File 01 · the spine

The loop.

Good work comes from a loop, not a line.

Diverge wide. Converge hard against reality. Translate only when a gate needs someone's yes. Let what comes back decide which layer the next loop runs at.

FIG 02 // THE FULL LOOPSCROLL // RUN IT

01 · Goal stack

Start from the real objective. Keep asking "in service of what?" until the next answer wouldn't change anything you'd do.

02 · Diverge

Go wide. Open the parts of the space the default answer never searches.

03 · Converge

Then hard against reality. Kill what breaks. This is where the checks and the receipts live.

04 · Until it's sound

Back and forth until diverging runs dry and converging stops breaking. What's left is sound.

05 · The gate

No built gate between the result and the outcome? Go straight through. A gate in the way, someone whose yes you need? Only then translate. Nothing unchecked crosses.

06 · Translate

See it from their seat: their goals, their hard rules, where they have room to say yes. Lead with what helps them get their job done.

07 · Feedback climbs

Came back wrong? Ask which layer failed. Tactics: done badly. Strategy: wrong plan. Vision: wrong goal. Fix at the lowest layer that explains it. When a good fix keeps failing, climb.

Why it holds

Each step guards the one before it.

Diverge, no converge

Fantasy.

Every option survives, because nothing tests them.

Converge, no diverge

Consensus.

You check hard, but only the answer everyone already had.

Right, no translate

Nobody says yes.

The answer holds, and it stops at the gate.

How deep

Set before you start.

By consequences. Try it:

SHIP AND LEARN. Cheap to undo and fast to learn from. Don't overthink it.

When it's done

Read while it runs.

  • Diverge runs dry

    New passes return what's already been seen.

  • Converge stops breaking

    The strongest attack fails, and new evidence stops moving it.

  • The next round isn't worth it

    At these stakes, another pass costs more than it could change.

The operator makes the call.

The system's job is to show what's dry, what's holding and what's untested. The two ways to get it wrong: stopping on relief instead of evidence, and never stopping because openness feels safer.

FILE 01 · THE LOOP
ANNEX A · LIMITSWHERE IT BREAKS
  1. A failure doesn't name its own layer. Noise, a weak test or a sloppy fix can look like a strategy failure, and the loop climbs when it shouldn't. Climb only when a good-enough fix fails more than once.
  2. Same family, same blind spot. When one kind of model both generates and judges, diverge and converge agree because they're alike, not because the answer holds.
  3. The posted rules aren't the real ones. Translate frames to the stated checkboxes while the decision turns on something unwritten.
  4. No judge outside you. In work judged by taste, converge has nothing outside the operator to check against in the time you have.

Where the files sit

Seven files, one loop.

  • P1 · GOAL STACKENTRY
  • P3 · DIVERGENCEDIVERGE
  • P5 · ALREADY SOLVEDDIVERGE
  • P4 · SEATSCONVERGE
  • P6 · RECEIPTSCONVERGE
  • P7 · TRANSLATETRANSLATE
  • P2 · THE WALL SORTEVERYWHERE
OTHR · THE LOOPREV // 36231ef4 · BUILT // 2026.10.11NO CAPTURE · NO CTA