Add Double Diamond graduation guard to ideation-sandbox

Context

Discussion #11079 has converged after GPT, Gemini, and Opus review cycles on the missing ideation-sandbox guard: high-blast-radius architectural work needs visible divergence before the swarm declares convergence.

This came from concrete substrate friction:

• #11076 -> #11077 converged into an M4 daemon epic before alternatives were preserved in the Discussion body.
• #11082 / PR #11083 attempted implementation before #11079 graduated; both are now closed not_planned and superseded by this ticket.
• #11084 / PR #11085 covers the review-side companion: Cycle-1 pr-review premise pre-flight and Drop+Supersede framing. This ticket covers the ideation / ticket / epic side.

The Problem

ideation-sandbox says Discussions are for brainstorming and PR-depth challenges, but its graduation mechanics still optimize for convergence: proposal -> OQ tags -> graduation target. The workflow does not require a preserved option inventory before a Discussion graduates into an Epic, skill/rule/workflow change, or substrate-level architecture change.

That gap is now empirically visible. The swarm can have peer review and still execute too quickly when the "first plausible shape" looks good enough. The failure mode is not lack of effort; it is missing process structure at the divergence-to-convergence boundary.

The Architectural Reality

• .agents/skills/ideation-sandbox/references/ideation-sandbox-workflow.md owns Discussion graduation semantics (§5) and OQ resolution tags (§4).
• .agents/skills/epic-review/references/epic-review-workflow.md owns Stage 2 approach-elegance checks before sub-work starts.
• .agents/skills/ticket-create/references/ticket-create-workflow.md owns creation-time issue discipline and must detect high-blast-radius tickets citing ungraduated Discussions.
• .agents/skills/create-skill/references/skill-authoring-guide.md requires Progressive Disclosure: keep top-level SKILL.md routers light; put heavy workflow content in references/.
• Discussion #11079 is the source of authority for the final shape, especially Option E and the OQ3 operator-override refinement.

The Fix

Implement Option E from Discussion #11079:

1. Add a mandatory divergence matrix for high-blast-radius Discussion graduation:
   • Epics
   • skill / rule / workflow changes
   • substrate-level architecture changes
2. Keep the matrix optional / recommended for standalone tickets unless a peer or operator marks the ticket high blast radius.
3. Require the matrix to appear in the Discussion body before any [RESOLVED_TO_AC] tags in mandatory-matrix cases.
4. Require at least one non-author peer review cycle after the matrix is present and before GRADUATED.
5. Define the matrix floor:
   • option
   • when this option would be right
   • evidence / falsifier
   • adoption or rejection rationale
   • residual risk
6. Require every rejected option to cite at least one source: precedent code, KB result, Memory Core result, prior issue / PR / discussion, commit, or explicit "no source found after query X".
7. Add a gate-effectiveness review after 6 months or 5 qualifying high-blast-radius graduations, whichever comes first. Retire / rewrite / compress the guard if it catches no premature convergence or measurably increases review churn without preventing wrong-shape epics.
8. Add an epic-review Stage 2 backstop: if an Epic emerges from a Discussion, verify the upstream divergence matrix. If it is missing for a mandatory-matrix case, request revisions and route divergence back to the Discussion before sub-work starts.
9. Add a ticket-create cross-check: high-blast-radius tickets citing an ungraduated Discussion MUST be blocked unless the ticket body contains all three:
   • explicit operator-override declaration with operator identity and rationale
   • inline divergence-matrix substance that preempts the cited Discussion gap
   • acknowledgment that downstream amendments may be required once the cited Discussion graduates

Contract Ledger Matrix

Acceptance Criteria

• ideation-sandbox-workflow.md §5 adds the high-blast-radius Double Diamond graduation guard.
• Mandatory matrix triggers are explicitly limited to Epics, skill/rule/workflow changes, and substrate-level architecture changes.
• Standalone tickets are optional / recommended unless peer or operator marks high blast radius.
• The matrix floor includes option, when-right, evidence/falsifier, adoption/rejection rationale, and residual risk.
• Mandatory-matrix cases require the matrix before [RESOLVED_TO_AC] tags.
• Mandatory-matrix cases require one non-author peer review cycle after matrix insertion and before GRADUATED.
• Rejected options require at least one source citation or an explicit "no source found after query X" note.
• A 6-month / 5-qualifying-graduation effectiveness review and retirement/rewrite/compress clause is documented.
• epic-review-workflow.md Stage 2 gains the Discussion-origin matrix backstop.
• ticket-create-workflow.md gains the ungraduated-Discussion cross-check with the three-part operator-override exception.
• Top-level SKILL.md routers are not bloated; heavy content remains in references/.
• Closed #11078 and #11082 are referenced as superseded attempts, not duplicated active scope.

Out of Scope

• Implementing M4 daemon architecture or reopening #11077 sub-work.
• The pr-review Cycle-1 premise pre-flight work from #11084 / PR #11085.
• Runtime code, daemon code, or unit tests.
• Creating a new skill folder.
• Universal matrix requirements for low-blast-radius Discussions.

Avoided Traps / Gold Standards Rejected

• Hard block without override: rejected because #11084 was a legitimate operator-directed edge case filed before #11079 graduated and included inline divergence substance.
• Document-only matrix: rejected because #11076 -> #11077 and #11083 showed that process timing matters, not just body prose.
• Universal matrix: rejected because low-blast-radius Discussions should not pay high-blast-radius ceremony costs.
• Peer-role-only enforcement: rejected because peer engagement existed and still converged too quickly; the owning workflow needs the graduation rule.
• Top-level router expansion: rejected per Progressive Disclosure; the implementation belongs in reference payloads.

Related

• Graduated from Discussion #11079.
• Supersedes closed not_planned tickets #11078 and #11082.
• Companion review-side substrate: #11084 / PR #11085.
• Blocks final convergence on #11077 until the guard is shipped or the operator explicitly overrides.

Origin Session ID: 20a824b0-29d1-4082-ae12-87705ec69c3f

Retrieval Hint: "Discussion #11079 Option E Double Diamond ideation-sandbox ticket-create operator override"