Explore Docs

Evaluation / 11

The requirements behind the concepts.

Public summaries and stable acceptance IDs connect the documentation to Merlin’s SpecKits. Internal implementation details remain private.

On this page
  1. How to read these references
  2. SpecKit 001 — Platform and lifecycle
  3. SpecKit 002 — Task outcomes
  4. SpecKit 003 — Staged workspace changes
  5. SpecKit 004 — Final authorization
  6. SpecKit 005 — Evidence and grounding
  7. SpecKit 006 — Analysis v2
  8. SpecKit 007 — Context and retrieval
  9. SpecKit 008 — Plans and program execution
  10. SpecKit 009 — Planning, routing and budgets
  11. SpecKit 010 — Sessions and workspace scope
  12. SpecKit 014 — Document indexing
  13. SpecKit 018 — Specification authoring

How to read these references

SpecKits are Merlin’s governing requirements. The IDs below connect public explanations to the acceptance criteria that define the relevant behavior. They are curated summaries, not exports of the internal specifications.

A requirement is not an independent certification or a guarantee that a deployment has no defects. Evaluate the actual configured system and its evidence. Research proposals, retired implementation details, and historical measurements are not presented here as shipping capabilities. This edition was reviewed on 28 September 2026.

SpecKit 001 — Platform and lifecycle

Structured operations, bounded progress controls, and explicit pause/resume behavior. These references identify public behavior; they do not publish runtime internals.

Acceptance references: AC-FR001-05, AC-FR001-09, AC-FR001-10, AC-FR003-01, AC-FR003-04.

SpecKit 002 — Task outcomes

Task completion and blocking outcomes remain explicit. A completed task record and an achieved semantic outcome are related but distinct responsibilities.

A completed dependency releases the tasks waiting on it, and its residue travels with them as a stated fact.

Acceptance references: AC-002-004-01, AC-002-007-01, AC-002-009-08.

SpecKit 003 — Staged workspace changes

Changes enter a staged lifecycle with required content verification and defined rollback states. This contract does not promise reversibility for every external side effect.

Acceptance references: AC-003-002-01, AC-003-003-01, AC-003-005-01.

SpecKit 004 — Final authorization

Final claims are reconciled with actual work, verification and blockers. A verification receipt proves its named assertion, not arbitrary correctness.

Acceptance references: AC-004-004-01, AC-004-004-02, AC-004-004-03, AC-004-004-04, AC-004-006-17, AC-004-006-22.

SpecKit 005 — Evidence and grounding

Evidence attribution and concrete grounding checks connect claims to inspected material. Source fidelity and interpretation are separate questions.

Acceptance references: AC-005-001-01, AC-005-001-03, AC-005-004-01, AC-005-004-02, AC-005-004-04, AC-005-005-01.

SpecKit 006 — Analysis v2

Structured investigation, challenge, verification and consolidation. Rebuttal and verification can preserve uncertainty instead of manufacturing confirmation.

Acceptance references: AC-006-001-01, AC-006-005-01, AC-006-006-01, AC-006-008-01.

SpecKit 007 — Context and retrieval

Context assembly and source-aware discovery. A retrieval hit is not proof that a whole document was read.

Acceptance references: AC-007-003-07, AC-007-005-01, AC-007-008-01, AC-007-008-02.

SpecKit 008 — Plans and program execution

A plan is a list of tasks, each with a short key, its text and a step kind. Routing hints name each task’s dependencies by key, and an unknown or duplicate key is refused by name.

Acceptance references: AC-008-002-14.

SpecKit 009 — Planning, routing and budgets

Runtime accounting governs the work without claiming a model is deterministic. Stage budgets gate how many iterations planning, analysis and execution may run, a session ledger records token use, and steps whose dependencies did not complete are skipped.

Acceptance references: AC-009-005-01, AC-009-008-01, AC-009-008-02.

SpecKit 010 — Sessions and workspace scope

Session identity, lifecycle, and configured evidence scope define continuity and reuse boundaries. Workspace reuse must account for source freshness.

Acceptance references: AC-010-016-35, AC-010-016-36.

SpecKit 014 — Document indexing

Configured documents participate in discovery alongside code. Indexing and extraction do not establish that every source has been reviewed.

Acceptance references: AC-014-002-01.

SpecKit 018 — Specification authoring

Repository-aware artifact generation, authorization boundaries and cross-artifact checks. Documenting a feature is not implementing it.

Acceptance references: AC-018-001-02, AC-018-002-01, AC-018-002-02, AC-018-002-04, AC-018-003-01, AC-018-003-02, AC-018-003-03, AC-018-004-01, AC-018-004-04, AC-018-005-01, AC-018-005-03.

Discuss a workflow or deployment. Contact RLM-Labs.ai →