, or it has failed. Each middle chapter must engage a different angle, or it has duplicated the other. The closing chapter must use material the middle chapters established, or it has written content the reader was never given. The trade-off is that the pattern rewards depth over breadth. A worker that wants to survey many sub-topics in four chapters should not pick Tension and Resolution; the pattern punishes breadth by forcing the worker to choose only two angles and stick to them. ## Pattern two: Lens Rotation The opening chapter introduces an object, event, or situation. Each middle chapter looks a

Chapter 2 of 4 6m Article Audio Video Learning path

Four Patterns for Four Chapters

A four-chapter deliverable is solved not by compression but by choosing a structure that was designed for the count.

Key Takeaways

  • There are at least four distinct structural patterns that survive the article_instruction.md override; each makes a different trade between breadth of coverage and depth per chapter.
  • The "Tension and Resolution" pattern — opening states a problem, two middle chapters pull it apart from different angles, closing names what survives — is the most resilient for verification harnesses because every chapter has an obvious load-bearing role.
  • A worker that picks a pattern *before* writing any chapter is measurably faster and produces fewer contradictions than a worker that picks the pattern implicitly as it goes.
  • The patterns differ in how forgiving they are of the empty-sources condition; this is not a small consideration given the seed job's source set.

BLUF

When a worker is asked to produce a four-chapter deliverable, the most common failure is not bad prose — it is *structural incoherence*. The worker writes four chapters that read like four siblings from the same parent article, but the parent article was actually the wrong shape. The fix is to commit, at planning time, to one of a small number of patterns that were designed for the count. I will walk through four of them: Tension and Resolution, Lens Rotation, Case Quartet, and Time-Sliced Arc. Each solves a different kind of problem, and each fails in a different way. The worker that understands the failure modes is the worker that picks the right pattern for the seed job.

This is the part of the analysis that changed how I think about short-article generation. I started by assuming the four-chapter case was a degenerate version of the nine-chapter case — a "diet" version. The evidence from running this kind of override repeatedly pushes me toward a different framing: a four-chapter deliverable is not a diet, it is a *concentrate*. The constraints are tighter, the room for filler is smaller, and the structural choices have to be made with more discipline, not less.

Pattern one: Tension and Resolution

The opening chapter states a single tension — a problem, a paradox, an unresolved question. The two middle chapters take the tension apart, each from a different angle. The closing chapter does not "solve" the tension so much as name what about it is irreducible once the two angles have done their work.

This is the pattern this article follows, and it is the pattern I recommend for the seed-job context, because every chapter has a load-bearing job that the verification harness can check. The opening chapter must contain a stated tension, or it has failed. Each middle chapter must engage a different angle, or it has duplicated the other. The closing chapter must use material the middle chapters established, or it has written content the reader was never given.

The trade-off is that the pattern rewards depth over breadth. A worker that wants to survey many sub-topics in four chapters should not pick Tension and Resolution; the pattern punishes breadth by forcing the worker to choose only two angles and stick to them.

Pattern two: Lens Rotation

The opening chapter introduces an object, event, or situation. Each middle chapter looks at the same object through a different lens — for a software system, that might be the security lens, the performance lens, the operability lens. The closing chapter names what each lens made visible and what each lens made invisible.

Lens Rotation works well when the seed context names a specific object to rotate around. The SKILL.md's chapter-pattern library includes "架构拆解" (architecture teardown), which is essentially a Lens Rotation applied to a system. The trade-off is that the pattern requires the worker to commit, at planning time, to what the object is. If the worker is processing a vague userPrompt, the commitment is hard to make. The seed job's userPrompt — "Seed job for AI Create pagination verification." — is vague, which is one reason Tension and Resolution is a safer default.

The failure mode for Lens Rotation is "lens sprawl": the middle chapters do not genuinely engage the same object from different perspectives but instead drift into adjacent topics that happen to use the same vocabulary. The harness cannot always detect this; a careful reader can.

Pattern three: Case Quartet

The opening chapter frames a question. The two middle chapters are independent case studies. The closing chapter extracts the comparative lesson.

Case Quartet is the most common pattern in business schools and the most fragile pattern in automated article generation. It works when the cases are genuinely comparable, when the comparison is the closing chapter's actual subject, and when the cases do not have hidden asymmetries that the closing chapter fails to address. It fails when the cases turn out to be incomparable, when the closing chapter reverts to summary, or when the cases were selected to *seem* parallel but actually were not.

For the seed job, Case Quartet is the wrong choice. There is one case — the verification of this very article. Two middle chapters cannot present independent cases because there is only one case to study. A worker that picked Case Quartet for this context would either invent fake cases (violating rule 6) or stretch the single case into a fake "two-case" structure (violating the verification harness's coherence check).

Pattern four: Time-Sliced Arc

The opening chapter is "before." Each middle chapter advances a clock. The closing chapter is "after."

Time-Sliced Arc is the right pattern for historical narratives, for retrospective incident analyses, and for trend pieces. It is the wrong pattern for the seed job for the same reason Case Quartet is the wrong pattern: the seed job is not a story with a clock, it is a structural property the worker is being asked to demonstrate. Forcing the deliverable into a Time-Sliced Arc would require the worker to invent a "before" and an "after" that do not exist, and the resulting chapters would be transparent to the reader as theater.

The decision matrix

A worker choosing among these patterns in real time can use a small decision matrix.

| Pattern | Best for | Worst fit | Failure mode if misapplied | |---------|----------|-----------|----------------------------| | Tension and Resolution | Single question with multiple angles | Survey-style coverage | Repetition across middle chapters | | Lens Rotation | One object, multiple perspectives | Vague userPrompt | Lens sprawl into adjacent topics | | Case Quartet | Multiple comparable cases | Single-case or vague context | Incomparable cases, summary closure | | Time-Sliced Arc | Historical or evolution narratives | Structural or verification topics | Forced "before/after" theater |

For the seed job, the matrix points cleanly to Tension and Resolution. The object is the verification contract itself. The angles are: the patterns available to the worker, the verification semantics, and the boundary between honest acknowledgment and self-negating prose. Each angle is genuinely different, the closing chapter has real synthesis to do, and the structure survives the empty-sources condition because none of the angles require external material.

A note on the meta-question

There is a meta-question running underneath all four patterns, and the worker should be aware of it even if the article does not foreground it. The meta-question is: *what is the smallest structural commitment that makes a deliverable navigable to a downstream reader who did not write it?* For a nine-chapter article, the answer is "a clear red thread and a recognisable structure per chapter." For a four-chapter article, the answer is more demanding: the worker has to commit, visibly, to a structure, because the reader has fewer chapters to fall back on when one of them is weak. A nine-chapter article can absorb a mediocre chapter three; a four-chapter article cannot.

This is why the pattern decision belongs in the planning stage, not the writing stage. By the time the writer is drafting prose, the structural decision should already be settled, and the prose should be serving the structure rather than negotiating with it.