Authoring an opinions post

An opinions post is a personal take where the take itself is the point — not a feature, not a pattern, not a procedure. Length is whatever the take needs. Voice is yours, full stop. Posts can live in draft = true for weeks or months without that being workflow failure.

The form has a unique optimization target: willingness to post. Not length, not polish, not completeness. A half-polished take that ships beats a fully-polished take that never does. The form is named "opinions" rather than "essays" deliberately — to escape the connotations (rigor, length, defensibility) that would raise the bar past the point of usability.

This is also the form to use as a deliberate vehicle to grow as a writer — not as a performance vehicle. The LLM's role here is thinking partner, not first-drafter — it helps find the take (Phase 1 prompting) and clean up the prose (Phase 6 mechanical), but it does not write the post.

When to write an opinions post

Use this form when:

• You have a take you want to put words to.
• A perspective has been simmering long enough that you want to write it down.
• An opinion surfaced while drafting another form and wants to be pulled out as standalone.
• An external trigger (article, conversation, mid-work frustration) prompted a reaction worth keeping.

Don't use this form when:

• The post is about a feature you built. (Use a logbook post.)
• The post is an interactive demo built to understand a concept. (Use a concepts post.)
• The post is a prescriptive procedure. (Use a workflows post.)
• The value is mostly in pointing at external content. (Use a resources post.)

Wherever the impetus to write comes from, the handling is the same: seed → flesh into draft with LLM prompting (depth-dialed) → revise into something you're willing to post.

Author disciplines

Two practices live with the author, not with any LLM check. They're not gates — just observational nudges as you build the writing habit.

• Own the take. First-person — "I think," "in my experience," "I suspect" — rather than hedging into "people often say" or "many would argue." If the take is yours, claim it.
• Length follows the take. A three-sentence take is fine; a two-thousand-word take is fine. Padding for the appearance of seriousness is the failure mode. So is truncating because the take feels too short to count.

Other writing-quality concerns (specificity, internal consistency, overclaiming) live inside Phase 4's substantive feedback tiers, where they're applied on request rather than carried as standing rules.

The workflow

Seven phases, à la carte. You can enter at any phase and walk forward from there. "Slow-refinement" and "fast-spontaneous" are convenient labels for the all-phases and minimal-phases endpoints, but the real shape is a spectrum.

The asymmetry between Phase 4 and Phase 6 is deliberate. Substantive feedback names observations the user has to act on themselves (their voice, their judgment). Mechanical fixes are surface-level and unambiguous once approved, so the LLM implements them directly — faster, and not risky.

Phase 1 — Point-finding

Phase 1 is the conversation that helps you figure out what you're trying to say before drafting. Three prompting modes are in the active toolkit.

Provocation (default opener) — LLM proposes an extreme version of your own argument. Same direction, pushed along whatever dimension is in play, until it twists into something you didn't mean. You react against it; what you keep is your actual position, and what you reject locates where the absurd version stops matching your stake.

Provocation is the default opener because reacting to a distortion of your own take forces examination of what you actually believe more reliably than reflecting on flat substance.

Crucially: provocation is not a counter-argument. The LLM should not draft an opposing position. If the proposed extreme version disagrees with your direction, it has misread the mode.

Mirror-and-probe (the consolidator) — you share raw thoughts; the LLM restates ("what I'm hearing is...") and asks one focused follow-up; iterate. Used after provocation or multiple-framings to consolidate the resulting pushback into a clearer position.

Multiple framings (branching alternative to provocation) — the LLM offers three candidate angles for what you might be getting at; you pick the closest, refuse them all, or use one as a springboard. Useful when the seed is too thin to provoke against, or when the take has branched and you want help organizing.

Default chain for a Phase 1 session:

session start
    |
    v
Provocation        <- forces reactive examination
    |
    v
Mirror-and-probe   <- consolidates the pushback into a clearer position
    |
    v
[pause - user picks next]
    |       |
    v       v
Provocation    Multiple framings
    |       |
    v       v
Mirror-and-probe (in both branches)
    |
    v
[iterate; pause between cycles]
    |
    v
Summary on request    <- LLM restates the user's opinions clearly
    |
    v
End of Phase 1

Properties of the chain:

• Pause-and-ask between cycles — the LLM doesn't pick the next mode unilaterally.
• Mirror-and-probe is the consolidator after each provocation or framings move, not a standalone opener.
• Summary on request closes Phase 1. When you feel like you've thought through the take cleanly, ask for a summary — the LLM restates the surfaced opinions back to you in a clear list. That list is Phase 1's terminal output, ready for Phase 2 (structure) or Phase 3 (drafting) to pick up.

Direct interrogative questions ("what's the point of this?") are explicitly off-limits. They trigger school-essay defences and shut down thinking. Use the three named modes; nothing flatter.

Phase 2 — Structure pass

Talk through how to organize the take. Output is an outline. Skip if the structure is obvious from the Phase 1 summary or if the take is short enough that scaffolding is overhead.

Phase 3 — Drafting

You write. The LLM is silent unless asked. This is the form's non-negotiable rule: the take is yours, in your voice, written by you. The LLM never first-drafts opinion prose.

Phase 4 — Substantive feedback

You share a draft and request the tier you want. Tier definitions:

Tier 1 — default suggest. When you share a draft, the LLM asks if you want Tier 1 feedback. You decide per-post. The Tier 1 set:

• Clarity — can the LLM restate the central claim in one sentence? If not, what's blurry?
• Internal consistency — do later paragraphs contradict earlier ones?
• What's missing — is there an obvious counterargument or piece of context the post doesn't acknowledge?

Tier 2 — on request.

• Steelman — the strongest version of the opposing view. You decide whether your take survives.

Tier 2 is on-request specifically because steelmanning too early shuts an embryonic take down before it forms.

Tier 3 — demonstrate on a single paragraph on request.

• Specificity gap — pick a vague claim; suggest a concrete example or counterexample.
• Overclaiming / underclaiming — flag where the claim is stronger or weaker than the evidence supports.
• Emotional truth — does the prose feel lived or hollow?

Tier 3 is paragraph-scoped because applying these to a whole post is overload; applying them to one paragraph lets you carry the lesson forward to the rest yourself.

Tier 4 — deferred.

• Voice authenticity — needs a corpus of opinion posts to compare against. Revisit once around five posts exist.
• Audience match — needs an actual audience. Revisit when there are readers.

After feedback, you do the rework. The LLM stays silent unless asked.

Phase 5 — Revision

You revise. You can loop back to Phase 4 with a revised draft if you want another round. The LLM is silent unless asked.

Phase 6 — Mechanical pass

The LLM produces a list of mechanical issues:

• Grammar and typos
• Awkward phrasing at the sentence level (not structural)
• Broken or missing links
• Title revisit — a low-stakes glance: does the working title still fit the finished post? If nothing obviously better surfaces in a minute or two, ship the working title. Title perfectionism is a willingness-to-post killer dressed up as craft.

You approve or reject each item (or batch-approve). The LLM implements the approved fixes directly. This is the one phase where the LLM modifies the post for you — because mechanical fixes are surface-level and unambiguous once approved.

Phase 7 — Publish

Flip draft = false. Run zola check.

Entry points

Common entry points into the phase sequence:

The LLM figures out which entry point you're at via:

1. You declare. "Want feedback on this draft" → Phase 4. "Just check typos before I publish" → Phase 6. "Help me think through this" → Phase 1. Most sessions land here cleanly.
2. LLM asks if ambiguous. When you share an artifact without context, one short disambiguation question: "Where do you want to enter — point-finding, substantive feedback, or mechanical pass?"
3. Inference from artifact shape. A seed sentence → Phase 1. A finished-looking draft + "ready to publish" → Phase 6.

Anti-patterns

1. Optimizing for polish over willingness. The form's optimization target is willingness to post; the failure mode is polishing past the point of shipping. Half-polished that ships beats fully-polished that doesn't.
2. LLM first-drafting the prose. The take is yours. The LLM helps find it (Phase 1) and clean it up (Phase 6). It does not write it.
3. Treating draft = true as workflow failure. Posts living in draft for weeks or months is the slow flow working as intended, not a stuck workflow.
4. Steelmanning an embryonic take. Phase 4 Tier 2 is on-request because steelmanning too early kills a take before it forms.
5. Agonizing over the title. Phase 6 title revisit is a time-boxed glance. If no obviously-better title surfaces in a minute or two, ship the working title.
6. Adding school-essay scaffolding when the take doesn't need it. Intro / thesis / three body paragraphs / conclusion is the structure school taught — and the connotation the form was named away from. Some takes need structure; many don't. Forcing scaffolding onto a short take is a hedge against feeling exposed by brevity.

Worked example

A full slow-refinement walk-through on a single seed, followed by two compressed variants showing entry at Phase 3 and Phase 6.

The example seed is intentionally low-stakes and not a real take — substitute your own when running this for real.

Slow refinement (Phase 1 → 7)

Seed: I think people overrate the productivity boost from a second monitor.

Phase 1 — Point-finding:

You: I think people overrate the productivity boost from a second
     monitor.

LLM (provocation, opening):
  Extreme version: there is no productivity gain from second monitors
  at all. The entire dual-monitor industry is performative theater.
  Every developer who claims a real benefit is rationalizing a $300
  desk-decoration purchase into a productivity tool. Push back.

You: No - I do think there's a real gain, just smaller than the hype.
     The marketing acts like 5x productivity, when realistically it's
     maybe 5-10% for tasks that actually need two surfaces.

LLM (mirror-and-probe):
  What I'm hearing is that you accept the gain is real but reject the
  magnitude that's marketed - the take is about the gap between
  marketed and actual gains, not about second monitors being useless.
  Is that closer?

You: Yes, and also that for a lot of devs the gain is zero because
     they don't actually use both screens - one becomes a Slack and
     distraction parking lot.

LLM (pause):
  Take is sharpening. Two threads have surfaced: (1) the magnitude
  gap between marketing and real gains, (2) for many devs the second
  screen becomes a distraction sink. Next: another provocation,
  multiple framings to organize these, or summary?

You: summary

LLM (summary - closes Phase 1):
  Your opinions so far:
  1. Second monitors provide a real but modest gain (~5-10%), well
     below the marketed promise.
  2. The benefit materializes only when tasks genuinely use two
     surfaces.
  3. For many developers, the second screen becomes a Slack and
     distraction parking lot - zero or negative value.
  4. The marketing-vs-reality gap is the actual story you want to
     tell.

Phase 2 — Structure pass: discuss organization, output an outline. (Content varies per take; skipped here for space.)

Phase 3 — Drafting: you write the prose. LLM silent.

Phase 4 — Substantive feedback:

You: [shares draft]

LLM: Want Tier 1 feedback (clarity, internal consistency, what's
     missing)?

You: yes

LLM (Tier 1):
  - Clarity: the central claim restates as "second-monitor
    productivity gains are real but marketed at roughly 5x reality."
    That sentence isn't in the draft - consider adding it explicitly
    in the lede.
  - Internal consistency: paragraph 2 says "many devs get zero
    benefit"; paragraph 4 says "everyone gets some benefit from a
    second screen for code review." Reconcile.
  - What's missing: no acknowledgement of why the marketing-vs-reality
    gap exists - is it pure marketing, or do devs self-deceive after
    purchase?

You decide which observations to act on, then rework the prose yourself.

Phase 5 — Revision: you revise. LLM silent.

Phase 6 — Mechanical pass:

LLM: Mechanical review:
  - 3 typos: para 2 "definately" -> "definitely"; para 5 "thier" ->
    "their"; para 6 missing closing quote.
  - 1 awkward phrasing: "the productivity that is gained when one is
    using two screens" -> "the productivity gained from two screens."
  - All links resolve.
  - Title revisit (low-stakes glance): working title is
    "second-monitor-take." Quick alternatives:
    "What Two Monitors Actually Give You" or
    "The Dual-Monitor Productivity Gap." Pick one or keep working
    title - don't agonize.

You: approve all typo and phrasing fixes; keep working title.

LLM: [implements the approved fixes directly in the post]

Phase 7 — Publish: flip draft = false, run zola check.

Variant: entry at Phase 3 (clear-idea-then-feedback)

You have a clear take and want to draft solo before involving the LLM. Open the file, write the post, then enter the workflow at Phase 4 by sharing the draft and saying "ready for substantive feedback." Continues through Phase 5, 6, 7 normally.

Variant: entry at Phase 6 (mechanical-only)

You have a finished draft and just want polish before publish. Share the draft with "just need mechanical review." LLM produces the mechanical-issue list as in Phase 6 above; you approve/reject; LLM implements; you publish.

Optional starter template

+++
title = "..."  # working title - revisit before publish, don't agonize
slug = "..."
date = ...
draft = true
[taxonomies]
tags = []
+++

<!-- Pre-writing modes (optional; ask the LLM for one or several,
     or skip entirely):
  Provocation        - LLM proposes an extreme version of your own
                       argument; you push back to find your actual
                       position. Recommended opener.
  Multiple framings  - LLM offers candidate angles; you pick closest.
  Mirror-and-probe   - LLM restates and probes; you iterate.
-->

[your post starts here]

The HTML comments are visible while drafting and invisible if the post is rendered without them being removed — the lowest-bar scaffolding the form has.