diff --git a/docs/guide/benchmarking/2026-06-11-measurement-coverage-audit.md b/docs/guide/benchmarking/2026-06-11-measurement-coverage-audit.md
new file mode 100644
index 0000000..862395b
--- /dev/null
+++ b/docs/guide/benchmarking/2026-06-11-measurement-coverage-audit.md
@@ -0,0 +1,243 @@
+# ELF Benchmark Measurement Coverage Audit - June 11, 2026
+
+Goal: Record what is actually measured today, where competitor comparisons are still
+not comparable, and which measurement reports should guide future ELF iteration.
+Read this when: You need to answer whether ELF has enough empirical evidence to
+claim a win, tie, loss, or non-claim against tracked memory, RAG, graph, and
+agent-continuity projects.
+Inputs: Fresh local runs of `cargo make real-world-memory` and
+`cargo make real-world-memory-live-adapters` on commit `286af8b`, plus
+`apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json`,
+`2026-06-11-competitor-strength-evidence-matrix.md`, and
+`2026-06-11-elf-iteration-direction-from-competitor-benchmarks.md`.
+Outputs: Fresh measured counters, scenario coverage, project coverage, and the next
+measurement reports needed before stronger ELF claims.
+
+## Executive Judgment
+
+The benchmark program is useful and already prevents misleading claims, but the
+current measured comparison is not complete enough to say ELF beats or ties every
+tracked project's strongest scenario.
+
+What is proven today:
+
+- ELF has a strong fixture-backed real-world benchmark contract: 38 jobs, 36 pass,
+  2 blocked operator boundaries, and no wrong results in the fixture aggregate.
+- ELF and qmd have comparable full-suite live real-world sweeps. They are effectively
+  tied on pass/fail shape: each has 38 jobs, 18 pass, 5 wrong_result, 2 blocked, and
+  13 not_encoded.
+- ELF is ahead on production-operation evidence among tracked systems because it has
+  checked-in provider synthetic, stress, backfill, backup/restore, and Qdrant rebuild
+  evidence.
+- The current comparison still undermeasures most competitor strengths. OpenViking
+  trajectory, mem0/OpenMemory entity history and UI, Letta core-vs-archival memory,
+  Graphiti/Zep temporal graph behavior, graph/RAG navigation, agentmemory and
+  claude-mem capture/continuity, and knowledge-page workflows remain non-claims.
+
+So the current adoption decision can remain "credible for bounded personal
+production," but the competitiveness objective remains open.
+
+## Fresh Runs
+
+These commands were run from an isolated report worktree based on `origin/main`:
+
+| Command | Result | Runtime |
+| --- | --- | ---: |
+| `cargo make real-world-memory` | pass | 42.38 seconds |
+| `cargo make real-world-memory-live-adapters` | pass | 121.93 seconds |
+
+The live adapter run emitted repeated Qdrant client/server compatibility warnings, but
+the command completed successfully and produced ELF and qmd JSON/Markdown reports.
+Treat that warning as a measurement-harness risk to keep visible, not as a current run
+failure.
+
+## Fixture Aggregate
+
+`cargo make real-world-memory` produced:
+
+| Metric | Value |
+| --- | ---: |
+| Jobs | `38` |
+| Encoded suites | `11` |
+| Pass | `36` |
+| Blocked | `2` |
+| Wrong result | `0` |
+| Lifecycle fail | `0` |
+| Incomplete | `0` |
+| Not encoded | `0` |
+| Unsupported claim | `0` |
+| Mean score | `0.947` |
+| Mean latency | `4.411 ms` |
+| Expected evidence recall | `77/77` |
+| Evidence coverage | `84/84` |
+| Source-ref coverage | `84/84` |
+| Quote coverage | `84/84` |
+
+This proves fixture contract breadth and scoring behavior. It does not prove every
+live adapter or competitor runtime can complete those jobs.
+
+## Live ELF/qmd Sweep
+
+`cargo make real-world-memory-live-adapters` produced:
+
+| Adapter | Jobs | Pass | Wrong result | Blocked | Not encoded | Mean score | Mean latency | Evidence recall | Evidence coverage |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| ELF live service adapter | `38` | `18` | `5` | `2` | `13` | `0.525` | `5.100 ms` | `41/77` | `48/84` |
+| qmd live CLI adapter | `38` | `18` | `5` | `2` | `13` | `0.512` | `719.758 ms` | `41/77` | `48/84` |
+
+This supports a narrow tie on the currently encoded live real-world suite shape. It
+does not support a broad ELF-over-qmd claim because qmd remains the stronger
+retrieval-debug UX reference and its deep profile is still not encoded.
+
+### Live Suite Breakdown
+
+ELF and qmd had the same suite status shape:
+
+| Suite | Jobs | Status breakdown |
+| --- | ---: | --- |
+| `trust_source_of_truth` | `1` | `pass:1` |
+| `work_resume` | `5` | `pass:5` |
+| `retrieval` | `5` | `pass:5` |
+| `project_decisions` | `5` | `pass:5` |
+| `personalization` | `1` | `pass:1` |
+| `memory_evolution` | `6` | `pass:1`, `wrong_result:5` |
+| `capture_integration` | `2` | `not_encoded:2` |
+| `consolidation` | `4` | `not_encoded:4` |
+| `knowledge_compilation` | `2` | `not_encoded:2` |
+| `operator_debugging_ux` | `1` | `not_encoded:1` |
+| `production_ops` | `6` | `blocked:2`, `not_encoded:4` |
+
+The five live wrong results are all memory-evolution jobs. The live adapters retrieve
+current evidence but do not yet provide the required historical conflict evidence
+links for current-vs-historical reasoning.
+
+## External Adapter Ledger
+
+The checked-in manifest records 21 adapter records across 17 unique project names.
+
+| Evidence class | Adapter records | Meaning |
+| --- | ---: | --- |
+| `fixture_backed` | `1` | ELF fixture scoring only. |
+| `live_baseline_only` | `6` | Docker same-corpus or lifecycle evidence without real-world job scoring. |
+| `live_real_world` | `2` | ELF and qmd live real-world sweeps. |
+| `research_gate` | `12` | Setup, source, resource, or output-contract gate only. |
+
+| Overall status | Adapter records |
+| --- | ---: |
+| `pass` | `1` |
+| `wrong_result` | `6` |
+| `lifecycle_fail` | `1` |
+| `blocked` | `6` |
+| `not_encoded` | `7` |
+
+The generated JSON report also emits `external_project_count: 19`, while the unique
+project-name count from the manifest is 17. The runner currently computes that field
+as adapter records whose project is not `ELF`, not as unique external project names.
+Interpret the unique manifest project list as the project coverage count.
+
+## Project Coverage
+
+| Project | Best current evidence | Current measured state | Strongest unproven scenario | Next measurement before claim |
+| --- | --- | --- | --- | --- |
+| ELF | `fixture_backed` plus `live_real_world` | Fixture aggregate passes except 2 blocked operator boundaries; live full sweep is `wrong_result`. | Full live memory evolution, live consolidation, live knowledge pages, live capture, live production ops. | Memory-evolution diagnostic report, then live operator/capture/consolidation reports. |
+| qmd | `live_real_world` plus `live_baseline_only` | Same live sweep shape as ELF; same-corpus baseline passes. | Deep retrieval-debug ergonomics and trace replay. | qmd/ELF deep retrieval-debug profile with expansion, fusion, rerank, and dropped-candidate traces. |
+| agentmemory | `live_baseline_only` | `lifecycle_fail`. | Durable coding-agent continuity and capture hooks. | Durable lifecycle and work-resume/capture adapter report. |
+| mem0/OpenMemory | `live_baseline_only` | `wrong_result`. | Entity history, lifecycle UI, OpenMemory inspection. | Same-corpus repair first, then entity-history and UI-readback report. |
+| memsearch | `live_baseline_only` | `wrong_result`; source-of-truth is `incomplete`. | Markdown canonical store and local reindex clarity. | Reindex/update/delete/reload plus source-of-truth report. |
+| OpenViking | `live_baseline_only` plus `research_gate` | Same-corpus retrieval is `wrong_result`; trajectory is `not_encoded`. | Hierarchical staged context trajectory. | Evidence-bearing retrieval fix, then staged trajectory report. |
+| claude-mem | `live_baseline_only` | `wrong_result`. | Progressive disclosure and automatic capture review. | Work-resume, operator-debugging, and capture/write-policy report. |
+| RAGFlow | `research_gate` | `blocked`. | RAG app workflow with document/chunk references. | Tiny Docker evidence-smoke with `reference.chunks` mapped to evidence ids. |
+| LightRAG | `research_gate` | `blocked`. | Graph/RAG context export with source-path citations. | Docker context-export report with explicit provider config and source citation mapping. |
+| GraphRAG | `research_gate` | `blocked`. | Graph summaries and document/text-unit evidence tables. | Cost-bounded Docker adapter report over a tiny corpus. |
+| Graphiti/Zep | `research_gate` | `blocked`. | Temporal graph facts and validity windows. | Docker-local temporal graph adapter report for current and historical facts. |
+| Letta | `research_gate` | `not_encoded`. | Core memory blocks versus archival memory. | Contained export contract, then core-vs-archival and decision-memory report. |
+| LangGraph | `research_gate` | `not_encoded`; direct memory backend is unsupported. | Checkpoint replay and fork/regression debugging. | Treat as benchmark-infra reference unless a memory-output contract emerges. |
+| nanograph | `research_gate` | `not_encoded`; full memory backend is unsupported. | Typed graph schema and query ergonomics. | Typed relation query report only if evidence ids can be emitted. |
+| llm-wiki | `research_gate` | `not_encoded`. | Wiki/page generation, query-save, lint and repair loops. | Contained page-generation report with citation and unsupported-claim lint. |
+| gbrain | `research_gate` | `not_encoded`; setup path is blocked. | Compiled truth pages, timelines, and brain navigation. | Docker-local brain repo setup proof, then compiled-truth/timeline report. |
+| graphify | `research_gate` | `blocked`. | Graph-compressed navigation with `graph.json` and `GRAPH_REPORT`. | Docker graph/report output report mapped to benchmark evidence ids. |
+
+## Scenario Coverage And Claims
+
+| Scenario | Current measured position | Claim allowed today | Missing measurement |
+| --- | --- | --- | --- |
+| Retrieval/debug | ELF and qmd live retrieval pass; qmd same-corpus baseline passes. | Tie on encoded live retrieval; no ELF-over-qmd UX claim. | qmd/ELF deep trace replay and debug ergonomics scoring. |
+| Work resume | ELF and qmd live pass. | ELF is credible on encoded work resume. | agentmemory, claude-mem, and OpenViking comparable continuity adapters. |
+| Project decisions | ELF and qmd live pass. | ELF is credible on encoded project-decision recovery. | Letta core/archival decision memory comparison. |
+| Source of truth | ELF and qmd live pass; ELF has stronger production restore/rebuild evidence. | ELF has strongest measured source-of-truth discipline. | memsearch source-of-truth reindex/reload evidence. |
+| Memory evolution | ELF and qmd live fail 5/6 jobs; fixture aggregate passes. | No live superiority claim. | Historical conflict evidence links and Graphiti/Zep temporal comparison. |
+| Consolidation | Fixture aggregate passes; live adapters are not encoded. | Fixture-only claim. | Live proposal generation with lineage, confidence, and review-action audit. |
+| Knowledge pages | Fixture aggregate passes; live adapters are not encoded. | Fixture-only claim. | Live page rebuild/lint plus llm-wiki, gbrain, GraphRAG, and graphify comparisons. |
+| Operator debugging | Fixture aggregate passes; live adapters are not encoded. | Fixture-only claim. | Trace hydration, stage attribution, dropped-candidate, and repair-action scoring. |
+| Capture/write policy | Fixture aggregate passes; live adapters are not encoded. | Fixture-only claim. | agentmemory/claude-mem style capture with redaction and evidence binding. |
+| Production ops | ELF has separate production-provider/backfill/restore evidence; live sweep is not a full production-ops pass. | Bounded personal-production adoption claim with caveats. | Private corpus manifest and credentialed provider gates. |
+| Personalization | ELF and qmd live pass one scoped preference job. | Narrow encoded pass only. | mem0/OpenMemory and Letta entity/preference history comparison. |
+| Context trajectory | Not comparable. | No claim. | OpenViking staged hierarchy/trajectory scoring. |
+| Core-vs-archival memory | Not comparable. | No claim. | Letta contained export and ELF core-block benchmark. |
+| Graph/RAG navigation | Research gates and blocked adapters only. | No claim. | RAGFlow, LightRAG, GraphRAG, Graphiti/Zep, and graphify Docker reports. |
+
+## Next Measurement Reports
+
+Order these by decision value, not implementation convenience:
+
+1. ELF/qmd retrieval-debug deep profile
+   - Why: qmd is the closest measured live competitor and still stronger as a
+     debugging reference.
+   - Output: trace-level comparison of expansion, dense/sparse retrieval, fusion,
+     rerank, dropped candidates, and command-line replay.
+
+2. ELF/qmd live memory-evolution diagnostic
+   - Why: both systems currently fail 5/6 live memory-evolution jobs.
+   - Output: per-job evidence-link failure analysis for current-vs-historical facts,
+     supersession, and relation temporal validity.
+
+3. Live operator-debugging and capture/write-policy report
+   - Why: these are daily-use agent-memory qualities, currently fixture-only or
+     not_encoded in live sweeps.
+   - Output: trace hydration, raw-SQL avoidance, redaction, exclusion, write-policy,
+     and repair-action scoring.
+
+4. Continuity and context-trajectory report
+   - Why: agentmemory, claude-mem, and OpenViking represent real user expectations
+     around automatic capture, progressive disclosure, and staged context.
+   - Output: comparable work-resume/capture/trajectory jobs or typed blockers.
+
+5. Personalization and core-memory report
+   - Why: mem0/OpenMemory and Letta represent product expectations ELF should absorb
+     before claiming better personalization or operating context.
+   - Output: entity history, preference correction, UI/readback, core-vs-archival,
+     and project-decision scoring.
+
+6. Knowledge and graph/RAG report pack
+   - Why: llm-wiki, gbrain, graphify, GraphRAG, LightRAG, RAGFlow, and Graphiti/Zep
+     cover knowledge synthesis and graph navigation that ELF currently cannot claim.
+   - Output: Docker-contained artifacts mapped to evidence ids, or typed setup and
+     resource blockers.
+
+Before publishing the next aggregate report, clarify or rename the generated
+`external_project_count` field so readers do not confuse non-ELF adapter records with
+unique external projects.
+
+## Fail Criteria
+
+Use these criteria for future reports:
+
+- `pass`: comparable scenario is encoded, run, and evidence-backed.
+- `wrong_result`: the system ran but answered with wrong, stale, unsupported, or
+  insufficiently evidenced memory.
+- `not_encoded`: the runner does not yet exercise the scenario. This is not a win or
+  loss.
+- `blocked`: safe measurement needs missing credentials, private data, resource
+  envelope acceptance, setup proof, or an export contract.
+- `unsupported`: the project shape is not a direct memory-system comparison target.
+- Fixture evidence cannot be promoted into live runtime evidence.
+- Live baseline evidence cannot be promoted into real-world job evidence.
+- Research-gate evidence cannot be promoted into pass/fail product quality evidence.
+
+## Bottom Line
+
+ELF is on a strong path because its benchmark methodology is stricter than a normal
+leaderboard, and its production evidence is unusually concrete. The next work is not
+to declare victory. The next work is to measure the strongest user-facing patterns in
+adjacent projects, then decide which ones ELF should absorb behind fresh benchmark
+gates.
diff --git a/docs/guide/benchmarking/index.md b/docs/guide/benchmarking/index.md
index b273eae..fd2569d 100644
--- a/docs/guide/benchmarking/index.md
+++ b/docs/guide/benchmarking/index.md
@@ -54,6 +54,10 @@ cleanup, use `docs/guide/single_user_production.md`.
 - `2026-06-11-elf-iteration-direction-from-competitor-benchmarks.md`: current
   optimization-direction report that translates measured benchmark data and competitor
   strengths into prioritized ELF iteration themes and explicit non-claims.
+- `2026-06-11-measurement-coverage-audit.md`: fresh coverage audit that separates
+  current measured ELF/qmd data, fixture evidence, external adapter ledger coverage,
+  scenario non-claims, and the next measurement reports needed before stronger
+  competitor claims.
 - `real_world_agent_memory_benchmark.md`: operator overview for the v1 real-world
   agent memory benchmark contract, including suite taxonomy, typed report states,
   knowledge-compilation fixture tasks, and the production-ops fixture target.
diff --git a/docs/research/2026-06-11-measurement-coverage-audit.json b/docs/research/2026-06-11-measurement-coverage-audit.json
new file mode 100644
index 0000000..b04d86e
--- /dev/null
+++ b/docs/research/2026-06-11-measurement-coverage-audit.json
@@ -0,0 +1,124 @@
+{
+  "schema": "elf.benchmark_measurement_coverage_audit/v1",
+  "run_id": "2026-06-11-measurement-coverage-audit",
+  "commit": "286af8b",
+  "created_at": "2026-06-11",
+  "scope": "ELF memory-system competitiveness measurement coverage, external competitor comparison evidence, and next report directions",
+  "commands": [
+    {
+      "command": "cargo make real-world-memory",
+      "status": "pass",
+      "runtime_seconds": 42.38,
+      "artifact": "tmp/real-world-memory/real-world-memory-report.json"
+    },
+    {
+      "command": "cargo make real-world-memory-live-adapters",
+      "status": "pass",
+      "runtime_seconds": 121.93,
+      "artifact": "tmp/real-world-memory/live-adapters/"
+    }
+  ],
+  "fixture_aggregate": {
+    "job_count": 38,
+    "encoded_suite_count": 11,
+    "pass": 36,
+    "wrong_result": 0,
+    "lifecycle_fail": 0,
+    "incomplete": 0,
+    "blocked": 2,
+    "not_encoded": 0,
+    "unsupported_claim": 0,
+    "mean_score": 0.947,
+    "mean_latency_ms": 4.411,
+    "expected_evidence_total": 77,
+    "expected_evidence_matched": 77,
+    "evidence_required_count": 84,
+    "evidence_covered_count": 84
+  },
+  "live_real_world_adapters": [
+    {
+      "adapter": "ELF live service adapter",
+      "job_count": 38,
+      "encoded_suite_count": 11,
+      "pass": 18,
+      "wrong_result": 5,
+      "blocked": 2,
+      "not_encoded": 13,
+      "mean_score": 0.525,
+      "mean_latency_ms": 5.1,
+      "expected_evidence_total": 77,
+      "expected_evidence_matched": 41,
+      "evidence_required_count": 84,
+      "evidence_covered_count": 48
+    },
+    {
+      "adapter": "qmd live CLI adapter",
+      "job_count": 38,
+      "encoded_suite_count": 11,
+      "pass": 18,
+      "wrong_result": 5,
+      "blocked": 2,
+      "not_encoded": 13,
+      "mean_score": 0.512,
+      "mean_latency_ms": 719.758,
+      "expected_evidence_total": 77,
+      "expected_evidence_matched": 41,
+      "evidence_required_count": 84,
+      "evidence_covered_count": 48
+    }
+  ],
+  "live_suite_breakdown": [
+    {"suite": "trust_source_of_truth", "jobs": 1, "status_counts": {"pass": 1}},
+    {"suite": "work_resume", "jobs": 5, "status_counts": {"pass": 5}},
+    {"suite": "retrieval", "jobs": 5, "status_counts": {"pass": 5}},
+    {"suite": "project_decisions", "jobs": 5, "status_counts": {"pass": 5}},
+    {"suite": "personalization", "jobs": 1, "status_counts": {"pass": 1}},
+    {"suite": "memory_evolution", "jobs": 6, "status_counts": {"pass": 1, "wrong_result": 5}},
+    {"suite": "capture_integration", "jobs": 2, "status_counts": {"not_encoded": 2}},
+    {"suite": "consolidation", "jobs": 4, "status_counts": {"not_encoded": 4}},
+    {"suite": "knowledge_compilation", "jobs": 2, "status_counts": {"not_encoded": 2}},
+    {"suite": "operator_debugging_ux", "jobs": 1, "status_counts": {"not_encoded": 1}},
+    {"suite": "production_ops", "jobs": 6, "status_counts": {"blocked": 2, "not_encoded": 4}}
+  ],
+  "adapter_ledger": {
+    "adapter_records": 21,
+    "unique_project_names": 17,
+    "external_project_count_note": "The generated report field external_project_count currently counts non-ELF adapter records, not unique external project names.",
+    "evidence_class_counts": {
+      "fixture_backed": 1,
+      "live_baseline_only": 6,
+      "live_real_world": 2,
+      "research_gate": 12
+    },
+    "overall_status_counts": {
+      "pass": 1,
+      "wrong_result": 6,
+      "lifecycle_fail": 1,
+      "blocked": 6,
+      "not_encoded": 7
+    }
+  },
+  "claim_boundary": {
+    "elf_vs_qmd": "tie_on_current_encoded_live_real_world_shape_not_overall_win",
+    "elf_personal_production": "credible_with_bounded_caveats",
+    "broad_competitor_superiority": "not_proven",
+    "major_unmeasured_strengths": [
+      "qmd_deep_retrieval_debug",
+      "OpenViking_context_trajectory",
+      "mem0_OpenMemory_entity_history_ui",
+      "agentmemory_claude_mem_capture_continuity",
+      "Letta_core_vs_archival_memory",
+      "Graphiti_Zep_temporal_graph",
+      "RAG_graph_navigation",
+      "llm_wiki_gbrain_graphify_knowledge_workflows"
+    ]
+  },
+  "next_reports": [
+    "ELF/qmd retrieval-debug deep profile",
+    "ELF/qmd live memory-evolution diagnostic",
+    "Live operator-debugging and capture/write-policy report",
+    "Continuity and context-trajectory report",
+    "Personalization and core-memory report",
+    "Knowledge and graph/RAG report pack"
+  ]
+}