The Four Phases of Work — Design, Build, Fix, Refactor¶

Status: BINDING doctrine — made lawful 2026-07-12 by ADR-0.33.0-airlock-membrane / OBPI-0.33.0-06 (the campaign's Movement III section-8 "1.0 gate"), companion to its now-binding sibling work-phases-and-airlock.md. That doc holds the airlock (the in/out ritual + the two-graph); this doc holds the theory of the four phases the airlock is traversed in. Both originate in the operator design dialogue of 2026-06-16 and are ratified into the Build-to-1.0 campaign (Magna Carta) as item E.7. Per operator ruling 2026-06-17, Magna Carta could not be complete until these theories were realized, substantiated, and made lawful — this promotion, gated behind the section-5 live negative control biting un-forced in production, discharges that gate.

Why this exists: the operator's opening ask was for "a theory of the major phases: design, build, fix, refactor." This is that theory. The airlock doc absorbed the mechanism; this doc holds the modes — what each phase is, and the distinct proof each one owes.

The core claim¶

Each phase is defined by what it must prove, and that proof obligation is exactly the surface that makes vibing inert in that phase. A phase is not chosen by intention; it is determined by the evidence you can produce (see § "A phase is the proof you can produce").

The four phases partition work along two questions, plus one tiebreaker:

Does the canonical intent change? — the declared statement of what should be (ADR / REQ / invariant).
Does observable behavior change?
Tiebreaker (build vs fix): was the intent previously realized?

The four phases¶

phase	intent	behavior	must prove	the vibing it kills
Design	changes	none yet	coherence & traceability — the new intent decomposes down and traces up; no orphan intent	"everyone uses X" architecture; training-corpus-driven choices
Build	consumes (unchanged)	new	intent → behavior — every REQ covered by a passing test	plausible-looking code written without reading the surface
Fix	unchanged	corrected	behavior → intent — a red test proving the defect, now green; fix the class	fixing the instance, not the family
Refactor	unchanged	preserved	invariance — behavior unchanged + non-regressing complexity (semantics-pin; xenon/radon)	"cleanup" that silently changes behavior; taste-driven churn

Build vs Fix — the correction-vs-enhancement tiebreaker¶

Build and Fix both change behavior while leaving intent unchanged. They differ on one question: was the intent previously realized?

Build realizes intent that was never yet realized — a new capability. The gap is "not built yet."
Fix restores intent that was realized but broke or fell short — a correction. The gap is "built, but wrong / incomplete against its own declared intent."

This boundary is the operator's verbatim doctrine: "discovering that more is needed to fulfill the intent of a feature is not an enhancement, it is a correction." The four-phase theory derives that doctrine rather than asserting it.

The forbidden fifth cell¶

The phases force intent-change and behavior-change into separate, gated steps. The cell where both change at once, in one uncontrolled motion, is not a fifth phase — it is the named failure class: V.I.B.E.S. ("Velocity Increased, Bugs Expected Software"). Changing the spec and the code in the same breath, with no gate between, is exactly what governance exists to forbid: design must precede build. The four phases are the four legal quadrants of work; vibing is the illegal one.

A phase is the proof you can produce, not the intention you declare¶

You do not get to call a change a refactor — you are in Refactor only if you can show behavior-invariance. You do not get to call a change a Fix unless you can produce a red→green. The actor's claim is precisely the vibing surface; the available proof is what is real. This is why the phase is determined by evidence, not intention — and why each phase's proof-obligation (the table above) is its anti-vibing surface.

Cross-phase tripwires — the phases are in conversation¶

A unit of work can discover mid-traversal that it is in a different phase. The tripwires:

Drifting the intent mid-build → you are actually in Design. Stop; re-enter through design.
A fix that reveals the intent itself was underspecified → that is a Design act, not a fix.
Tempted to restructure while building or fixing → that is Refactor. Pull it out into its own gated traversal — never smuggle it in. (Most "the model wrecked three things it didn't see" is unaccounted refactor riding inside a build or a fix.)

Relationship to the airlock¶

The four phases are the modes of traversal through the airlock (see work-phases-and-airlock.md). Each phase's proof-obligation above is the airlock's depth-proof for that phase — what you must prove about the change itself. The airlock's other invariant, lateral closure (pack-in / pack-out; account for every seam perturbed), is universal across all four. And each phase is an operation on the two-graph: design writes intent-edges (LAW); build extends fact-edges toward law; fix repairs fact-edges back onto law; refactor reshapes fact-edges with law held invariant (airlock doc § 4).

Provenance¶

Captured verbatim-anchored from the operator design dialogue of 2026-06-16 (the theory) and the 2026-06-17 ruling (separate doc + 1.0-gating). Operator anchors preserved unchanged:

"a theory of the major phases: design, build, fix, refactor."
"discovering that more is needed to fulfill the intent of a feature is not an enhancement, it is a correction." (the build/fix boundary)
"have magna carta reference them, make it clear that we will never be complete with magna carta until those theories are realized/substantiated/make lawful."