diff --git a/.specify/feature.json b/.specify/feature.json
index f1ada89..b90bf9b 100644
--- a/.specify/feature.json
+++ b/.specify/feature.json
@@ -1,3 +1,3 @@
 {
-  "feature_directory": "specs/007-phase-gates"
+  "feature_directory": "specs/008-clarify-gate"
 }
diff --git a/specs/008-clarify-gate/checklists/requirements.md b/specs/008-clarify-gate/checklists/requirements.md
new file mode 100644
index 0000000..d14d026
--- /dev/null
+++ b/specs/008-clarify-gate/checklists/requirements.md
@@ -0,0 +1,34 @@
+# Specification Quality Checklist: Clarify-Gate Pattern
+
+**Purpose**: Validate specification completeness and quality before proceeding to planning
+**Created**: 2026-05-20
+**Feature**: [spec.md](../spec.md)
+
+## Content Quality
+
+- [x] No implementation details (languages, frameworks, APIs)
+- [x] Focused on user value and business needs
+- [x] Written for non-technical stakeholders
+- [x] All mandatory sections completed
+
+## Requirement Completeness
+
+- [x] No [NEEDS CLARIFICATION] markers remain
+- [x] Requirements are testable and unambiguous
+- [x] Success criteria are measurable
+- [x] Success criteria are technology-agnostic (no implementation details)
+- [x] All acceptance scenarios are defined
+- [x] Edge cases are identified
+- [x] Scope is clearly bounded
+- [x] Dependencies and assumptions identified
+
+## Feature Readiness
+
+- [x] All functional requirements have clear acceptance criteria
+- [x] User scenarios cover primary flows
+- [x] Feature meets measurable outcomes defined in Success Criteria
+- [x] No implementation details leak into specification
+
+## Notes
+
+- All items pass. Spec ready for `/speckit-plan`.
diff --git a/specs/008-clarify-gate/spec.md b/specs/008-clarify-gate/spec.md
new file mode 100644
index 0000000..f177896
--- /dev/null
+++ b/specs/008-clarify-gate/spec.md
@@ -0,0 +1,100 @@
+# Feature Specification: Clarify-Gate Pattern
+
+**Feature Branch**: `feature/8-clarify-gate`
+
+**Created**: 2026-05-20
+
+**Status**: Draft
+
+**Input**: Implement the clarify-gate pattern — structured CLR-NNN questions that persist across phase re-runs, with merge behaviour control.
+
+## User Scenarios & Testing
+
+### User Story 1 - AI agent raises clarification during phase execution (Priority: P1)
+
+An AI agent running `/adm.lineage` encounters an ambiguous column name. It raises a structured CLR-NNN question that gets stored in the domain's state. When the phase is re-run later (with new data or after resolving questions), existing resolutions are preserved.
+
+**Why this priority**: This is the core mechanism — without it, analyst decisions get lost on every re-run, violating the "preserve analyst work" principle.
+
+**Independent Test**: Raise a CLR, resolve it, re-run the phase, verify the resolution survives.
+
+**Acceptance Scenarios**:
+
+1. **Given** a domain with no resolutions, **When** a new CLR-001 question is raised, **Then** it appears in `domain.resolutions` with status "open".
+2. **Given** CLR-001 is resolved, **When** the phase is re-run and raises CLR-001 again, **Then** the existing resolution is preserved (merge behaviour).
+3. **Given** CLR-001 is resolved and CLR-002 is new, **When** the phase re-runs, **Then** CLR-001 stays resolved and CLR-002 is added as open.
+
+---
+
+### User Story 2 - Developer controls merge behaviour with --resolutions flag (Priority: P1)
+
+A developer re-running a phase can choose how new questions interact with existing resolutions: merge (default — preserve existing), overwrite (discard all, re-raise fresh), or skip (don't raise any new questions).
+
+**Why this priority**: Different re-run scenarios need different behaviour. Fresh data may invalidate old answers (overwrite), while minor updates should preserve work (merge).
+
+**Independent Test**: Run with each flag value and verify the expected merge behaviour.
+
+**Acceptance Scenarios**:
+
+1. **Given** existing resolutions and `--resolutions merge` (default), **When** new CLRs are raised, **Then** existing resolutions are kept, new questions are added as open.
+2. **Given** existing resolutions and `--resolutions overwrite`, **When** new CLRs are raised, **Then** all prior resolutions are discarded, all questions are fresh and open.
+3. **Given** existing resolutions and `--resolutions skip`, **When** the phase runs, **Then** no new questions are raised, only existing resolutions are used.
+
+---
+
+### User Story 3 - adm check reports clarification status (Priority: P2)
+
+A developer runs `adm check` and sees the count of open vs resolved clarifications per domain, giving visibility into analyst work remaining before ratcheting.
+
+**Why this priority**: Already partially implemented in issue #5 — this story ensures the clarify module integrates cleanly with the existing check command.
+
+**Independent Test**: Already covered by test_check.py (CLR reporting tests from issue #5).
+
+**Acceptance Scenarios**:
+
+1. **Given** 3 open and 2 resolved CLRs, **When** `adm check` runs, **Then** it reports "3 open, 2 resolved" as a warning.
+2. **Given** zero open CLRs, **When** `adm check` runs, **Then** it reports "ready for ratchet".
+
+---
+
+### Edge Cases
+
+- CLR-NNN IDs must be globally unique within a domain — never reused even if the question is removed.
+- What if a resolution references a column that no longer exists after data refresh? The resolution is still preserved (analyst decides if it's still valid).
+- Maximum number of CLRs per domain? No limit — but `adm status` shows the count.
+
+## Requirements
+
+### Functional Requirements
+
+- **FR-001**: The system MUST support structured clarification questions with format: ID (CLR-NNN), question text, context, resolution status (open/resolved)
+- **FR-002**: Resolutions MUST be stored in `.adm/project.json` under `domain.resolutions` as a `{CLR-ID: status}` mapping
+- **FR-003**: Re-running a phase MUST preserve existing resolutions by default (merge behaviour)
+- **FR-004**: The `--resolutions` flag MUST support three modes: `merge` (default), `overwrite`, `skip`
+- **FR-005**: CLR-NNN IDs MUST be globally unique within a domain and never reused
+- **FR-006**: The system MUST provide a function to raise a new CLR question (returns the next available ID)
+- **FR-007**: The system MUST provide a function to resolve a CLR by ID (sets status to "resolved" with answer text)
+- **FR-008**: `adm check` MUST report open vs resolved counts per domain (already implemented in issue #5)
+
+### Key Entities
+
+- **Clarification (CLR)**: A structured question raised during phase execution with ID, question, context, and status
+- **Resolution**: The analyst's answer to a CLR, persisted in the state file
+- **Merge Mode**: Controls how new CLRs interact with existing resolutions on phase re-run
+
+## Success Criteria
+
+### Measurable Outcomes
+
+- **SC-001**: Resolutions survive 100% of phase re-runs under default (merge) behaviour
+- **SC-002**: No analyst decision is ever silently discarded without explicit `--resolutions overwrite`
+- **SC-003**: The next available CLR-NNN ID is always correctly computed (no collisions)
+- **SC-004**: All three merge modes produce the expected state after re-run
+
+## Assumptions
+
+- The state schema already has `resolutions: dict[str, ResolutionStatus]` (implemented in issue #1)
+- Resolution status is either "open" or "resolved" (the ResolutionStatus enum already exists)
+- The actual question text and context are stored in artefact files (not in project.json — only the ID and status live in state)
+- The `--resolutions` flag is passed to phase commands, not to `adm check` or `adm status`
+- CLR IDs are sequential within a domain: CLR-001, CLR-002, etc.
diff --git a/src/adm_cli/clarify.py b/src/adm_cli/clarify.py
new file mode 100644
index 0000000..66ceeed
--- /dev/null
+++ b/src/adm_cli/clarify.py
@@ -0,0 +1,111 @@
+"""Clarify-gate pattern — structured CLR-NNN questions that persist across re-runs."""
+
+from __future__ import annotations
+
+from enum import Enum
+from pathlib import Path
+
+from .schema import ProjectState, ResolutionStatus, save_state
+
+
+class MergeMode(str, Enum):
+    MERGE = "merge"
+    OVERWRITE = "overwrite"
+    SKIP = "skip"
+
+
+def next_clr_id(domain: str, state: ProjectState) -> str:
+    """Compute the next available CLR-NNN ID for a domain."""
+    if domain not in state.domains:
+        raise ValueError(f"Domain '{domain}' does not exist")
+
+    existing = state.domains[domain].resolutions
+    if not existing:
+        return "CLR-001"
+
+    max_num = 0
+    for key in existing:
+        if key.startswith("CLR-") and key[4:].isdigit():
+            max_num = max(max_num, int(key[4:]))
+    return f"CLR-{max_num + 1:03d}"
+
+
+def raise_clr(
+    domain: str,
+    state: ProjectState,
+    state_path: Path,
+    question: str,
+    context: str | None = None,
+) -> str:
+    """Raise a new clarification question. Returns the assigned CLR-NNN ID."""
+    if domain not in state.domains:
+        raise ValueError(f"Domain '{domain}' does not exist")
+
+    clr_id = next_clr_id(domain, state)
+    state.domains[domain].resolutions[clr_id] = ResolutionStatus.OPEN
+    save_state(state, state_path)
+    return clr_id
+
+
+def resolve_clr(
+    domain: str,
+    clr_id: str,
+    state: ProjectState,
+    state_path: Path,
+) -> None:
+    """Mark a CLR as resolved."""
+    if domain not in state.domains:
+        raise ValueError(f"Domain '{domain}' does not exist")
+
+    resolutions = state.domains[domain].resolutions
+    if clr_id not in resolutions:
+        raise ValueError(f"CLR '{clr_id}' does not exist in domain '{domain}'")
+
+    resolutions[clr_id] = ResolutionStatus.RESOLVED
+    save_state(state, state_path)
+
+
+def merge_clarifications(
+    domain: str,
+    new_clr_ids: list[str],
+    state: ProjectState,
+    state_path: Path,
+    mode: MergeMode = MergeMode.MERGE,
+) -> None:
+    """Apply merge behaviour when a phase is re-run with new CLRs.
+
+    - merge: preserve existing resolutions, add new as open
+    - overwrite: discard all prior resolutions, all new are open
+    - skip: don't add any new CLRs, keep existing only
+    """
+    if domain not in state.domains:
+        raise ValueError(f"Domain '{domain}' does not exist")
+
+    resolutions = state.domains[domain].resolutions
+
+    if mode == MergeMode.OVERWRITE:
+        resolutions.clear()
+        for clr_id in new_clr_ids:
+            resolutions[clr_id] = ResolutionStatus.OPEN
+
+    elif mode == MergeMode.SKIP:
+        pass  # No changes — existing resolutions stay, no new ones added
+
+    elif mode == MergeMode.MERGE:
+        for clr_id in new_clr_ids:
+            if clr_id not in resolutions:
+                resolutions[clr_id] = ResolutionStatus.OPEN
+            # If already exists (resolved or open), leave as-is
+
+    save_state(state, state_path)
+
+
+def get_clr_summary(domain: str, state: ProjectState) -> tuple[int, int]:
+    """Return (open_count, resolved_count) for a domain."""
+    if domain not in state.domains:
+        raise ValueError(f"Domain '{domain}' does not exist")
+
+    resolutions = state.domains[domain].resolutions
+    open_count = sum(1 for s in resolutions.values() if s == ResolutionStatus.OPEN)
+    resolved_count = sum(1 for s in resolutions.values() if s == ResolutionStatus.RESOLVED)
+    return open_count, resolved_count
diff --git a/tests/test_clarify.py b/tests/test_clarify.py
new file mode 100644
index 0000000..2133534
--- /dev/null
+++ b/tests/test_clarify.py
@@ -0,0 +1,129 @@
+"""Tests for clarify-gate pattern."""
+
+from pathlib import Path
+
+import pytest
+
+from adm_cli.clarify import (
+    MergeMode,
+    get_clr_summary,
+    merge_clarifications,
+    next_clr_id,
+    raise_clr,
+    resolve_clr,
+)
+from adm_cli.schema import DomainState, ProjectState, ResolutionStatus, save_state
+
+
+def _setup(tmp_path: Path, **kwargs) -> tuple[ProjectState, Path]:
+    state = ProjectState(domains={"holdings": DomainState(**kwargs)})
+    state_path = tmp_path / ".adm" / "project.json"
+    save_state(state, state_path)
+    return state, state_path
+
+
+class TestNextClrId:
+    def test_first_id(self, tmp_path: Path):
+        state, _ = _setup(tmp_path)
+        assert next_clr_id("holdings", state) == "CLR-001"
+
+    def test_sequential(self, tmp_path: Path):
+        state, _ = _setup(tmp_path, resolutions={"CLR-001": ResolutionStatus.OPEN})
+        assert next_clr_id("holdings", state) == "CLR-002"
+
+    def test_gaps_use_max(self, tmp_path: Path):
+        state, _ = _setup(
+            tmp_path,
+            resolutions={"CLR-001": ResolutionStatus.RESOLVED, "CLR-005": ResolutionStatus.OPEN},
+        )
+        assert next_clr_id("holdings", state) == "CLR-006"
+
+    def test_nonexistent_domain(self, tmp_path: Path):
+        state, _ = _setup(tmp_path)
+        with pytest.raises(ValueError, match="does not exist"):
+            next_clr_id("nonexistent", state)
+
+
+class TestRaiseClr:
+    def test_raises_open(self, tmp_path: Path):
+        state, state_path = _setup(tmp_path)
+        clr_id = raise_clr("holdings", state, state_path, "What does column X mean?")
+        assert clr_id == "CLR-001"
+        assert state.domains["holdings"].resolutions["CLR-001"] == ResolutionStatus.OPEN
+
+    def test_sequential_ids(self, tmp_path: Path):
+        state, state_path = _setup(tmp_path)
+        id1 = raise_clr("holdings", state, state_path, "Q1")
+        id2 = raise_clr("holdings", state, state_path, "Q2")
+        assert id1 == "CLR-001"
+        assert id2 == "CLR-002"
+
+    def test_persists_to_disk(self, tmp_path: Path):
+        from adm_cli.schema import load_state
+
+        state, state_path = _setup(tmp_path)
+        raise_clr("holdings", state, state_path, "Q1")
+        reloaded = load_state(state_path)
+        assert "CLR-001" in reloaded.domains["holdings"].resolutions
+
+
+class TestResolveClr:
+    def test_marks_resolved(self, tmp_path: Path):
+        state, state_path = _setup(tmp_path, resolutions={"CLR-001": ResolutionStatus.OPEN})
+        resolve_clr("holdings", "CLR-001", state, state_path)
+        assert state.domains["holdings"].resolutions["CLR-001"] == ResolutionStatus.RESOLVED
+
+    def test_nonexistent_clr(self, tmp_path: Path):
+        state, state_path = _setup(tmp_path)
+        with pytest.raises(ValueError, match="does not exist"):
+            resolve_clr("holdings", "CLR-999", state, state_path)
+
+
+class TestMergeClarifications:
+    def test_merge_preserves_existing(self, tmp_path: Path):
+        state, state_path = _setup(
+            tmp_path,
+            resolutions={"CLR-001": ResolutionStatus.RESOLVED},
+        )
+        merge_clarifications("holdings", ["CLR-001", "CLR-002"], state, state_path, MergeMode.MERGE)
+        assert state.domains["holdings"].resolutions["CLR-001"] == ResolutionStatus.RESOLVED
+        assert state.domains["holdings"].resolutions["CLR-002"] == ResolutionStatus.OPEN
+
+    def test_overwrite_discards_all(self, tmp_path: Path):
+        state, state_path = _setup(
+            tmp_path,
+            resolutions={"CLR-001": ResolutionStatus.RESOLVED, "CLR-002": ResolutionStatus.RESOLVED},
+        )
+        merge_clarifications("holdings", ["CLR-001", "CLR-003"], state, state_path, MergeMode.OVERWRITE)
+        resolutions = state.domains["holdings"].resolutions
+        assert resolutions == {"CLR-001": ResolutionStatus.OPEN, "CLR-003": ResolutionStatus.OPEN}
+
+    def test_skip_adds_nothing(self, tmp_path: Path):
+        state, state_path = _setup(
+            tmp_path,
+            resolutions={"CLR-001": ResolutionStatus.RESOLVED},
+        )
+        merge_clarifications("holdings", ["CLR-002", "CLR-003"], state, state_path, MergeMode.SKIP)
+        assert "CLR-002" not in state.domains["holdings"].resolutions
+        assert state.domains["holdings"].resolutions["CLR-001"] == ResolutionStatus.RESOLVED
+
+
+class TestGetClrSummary:
+    def test_counts(self, tmp_path: Path):
+        state, _ = _setup(
+            tmp_path,
+            resolutions={
+                "CLR-001": ResolutionStatus.OPEN,
+                "CLR-002": ResolutionStatus.OPEN,
+                "CLR-003": ResolutionStatus.RESOLVED,
+            },
+        )
+        open_count, resolved_count = get_clr_summary("holdings", state)
+        assert open_count == 2
+        assert resolved_count == 1
+
+    def test_empty(self, tmp_path: Path):
+        state, _ = _setup(tmp_path)
+        open_count, resolved_count = get_clr_summary("holdings", state)
+        assert open_count == 0
+        assert resolved_count == 0