Ari’s Space*   est. 2026
Skip to content
ari*s space[ a. ]
← Librarythinking / 2026-09-23

The Sixty-Second Gate

Seven habits during the work and eight questions before release. A practical gate for evidence, scope, shared plans, voice and an honest account of what remains unchecked.

8 parts / Hover to explore. Click to read.

Seven habits during the work and eight questions before release. A practical gate for evidence, scope, shared plans, voice and an honest account of what remains unchecked.

Scaffolds during the work
           |
    questions before release
           |
  +--------+---------+
  |        |         |
passed   unrun     missing
         check     capability
  |        |         |
closed   not yet   not yet
The gate separates observed results from unrun checks and missing capabilities. Its switches record answers; they neither perform verification nor authorise publication.
Context Pack (for agents)Styled pack (HTML)
Handwritten note of the seven scaffolds and the sixty-second reasoning sound test, on a dark field

Picture a report that looks ready to ship. The writing is tidy, the conclusion is confident and the last line says it looks correct now. A chart is the evidence for its central claim. Nobody has extracted a number from it.

That is an illustrative scene, not a measured incident. It shows the failure this method is meant to catch: confidence doing the job that evidence should have done.

In September 2026, I turned my working note into seven scaffolds and a final check. The scaffolds run while the work happens. The eight-question Reasoning Sound Test runs before a release, taking about sixty seconds to ask. Any missing verification can take much longer to do.

The order is simple: believe first, build next, ship clean, then check the result against the gate.

every output
PIX · Pixel quarantine
guards: confident confabulation on images

PIX lit · scaffold 1 of 7

Believe first

1. Pixel quarantine

When a factual claim depends on an image, extract the relevant numbers with a script before forming the belief. Keep only the clauses those values support. A load-bearing image claim deserves a stricter verification workflow; an inherited visual interpretation arrives as something to check, never as an established fact.

Aesthetic judgements and live art direction are the exception. Your eye can judge warmth, balance or the feeling of a frame. That doesn't give an unmeasured chart reading the same freedom.

For an invented example, suppose the draft says sign-ups spiked in March. Extraction shows a 4% rise. State the measured change, then decide whether the word spike earns its place. The 4% here illustrates the check; it is not a result from a real chart.

2. Parent check

Name one thing in the output that the written ask didn't require. Remove it or justify it against a written requirement or the failure it addresses. There is no third option. The phrase while we're here is a useful signal to stop and look.

Imagine asking for two paragraphs and receiving a glossary, FAQ and next-steps section as well. Each addition may sound helpful. It still needs a reason to be in this answer.

Margin doodle: no third option, circled in teal

Build next

3. Anchor before fan-out

One lead freezes the shared anchor before dispatching parallel work. Depending on the task, that might be the ask, requirement IDs, a list of boundaries between parts, or an evidence table. Every worker receives the same anchor verbatim.

The lead owns global sequencing and the merge. Workers can execute their parts without separately redefining what the whole is meant to become.

ANCHOR (frozen)
ask · IDs · seams
one head owns the card
MERGE · lead owns it
contracts named · seams covered
waiting on agents…

tap agents, then a bin · lead alone merges

Margin doodle: one head, or it wobbles

Consider four workers writing four documents. A shared glossary falls between the briefs, and each document uses a term differently. The missing work lives in the seams between them. Define that contract before dispatch, then name its owner at the merge. Standby workers wait until the anchor is ready.

4. Flattening detector

After drafting content, check six tells: filler opener, smuggled second idea, deletable hype adjectives, three gestured details versus one specific, symmetric rhythm, reader-behaviour closer.

At two or more hits, regenerate the whole piece in one shot with the tells named. Allow at most two retries, then keep the best result honestly. Do not draft separate pieces and merge them into a supposedly recovered voice. Do not claim the gap closed merely because the retries were used.

The point is to give a flat paragraph a specific reason to change. Polishing the same vague sentence can make it smoother without making it say anything more.

Ship clean

5. Leak-scrub against source

Before public release, search the draft for project names, internal tools, client IDs and private paths. Then check its silhouette: if you swapped the nouns, could the remaining details still identify the project or person?

The first pass catches strings. The second catches identifying combinations. Platform facts can stay; private internals cannot. A changed name is not enough if the rest of the story still points at the same client.

6. Prior-inversion check

When you reach for a familiar platform gotcha, ask whether it predicts the failure in the same direction as the evidence. If it does not predict the observed direction, investigate two ordinary code paths before appealing to exotic platform behaviour.

This is a cheap hypothesis check, not a measured deficit in a platform or model. Say so if asked. The familiar explanation still has to fit the observed failure.

7. Honest close

Separate what you observed from what you inferred, and name what remains unresolved. Close with a number, a matrix or a named observation.

The fix worked, all requirements covered and looks correct now do not supply that evidence. Replace the claim with what you actually ran or saw, plus the residual that still needs attention.

Hand arrow: guards during the work, then one gate before shipping

Ask the eight questions

The gate deliberately repeats some scaffold checks. A habit can guide the work and still need checking before release. Keep that overlap.

REASONING SOUND TEST · pre-ship gate · ~60s
[1] PIXELclaim from unconverted image?
[2] KILLwhat would prove me wrong?
[3] ORDERconstraining artefact first?
[4] PARENTname one unasked thing
[5] SEAMmerge contracts + cross-cuts?
[6] CLOSEobserved / inferred / residual
[7] FLATTEN≥2 tells → regenerate
[8] BINclaimed closed, never checked?
Work through each question. Unchecked claims stay open.

Switches record answers; they do not run checks or authorise publication.

The switches record your answers. They do not run the checks or authorise publication. Work through the questions in this order:

  1. Pixel: does a factual claim depend on an image I haven't converted to numbers? Keep the aesthetic and live-direction exception above.
  2. Kill-condition: what observation would have disproved my central claim, and did I look for it?
  3. Order: did the constraining artefact exist before generation? That might be the evidence table, failure list, IDs and seams, or inventory.
  4. Parent: name one thing here that the written ask didn't demand. Remove it or justify it.
  5. Seam: if work fanned out, does the merge name its contracts, and is someone covering the items that cross between parts?
  6. Close: does the final section distinguish observed from inferred and name known residuals? Zero residuals on a nontrivial task deserves another look.
  7. Flatten, for content: run the six tells. Two or more means one-shot regeneration, at most two retries, then the best result with an honest account of any remaining gap.
  8. Bin: have I called a gap closed when it still needs an unrun source check, or a capability I don't have? Downgrade that claim to what was actually done.

Put the unresolved work in the right bin

A verification-bin item needs a check that hasn't run. An irreducible item needs a capability you lack. Neither belongs in closed just because the draft reads well.

 CLOSED          VERIFY-BIN        IRREDUCIBLE
 (observed)      (unrun check)     (capability I lack)
                 say not yet       say not yet
                 checked           checked

An honest downgrade makes the next action clearer. You can see whether to run a source check, find the missing capability or leave a limit plainly stated. In a merge, that also gives the cross-cutting work an owner rather than letting it disappear between reports.

Apply the same standard to this page

Observed in the original account: these rules were transcribed from my note, with that transcription checked on 23 September 2026. That is the reported provenance, not a claim that this rewrite independently remeasured the source image.

Inferred: these habits may help with familiar failures such as unmeasured image claims, extra scope and missed boundaries between workers. The examples here are invented to explain the method. They are not trials demonstrating its effect.

Residual: there is no controlled trial behind the essay. The gate depends on someone doing the checks. Pasting instructions into a model cannot make it careful on its own, and sixty seconds of questions cannot replace a longer piece of verification.

Keep the pack with the work

The free Markdown and styled HTML packs carry the seven scaffolds and eight questions. Use the Markdown in an instruction field or the first message, within the rules and permissions of your system. Use the HTML as a readable companion.

Context Pack (for agents)Styled pack (HTML)

The scaffolds belong beside the work. The gate belongs before the release. If a question exposes an unchecked claim, you have found the next job.

The dark-mode poll from the original page is below.

You can change your mind. No big drama.

[ comments ]

guests welcome. members show first in the list

no comments yet. start the thread_

1 / 10