· ENPT-BR
Grok Build for Alembic · Evidence & selection

The overlap matrix: adopt, already, defer, reject

Turn repository admiration into an ownership decision that Alembic can maintain.

Primary source
Pinned Grok Build README

Pinned upstream commit: b189869b7755d2b482969acf6c92da3ecfeffd36

The Simple layer stands alone; open Technical for exact contracts and edge cases.
You will be able to
  • Apply all four verdicts to a concrete surface.
  • Separate useful learning from code adoption.
  • Name the blocker that turns ADOPT into DEFER.
1

The big idea

A strong upstream idea is not automatically a feature Alembic should copy. Every surface must answer two questions: does Alembic already own this job, and can we prove the new value at a real boundary? The four verdicts keep attractive duplication out while preserving a roadmap for genuine gaps.

Think of it like… Use the matrix like a transplant board: match the organ to a real need, reject incompatible duplicates, and defer surgery until prerequisites are ready. It breaks because software features can be re-evaluated cheaply; medical decisions cannot.

ADOPT requires a missing owner, narrow interface, bounded implementation, and a real Proof Gate. ALREADY routes learning into existing Alembic owners. DEFER records an attractive unit with an explicit blocker. REJECT protects architecture from a second control plane.

The key anti-pattern is feature-count thinking. Grok sessions, rewind, worktrees, hooks, plugins, TUI, and sandboxing are impressive, but importing them would duplicate mature Alembic layers. Their design can inform tests and future seams without becoming copied code.

Evidence: Pinned Grok Build README

fitoverlapproofblockerowner Safe: capture the decision, blocker, proof, and next owner.
Turn repository admiration into an ownership decision that Alembic can maintain.
2

Compare the boundaries

Safe
Safe: capture the decision, blocker, proof, and next owner.
Risky
Risky: equate stars or code volume with architectural fit.
SurfaceUpstreamAlembicVerdict
Streaming JSONLDocumented subprocess contractProvider seam waitingADOPT
Sessions / rewindCoordinated directory + snapshotsSingle-file storage seam todayDEFER
TUI / tools / sandboxComplete upstream implementationsExisting Alembic ownersREJECT

The key anti-pattern is feature-count thinking. Grok sessions, rewind, worktrees, hooks, plugins, TUI, and sandboxing are impressive, but importing them would duplicate mature Alembic layers. Their design can inform tests and future seams without becoming copied code.

3

Real code and commands

docs/grok-build-internalization/research.md (decision method)
type Verdict = "ADOPT" | "ALREADY" | "DEFER" | "REJECT";

const headlessProvider = {
  fit: "high",
  overlap: "low",
  proof: "real grok CLI",
  verdict: "ADOPT" satisfies Verdict,
};
How to reach it
rg -n "ADOPT|ALREADY|DEFER|REJECT" docs/grok-build-internalization docs/courses/grok-build-for-alembic

Evidence: RESOURCES.md

4

Visual atlas

inputprooffit
Trace the boundary: fit.
surfaceseamowneroverlap
Locate the owner: overlap.
claimwithout gateevidenceat boundaryvsproof
Compare claim and evidence: proof.
learnexecuteverifyblocker
Read the sequence: blocker.
complete?yes / noowner
Find the decision gate: owner.
inputproofADOPT
Trace the boundary: ADOPT.
surfaceseamownerALREADY
Locate the owner: ALREADY.
claimwithout gateevidenceat boundaryvsDEFER
Compare claim and evidence: DEFER.
learnexecuteverifyREJECT
Read the sequence: REJECT.
complete?yes / noTUI
Find the decision gate: TUI.
inputproofsessions
Trace the boundary: sessions.
surfaceseamownerJSONL
Locate the owner: JSONL.
claimwithout gateevidenceat boundaryvsACP
Compare claim and evidence: ACP.
learnexecuteverifyruntime
Read the sequence: runtime.
5

Practice and feedback

Predict before revealing

A feature is valuable but its storage prerequisite is missing. Which verdict is honest?

DEFER, with the missing prerequisite named. Value does not erase a blocker.

Retrieval cards

Retrieval cardsADOPTClick or press Enter to flip
ADOPTMissing owner + narrow proofable seam.
Retrieval cardsALREADYClick or press Enter to flip
ALREADYEquivalent capability has an Alembic owner.
Retrieval cardsDEFERClick or press Enter to flip
DEFERUseful, but a named prerequisite blocks honest delivery.
Retrieval cardsREJECTClick or press Enter to flip
REJECTDuplication or wrong ownership outweighs value.
A strong upstream idea is not automatically a feature Alembic should copy. Every surface must answer two questions: does Alembic already own this job, and can we prove the new value at a real boundary? The four verdicts keep attractive duplication out while preserving a roadmap for genuine gaps.
ADOPT requires a missing owner, narrow interface, bounded implementation, and a real Proof Gate. ALREADY routes learning into existing Alembic owners. DEFER records an attractive unit with an explicit blocker. REJECT protects architecture from a second control plane.

Filter the decision board

Select a verdict to see its operational rule.

Verdict

Four-slide summary

ADOPT

Missing owner + narrow proofable seam.

ALREADY

Equivalent capability has an Alembic owner.

DEFER

Useful, but a named prerequisite blocks honest delivery.

REJECT

Duplication or wrong ownership outweighs value.

Check your model

Spaced review

Where do Grok's multi-file sessions land in phase one?

The value is real, but Factory's current capture seam is not an atomic directory transfer.
6

Sources and next move

Pinned Grok Build READMEb189869b7755 · readmePinned headless mode guideb189869b7755 · headlessPinned session management guideb189869b7755 · sessionsRESOURCES.mdGrok Build for Alembic
Ask your teacher to challenge a verdict, replay an event, or review a future unit against the same proof discipline.