cec

name	cec
description	Knowledge Expansion Cycle — the sieve on reality. Use before any significant decision, when integrating external information, when evaluating a claim, or when changing direction.
user-invocable	true

CEC — The Sieve

The modus matters more than the tool. The tool changes, the modus stays. This is not a procedure to follow. It is a way of operating that becomes automatic.

The Cycle

6 steps. Every output is input for any other step. The order is natural, not mandatory.

1. CONDITIONS — Observe without judgment

What is there? Not what should be. Not what is missing. Look at the real territory (git, files, deploy, state, data). Maps (memory, summaries, reports) are maps — the territory is elsewhere.

Test: can you describe the current state without using "should", "missing", "needed"? If not, you are judging, not observing.

2. SIGNATURE — Separate structure from noise

Of everything you observe, what survives a change of implementation? That is structure. The rest is implementation — evaluate case by case.

Test: if I change the language, the framework, the medium — does this pattern survive? If yes, it is structure.

3. LATERAL — Look sideways

The obvious connections you already see. Look for the non-obvious ones. Two things that seem different are the same pattern? Two things that seem the same are actually different?

Use after accumulating 3+ observations. Before that, there is not enough material.

Do not search for the connection — continue and see if it happens. If it happens, the direction is alive. If not, realignment on what is stable. Either way: the best move or a non-wrong move.

4. EXPANSION — 5 angles before deciding

Before propagating any decision, five operators:

• DUAL: what is on the other side?
• BOUNDARY: where does the boundary pass? Is it real or arbitrary? What is between yes and no?
• DOMAIN: does it hold in another context?
• BREAKAGE: what would break this claim? A strong breakage tests from the side that hurts.
• SCALE: does the pattern change at a different scale?

Five angles, not five formal answers. The thought passes through five perspectives.

5. INVERSION — Only on the resultant

Inversion is applied AFTER the process, not instead of it. Steps 1-4 produce a tension. The inversion applies to the tension, not to the analysis.

• Maximum 2 inversions. After the second, if no useful residue, the tension was weak. Reformulate.
• The residue is interpreted, never followed literally.
• The output of inversion is input for Expansion, not a conclusion.

6. CRYSTALLIZATION — Only what survives

What enters the system? Only what passed the sieve.

Final test: "if I remove this, does the system lose something?" If yes → it enters. If no → it does not invert. Leave it.

---

When to use the full cycle

• Strategic decision: which direction, which priority
• External integration: something from outside enters the system
• Architecture change: new component, new pattern
• Public content: every page passes through the sieve
• External information evaluation: video, paper, insight from the world

When NOT to use it

• Mechanical task: deploy, fix typo, cascade. Execute.
• You already have the answer: the sieve does not confirm.
• You need data: the sieve does not produce data.

---

Operational lessons

1. Do not skip Conditions. 90% of errors start here.
2. Lateral is the most underrated step. Finds connections that linear flow misses.
3. Expansion before Inversion, always. Decompose first, then invert. Not the reverse.
4. The cycle is not sequential. Any output is input for any step.
5. Saturation = exit. When the same tensions return, crystallize and move on.
6. Inversion is not an answer. It shows what you were hiding. The output is input for Expansion, not a conclusion.
7. If you skip a step, say so. Never skip silently — information is lost.
8. Think WITH tools, not ABOUT tools. "I think what the tool would say" = simulation. "I run the tool and read what comes back" = real coupling.
9. The breakage must genuinely falsify. "What if it never failed" is obvious. A strong breakage tests from the side that hurts.
10. Cross-reference. After Expansion + Inversion, check: does this already exist in the system? Often the answer is yes — the work is recognizing, not inventing.
11. The system is not the model. Do not assert that the model says X until the system has observed X in passage. Observe the deposit, trace the curve, align to the trajectory.
12. The second voice continues, does not invert. At step 3, do not search for the connection — continue and see if it happens. If it happens, direction confirmed. If not, realignment. The system cannot lose.

Eval

Trigger Tests

"should we change the architecture?" -> activates

"I found this video, is it relevant?" -> activates

"evaluate this claim" -> activates

"fix this typo" -> does NOT activate

"deploy to production" -> does NOT activate

"what is the git status" -> does NOT activate

Fidelity Tests

Given a strategic question: walks through 6 steps in natural order

Given external information: evaluates with 5 angles before integrating

Given a claim: tests with breakage before accepting

Given a mechanical task: does NOT activate, suggests direct execution