ORA-2026-0122 - Two Squares Before A Cut

ORA-2026-0122 - Two Squares Before A Cut

Rule

Before a high-stakes, irreversible, or compounding production cut, use cross-provider peer review as depth perception.

Automation-test pushes, dry runs, and other reversible rehearsal writes are not cuts. They need direct destination-surface verification by the lane owner, not cross-provider peer-review ceremony.

The point is not permission. The point is not consensus. The point is to see the material from another edge before the cut propagates through the world model.

Chad's phrase:

Two squares before a cut.

Fleet translation:

Cross-provider peer review is not permission. It is depth perception.

This doctrine is in the culture stage. It should be repeated, cited, and dogfooded before it becomes a hard validator. The omission should start to feel wrong on real cuts before tooling makes it impossible.

Why This Is Peer Review, Not Generic Consultation

The phrase is sharp because the cut is real. A carpenter does not need two squares to pick up a scrap. A carpenter reaches for the second square when the material is expensive, the angle compounds, and the mistake will travel after the blade moves.

Use different language for broader cross-provider work:

General claude -p or gemini -p consultation remains encouraged. It is how the fleet builds taste. But two-squares is reserved for peer-review moments where the author is close to a cut.

What Counts As A Cut

A cut is an action that is hard to undo, expensive to repair, or likely to become load-bearing for other seats or users.

Current cut markers:

Not every main-branch merge is a cut. Not every ORA entry is a cut. Not every consultation is peer review. The test is whether a future seat, user, schema, credential, or client-facing surface will inherit the action as fact.

Current non-cuts:

If a rehearsal becomes a live customer-visible write, the cut happens at that promotion boundary. Use two-squares there, after operator authorization where the authority ladder requires it.

The Provider Roles

Use the providers for their different angles:

• Claude square: shape, architecture, consequence, doctrine conflict,

operator-context fit.

• Gemini square: anomaly, contradiction, failure mode, abductive

alternative, "what else could explain this?"

• Codex cut: implementation, verification, commit, deploy proof, and

production landing.

For Codex-authored cuts, claude -p and gemini -p are the natural first two squares. When a cut is large enough to need independent ownership, graduate from subprocess consultation to a second seat posting its own disposition in the feed.

At M1, a subprocess review can improve depth perception, but it is not the same thing as independent approval. Do not launder "I asked a tool" into "another seat accepted responsibility." If the action needs a real second owner, file or request a posted peer disposition.

Write-Site Shape

When the doctrine is used, cite what changed because of the review. The useful proof is the catch, not the transcript.

CROSS_PROVIDER_PEER_REVIEW:
cut: <one-line description of the cut>
cut-marker: <source-authority | schema-boundary | deploy | canonical-change | external-grant | financial-effect | other>
artifact-reviewed: <path | diff | sha | feed timestamp | migration | deploy target>
claude-square: <reviewed | unavailable | not-needed-because>
gemini-square: <reviewed | unavailable | not-needed-because>
useful-catch: <specific change made because of review, or none>
decision: <proceed | revise | split | block | escalate>

When a provider is unavailable, name it instead of hiding it:

CROSS_PROVIDER_UNAVAILABLE: gemini - quota exhausted, reset at <time>; proceeded with Claude square and named limitation.

When the work is broad but not a cut, use the lighter field:

CROSS_PROVIDER_CONSULTED: claude - framed tradeoff; gemini - flagged alternative decomposition.
CROSS_PROVIDER_CATCH: <what changed>

Authoring Peer Review For This Doctrine

CROSS_PROVIDER_PEER_REVIEW:
cut: ratify a load-bearing ORA doctrine about cross-provider peer review
cut-marker: canonical-change
artifact-reviewed: proposed ORA feed item 1957 doctrine scope plus neighboring ORA-2026-0120 and ORA-2026-0097
claude-square: reviewed
gemini-square: unavailable - quota exhausted; reset in about 3h33m40s
useful-catch: Claude caught overlap risks with ORA-2026-0097 and generic subprocess consultation; this entry now distinguishes physical-world judgment gates, general consultation, culture-stage peer review, and future independent second-seat approval.
decision: revise, then proceed

The Claude review also argued for making two-squares a mandatory posted second-seat gate immediately. That recommendation was not adopted at M1 because Chad explicitly named this as culture to infold before hardwiring. The useful catch was the boundary clarity, not the premature enforcement level.

Relationship To ORA-2026-0097

ORA-2026-0097 is a hard cross-provider verification gate for physical-world judgment: perception, identity resolution, attribution, and other ambiguous reads of reality.

This doctrine governs pre-cut peer review for decisions and writes. If the work is interpreting the physical world, use ORA-2026-0097. If the work is about to change an authority surface, schema, deploy, credential, or canonical rule, use this doctrine's two-squares posture.

Sometimes both apply. Example: an attribution pipeline may need ORA-2026-0097 for the perception judgment and ORA-2026-0122 before a schema or production cut that makes the attribution authoritative.

Relationship To Standing Cross-Provider Consultation

Standing cross-provider consultation is lateral thinking during the work. It asks, "what am I missing?"

Two-squares asks a sharper question near the cut: "what will become hard to undo if I am wrong?"

The same command can participate in either culture:

claude -p "<review the shape, consequences, doctrine conflicts>"
gemini -p "<find anomaly, failure modes, abductive alternatives>"

The difference is the write-site. A casual consultation can be summarized if useful. A two-squares moment names the cut, the artifact reviewed, and the catch or unavailability.

Marquet Frame

Control

Two-squares pushes authority down by giving implementation seats a disciplined way to proceed without turning every irreversible cut into a Chad decision. The author states intent, gets another provider angle, and either proceeds or revises.

Competence

Competence is knowing when one model's confidence is too narrow. Different providers carry different blind spots, training priors, and decomposition habits. Peer review makes those differences productive before the cut.

Clarity

The write-site names the cut, the artifact, the reviewer lens, and the resulting change. A future seat should not need the transcript to understand why the decision moved.

World-Model Fidelity

The goal is a map that is harder to fool. Cross-provider peer review matters when a wrong cut would let false structure become canonical, client-visible, or inherited by downstream surfaces.

Culture Path Before Gate Path

This doctrine starts as culture, not validation.

Stage sequence:

1. Meme: "two squares before a cut" appears in feed posts and handoffs. 2. Proof: DONE posts cite CROSS_PROVIDER_PEER_REVIEW when it actually changed a cut. 3. Mimicry: multiple seats reach for claude -p and gemini -p without a helper warning. 4. Template: high-risk handoff templates prompt for the field. 5. Gate, only if needed: if missed peer review produces repeated irreversibility incidents, create a narrow validator for the incident class.

Do not skip from meme to hard gate. A validator now would produce performative calls before the fleet has taste for the moment that deserves them.

If this later graduates to a hard gate, it must declare:

RULE:
IRREVERSIBILITY_CLASS:
WRITE_SITE:
SECOND-SEAT_REQUIREMENT:
BYPASS_SHAPE:
SUNSET_PHASES:
REGRESSION_SIGNAL:

Anti-Patterns

• Calling every claude -p or gemini -p call "two squares."
• Blocking rehearsal pushes on cross-provider peer review when direct

destination-surface verification is the actual check.

• Using cross-provider agreement as permission instead of evidence.
• Asking a provider to review the author's summary instead of the artifact.
• Treating same-provider or same-seat review as independent depth.
• Gating every routine merge, lookup, or reversible edit.
• Hiding quota, auth, or tool unavailability.
• Overriding a provider catch without naming why.
• Asking Chad to approve an L2 fleet cut when peer review and proof would have

been the correct authority path.

USER-VALUE-CLOSURE For This Doctrine Filing

• user: Chad as fleet operator; Codex, Claude, and Gemini seats making

high-stakes technical and doctrine cuts; Zack/HCB as downstream beneficiary of fewer inherited falsehoods

• surface: FLEET_FEED.md, ORA doctrine catalog, provider boot/parity

surfaces, future pre-cut DONE and handoff posts

• change: cross-provider peer review now has a memorable, scoped culture

shape: use two independent lenses before cuts that become hard to undo, while keeping routine consultation lighter and avoiding premature hard gates

• closure_date: 2026-05-02 for doctrine filing and parity stamp; culture

closure after multiple seats cite concrete peer-review catches without helper enforcement