diff --git a/apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json b/apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json
index 9ee1acb6..beea373a 100644
--- a/apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json
+++ b/apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json
@@ -918,7 +918,7 @@
       "overall_status": "blocked",
       "setup": {
         "status": "blocked",
-        "evidence": "RAGFlow remains a large RAG system watch item; D1/D2 research must prove a Docker-safe corpus ingest and query path before adapter implementation."
+        "evidence": "XY-882 marks RAGFlow as an adapter_candidate, but the runner still needs a Docker-safe tiny-corpus ingest/query smoke before any live adapter claim."
       },
       "run": {
         "status": "not_encoded",
@@ -930,9 +930,9 @@
       },
       "capabilities": [
         {
-          "capability": "d1_d2_research_before_adapter",
-          "status": "blocked",
-          "evidence": "The inventory marks RAGFlow as D0 pending deep dive."
+          "capability": "adapter_candidate_verdict",
+          "status": "not_encoded",
+          "evidence": "XY-882 completed D1/D2 feasibility research and marks RAGFlow adapter_candidate; no adapter run is encoded."
         },
         {
           "capability": "docker_service_setup",
@@ -985,20 +985,25 @@
             "label": "RAGFlow docs",
             "url": "https://ragflow.io/docs/",
             "evidence": "Official deployment and setup documentation."
+          },
+          {
+            "label": "RAGFlow HTTP API reference",
+            "url": "https://raw.githubusercontent.com/infiniflow/ragflow/main/docs/references/http_api_reference.md",
+            "evidence": "Official reference for OpenAI-compatible responses with reference chunks and document metadata."
           }
         ],
-        "setup_path": "Research the official Docker deployment, corpus ingest API, query API, and artifact export before adding a runner.",
+        "setup_path": "Implement a tiny Docker evidence-smoke runner using the official Docker deployment, dataset ingest API, and OpenAI-compatible query API.",
         "runtime_boundary": "Future runs must use docker-compose.baseline.yml or a nested Docker-isolated service profile without host-global installs.",
         "resource_expectation": "Large multi-service RAG stack; record CPU/GPU mode, memory, disk, startup time, and provider credential needs before scoring.",
         "retry_guidance": [
-          "Complete a D1/D2 setup and API deep dive.",
-          "Prototype a tiny Docker smoke that reaches ingest and query before adding quality checks."
+          "Start with CPU mode and a generated tiny text corpus.",
+          "Record image pull/build size, expanded disk use, startup time, vm.max_map_count handling, and provider boundaries before scoring."
         ],
-        "research_depth": "D0 watch item; D1/D2 required"
+        "research_depth": "D2 feasibility verdict: adapter_candidate (XY-882); research_gate only, adapter not encoded"
       },
       "follow_up": {
-        "title": "[ELF benchmark adapter] Research RAGFlow Docker adapter feasibility",
-        "reason": "The project is too large to score fairly without setup, resource, and API mapping research."
+        "title": "[ELF benchmark adapter] Implement RAGFlow Docker evidence-smoke adapter",
+        "reason": "Created as XY-885. XY-882 found a Docker boundary and reference-chunk output contract; implementation must prove a tiny ingest/query run before any quality claim."
       }
     },
     {
@@ -1011,7 +1016,7 @@
       "overall_status": "blocked",
       "setup": {
         "status": "blocked",
-        "evidence": "LightRAG requires D1/D2 research on Docker setup, LLM/embedding configuration, persistence, and context output before adapter implementation."
+        "evidence": "XY-882 marks LightRAG as an adapter_candidate, but the runner still needs a Docker context-export adapter before any live result."
       },
       "run": {
         "status": "not_encoded",
@@ -1024,8 +1029,8 @@
       "capabilities": [
         {
           "capability": "graph_augmented_rag_setup",
-          "status": "blocked",
-          "evidence": "The inventory marks LightRAG as D0 pending deep dive."
+          "status": "not_encoded",
+          "evidence": "XY-882 completed setup/output feasibility research; graph-augmented RAG execution is still not encoded."
         },
         {
           "capability": "retrieved_context_export",
@@ -1078,20 +1083,30 @@
             "label": "LightRAG Docker docs",
             "url": "https://github.com/HKUDS/LightRAG/blob/main/docs/DockerDeployment.md",
             "evidence": "Official Docker deployment reference."
+          },
+          {
+            "label": "LightRAG API server docs",
+            "url": "https://github.com/HKUDS/LightRAG/blob/main/docs/LightRAG-API-Server.md",
+            "evidence": "Official query-mode and context-output reference."
+          },
+          {
+            "label": "LightRAG core programming docs",
+            "url": "https://github.com/HKUDS/LightRAG/blob/main/docs/ProgramingWithCore.md",
+            "evidence": "Official source-id and file-path citation reference."
           }
         ],
-        "setup_path": "Research Docker Compose with explicit LLM, embedding, rerank, and storage configuration before adding a benchmark runner.",
+        "setup_path": "Implement Docker Compose with explicit LLM, embedding, rerank, storage, workspace, and data-volume configuration, then export context-only query output.",
         "runtime_boundary": "Docker-only service profile with generated corpus mounted as container-local input.",
         "resource_expectation": "Graph extraction and local model choices may dominate runtime; record backend choices, cache sizes, and provider needs.",
         "retry_guidance": [
           "Run a tiny Docker ingest/query smoke with deterministic or local providers.",
           "Verify returned contexts can be mapped to required evidence IDs."
         ],
-        "research_depth": "D0 watch item; D1/D2 required"
+        "research_depth": "D2 feasibility verdict: adapter_candidate (XY-882); research_gate only, adapter not encoded"
       },
       "follow_up": {
-        "title": "[ELF benchmark adapter] Research LightRAG graph-RAG adapter feasibility",
-        "reason": "Graph extraction, persistence, and context output must be understood before fair scoring."
+        "title": "[ELF benchmark adapter] Implement LightRAG Docker context-export adapter",
+        "reason": "Created as XY-886. XY-882 found a Docker service path and context/source mapping contract; implementation must prove evidence export before scoring."
       }
     },
     {
@@ -1104,7 +1119,7 @@
       "overall_status": "blocked",
       "setup": {
         "status": "blocked",
-        "evidence": "GraphRAG indexing cost and source-citation mapping require D1/D2 research before adapter implementation."
+        "evidence": "XY-882 marks GraphRAG as an adapter_candidate, but indexing cost and source mapping still need a cost-bounded Docker implementation before live scoring."
       },
       "run": {
         "status": "not_encoded",
@@ -1118,7 +1133,7 @@
         {
           "capability": "indexing_resource_envelope",
           "status": "blocked",
-          "evidence": "Official docs warn that indexing can be expensive; the benchmark must start small and record costs."
+          "evidence": "XY-882 requires the first adapter to start with a tiny corpus and record indexing cost before any scale or quality claim."
         },
         {
           "capability": "source_citation_mapping",
@@ -1171,20 +1186,25 @@
             "label": "GraphRAG docs",
             "url": "https://microsoft.github.io/graphrag/",
             "evidence": "Official documentation for indexing and querying."
+          },
+          {
+            "label": "GraphRAG output tables",
+            "url": "https://microsoft.github.io/graphrag/index/outputs/",
+            "evidence": "Official output schema with document, text unit, community, and relationship identifiers."
           }
         ],
-        "setup_path": "Research a tiny CLI index/query path with explicit model configuration and source mapping.",
+        "setup_path": "Implement a tiny CLI/API index/query path with explicit model configuration and source mapping from parquet output tables.",
         "runtime_boundary": "Docker-only Python CLI run with generated corpus and container-local artifacts.",
         "resource_expectation": "Indexing may be expensive; record model calls, cache size, elapsed time, and maximum corpus size used.",
         "retry_guidance": [
-          "Complete D1/D2 indexing and query-output research.",
-          "Add a cost-bounded smoke before any scale or quality claim."
+          "Add a cost-bounded smoke before any scale or quality claim.",
+          "Fail typed if source document or text_unit identifiers cannot be mapped to expected evidence IDs."
         ],
-        "research_depth": "D0 watch item; D1/D2 required"
+        "research_depth": "D2 feasibility verdict: adapter_candidate (XY-882); research_gate only, adapter not encoded"
       },
       "follow_up": {
-        "title": "[ELF benchmark adapter] Research GraphRAG cost-bounded adapter path",
-        "reason": "Indexing cost, graph summaries, and citation guarantees need proof before scoring."
+        "title": "[ELF benchmark adapter] Implement GraphRAG cost-bounded Docker adapter",
+        "reason": "Created as XY-887. XY-882 found a Docker-bounded CLI/API path and output-table evidence handles; implementation must stay tiny and cost-recorded."
       }
     },
     {
@@ -1197,7 +1217,7 @@
       "overall_status": "not_encoded",
       "setup": {
         "status": "not_encoded",
-        "evidence": "Graphiti/Zep is D1 reviewed as a temporal graph-memory reference, but no Docker adapter is implemented."
+        "evidence": "XY-882 marks Graphiti/Zep as an adapter_candidate, but no Docker temporal graph adapter is implemented."
       },
       "run": {
         "status": "not_encoded",
@@ -1211,7 +1231,7 @@
         {
           "capability": "temporal_graph_memory",
           "status": "not_encoded",
-          "evidence": "Temporal fact validity is a reference dimension but not an executable adapter output."
+          "evidence": "Temporal fact validity has a scoped adapter candidate path, but no executable adapter output is encoded."
         },
         {
           "capability": "docker_graph_store_setup",
@@ -1259,16 +1279,30 @@
             "label": "Zep Graphiti overview",
             "url": "https://www.getzep.com/platform/graphiti/",
             "evidence": "Official product documentation for temporal context graph behavior."
+          },
+          {
+            "label": "Graphiti quick start",
+            "url": "https://help.getzep.com/graphiti/getting-started/quick-start",
+            "evidence": "Official setup, episode ingest, and search output reference."
+          },
+          {
+            "label": "Graphiti FalkorDB configuration",
+            "url": "https://help.getzep.com/graphiti/configuration/falkor-db-configuration",
+            "evidence": "Official Docker-local FalkorDB setup reference."
           }
         ],
-        "setup_path": "Define a Docker-local graph store and provider configuration, then encode add/query current-versus-historical fact jobs.",
+        "setup_path": "Implement a Docker-local FalkorDB or Neo4j graph store and provider configuration, then encode add/query current-versus-historical fact jobs.",
         "runtime_boundary": "Docker-only service or SDK run with graph store state under benchmark artifacts.",
         "resource_expectation": "Requires graph store plus LLM/embedding configuration; record service startup, storage size, and provider boundaries.",
         "retry_guidance": [
           "Prototype a tiny temporal fact add/query run.",
           "Map valid_at/invalid_at evidence to memory_evolution scoring."
         ],
-        "research_depth": "D1 reviewed; adapter not encoded"
+        "research_depth": "D1 feasibility verdict: adapter_candidate (XY-882); research_gate only, adapter not encoded"
+      },
+      "follow_up": {
+        "title": "[ELF benchmark adapter] Implement Graphiti/Zep temporal graph adapter",
+        "reason": "Created as XY-888. XY-882 found a Docker-local graph-store path and fact/validity-window output contract for memory_evolution scoring."
       }
     },
     {
@@ -1357,7 +1391,7 @@
           "Create a tiny Docker agent with archival memory search.",
           "Score core-versus-archival retrieval only after source evidence can be exported."
         ],
-        "research_depth": "D1 reviewed; adapter not encoded"
+        "research_depth": "D1 feasibility verdict: research_only (XY-882); core/archival reference, adapter not encoded"
       }
     },
     {
@@ -1431,7 +1465,7 @@
           "Encode one replay/fork failure recovery job.",
           "Keep LangGraph classified as replay reference unless memory retrieval is actually exercised."
         ],
-        "research_depth": "D1 reviewed; adapter not encoded"
+        "research_depth": "D1 feasibility verdict: research_only (XY-882); replay/checkpoint reference, adapter not encoded"
       }
     },
     {
@@ -1505,7 +1539,7 @@
           "Define a minimal schema for memory_evolution facts.",
           "Score typed query output only if it cites fixture evidence IDs."
         ],
-        "research_depth": "D1 reviewed; adapter not encoded"
+        "research_depth": "D1 feasibility verdict: research_only (XY-882); typed graph DX reference, adapter not encoded"
       }
     },
     {
@@ -1579,7 +1613,7 @@
           "Prototype a fixture-only page build with explicit citations.",
           "Do not score until generated sections can be mapped to evidence IDs."
         ],
-        "research_depth": "D1 reviewed; adapter not encoded"
+        "research_depth": "D1 feasibility verdict: research_only (XY-882); derived wiki workflow reference, adapter not encoded"
       }
     },
     {
@@ -1663,7 +1697,7 @@
           "Prototype a tiny brain repo with one current-truth page and timeline.",
           "Score only if compiled truth cites the source timeline evidence."
         ],
-        "research_depth": "D1 reviewed; adapter not encoded"
+        "research_depth": "D1 feasibility verdict: blocked (XY-882); Docker-local brain repo and database path not proven"
       }
     },
     {
@@ -1676,7 +1710,7 @@
       "overall_status": "not_encoded",
       "setup": {
         "status": "not_encoded",
-        "evidence": "graphify is D1 reviewed as a graph-navigation reference, but no Docker adapter is implemented."
+        "evidence": "XY-882 marks graphify as an adapter_candidate for a Docker-only CLI/materializer path, but no adapter is implemented."
       },
       "run": {
         "status": "not_encoded",
@@ -1690,7 +1724,7 @@
         {
           "capability": "graph_report_generation",
           "status": "not_encoded",
-          "evidence": "Graph reports and assistant query flows are not executed by the runner."
+          "evidence": "Graph reports and query output have a candidate scoring path, but they are not executed by the runner."
         },
         {
           "capability": "multimodal_code_graph",
@@ -1733,16 +1767,25 @@
             "label": "graphify repository",
             "url": "https://github.com/safishamsi/graphify",
             "evidence": "Official source for graphify graph extraction and query workflow."
+          },
+          {
+            "label": "graphify README",
+            "url": "https://github.com/safishamsi/graphify/blob/v3/README.md",
+            "evidence": "Official CLI, output artifact, query, and source-location contract."
           }
         ],
-        "setup_path": "Install graphify inside Docker, build a graph/report from a generated corpus, and export query evidence.",
-        "runtime_boundary": "Docker-only CLI or skill run over mounted benchmark corpus.",
+        "setup_path": "Install graphify inside Docker, build a graph/report from a generated corpus, and export query evidence without installing host-global assistant hooks.",
+        "runtime_boundary": "Docker-only CLI/materializer run over mounted benchmark corpus.",
         "resource_expectation": "Graph build cost scales with corpus and model choices; record build time, graph size, and generated report size.",
         "retry_guidance": [
           "Start with a generated public code/document corpus.",
           "Score graph-guided answers only when report nodes cite source evidence IDs."
         ],
-        "research_depth": "D1 reviewed; adapter not encoded"
+        "research_depth": "D1 feasibility verdict: adapter_candidate (XY-882); research_gate only, adapter not encoded"
+      },
+      "follow_up": {
+        "title": "[ELF benchmark adapter] Implement graphify Docker graph-report adapter",
+        "reason": "Created as XY-889. XY-882 found a Docker-only CLI/materializer path and source-file/source-location output contract."
       }
     }
   ]
diff --git a/apps/elf-eval/tests/real_world_job_benchmark.rs b/apps/elf-eval/tests/real_world_job_benchmark.rs
index 45ac5b1f..48461dd4 100644
--- a/apps/elf-eval/tests/real_world_job_benchmark.rs
+++ b/apps/elf-eval/tests/real_world_job_benchmark.rs
@@ -321,7 +321,9 @@ fn assert_external_adapter_manifest_records(report: &Value) -> Result<()> {
 	assert_eq!(ragflow.pointer("/overall_status").and_then(Value::as_str), Some("blocked"));
 	assert_eq!(
 		ragflow.pointer("/execution_metadata/research_depth").and_then(Value::as_str),
-		Some("D0 watch item; D1/D2 required")
+		Some(
+			"D2 feasibility verdict: adapter_candidate (XY-882); research_gate only, adapter not encoded"
+		)
 	);
 	assert_eq!(
 		ragflow.pointer("/execution_metadata/sources/0/url").and_then(Value::as_str),
diff --git a/docs/guide/benchmarking/2026-06-10-real-world-comparison-report.md b/docs/guide/benchmarking/2026-06-10-real-world-comparison-report.md
index 490fecfb..0b91ce4e 100644
--- a/docs/guide/benchmarking/2026-06-10-real-world-comparison-report.md
+++ b/docs/guide/benchmarking/2026-06-10-real-world-comparison-report.md
@@ -110,6 +110,22 @@ separate:
 | `live_real_world` | 2 | Targeted ELF and qmd adapters execute representative `real_world_job` prompts and scoring. |
 | `research_gate` | 12 | Source/setup/runtime/resource/retry metadata for future adapter paths; not fixture-backed or live execution evidence. |
 
+XY-882 added D1/D2 feasibility verdicts inside the research-gate lane. RAGFlow
+([XY-885](https://linear.app/hack-ink/issue/XY-885/elf-benchmark-adapter-implement-ragflow-docker-evidence-smoke-adapter)),
+LightRAG
+([XY-886](https://linear.app/hack-ink/issue/XY-886/elf-benchmark-adapter-implement-lightrag-docker-context-export-adapter)),
+GraphRAG
+([XY-887](https://linear.app/hack-ink/issue/XY-887/elf-benchmark-adapter-implement-graphrag-cost-bounded-docker-adapter)),
+Graphiti/Zep
+([XY-888](https://linear.app/hack-ink/issue/XY-888/elf-benchmark-adapter-implement-graphitizep-temporal-graph-adapter)),
+and graphify
+([XY-889](https://linear.app/hack-ink/issue/XY-889/elf-benchmark-adapter-implement-graphify-docker-graph-report-adapter))
+are now adapter implementation candidates because they have scoped Docker boundaries
+and evidence-linked output contracts. Letta, LangGraph, nanograph, and llm-wiki remain
+`research_only`; gbrain remains `blocked` until a Docker-local brain repo and database
+path is proven. These verdicts do not change any record into live adapter pass
+evidence.
+
 Adapter-level status after refreshing the manifest:
 
 | Project | Evidence class | Overall status | What is proven | What is not proven |
@@ -125,8 +141,8 @@ Adapter-level status after refreshing the manifest:
 | claude-mem | `live_baseline_only` | `wrong_result` | Progressive disclosure and local viewer remain UX references. | Current Docker evidence is not a clean same-corpus pass and progressive disclosure jobs are not encoded. |
 | qmd deep profile | `research_gate` | `not_encoded` | The stress-profile command path and source metadata are recorded for a future deeper retrieval-debug run. | No expanded qmd stress artifact or broader real-world suite pass is checked in. |
 | OpenViking deep profile | `research_gate` | `incomplete` | The deeper context-trajectory gate inherits the current Docker local-embedding setup blocker. | No hierarchical trajectory suite result is claimed. |
-| RAGFlow, LightRAG, GraphRAG | `research_gate` | `blocked` | Official sources and setup/resource/retry expectations are recorded. | D1/D2 research, Docker runtime proof, and evidence-output mapping are required before adapter implementation. |
-| Graphiti/Zep, Letta, LangGraph, nanograph, llm-wiki, gbrain, graphify | `research_gate` | `not_encoded` | D1/D2-inspired adapter directions have source/setup/runtime/resource/retry metadata. | No Docker-isolated `real_world_job` adapter has run for these projects. |
+| RAGFlow, LightRAG, GraphRAG | `research_gate` | `blocked` | Official sources, setup/resource/retry expectations, and XY-882 adapter-candidate verdicts are recorded. | Docker runtime proof and real_world_job evidence-output mapping are still required before any live adapter claim. |
+| Graphiti/Zep, Letta, LangGraph, nanograph, llm-wiki, gbrain, graphify | `research_gate` | `not_encoded` | XY-882 records Graphiti/Zep and graphify as adapter candidates, Letta/LangGraph/nanograph/llm-wiki as research-only, and gbrain as blocked. | No Docker-isolated `real_world_job` adapter has run for these projects. |
 
 External summary counters: `21` adapter records, `19` non-ELF adapter records,
 `21` Docker-default, `0` host-global-install requirements, `2` live real-world
@@ -151,8 +167,9 @@ report:
 | memsearch same-corpus and real-world coverage | `wrong_result` / `incomplete` | Fix Docker same-corpus retrieval/reindex evidence before scoring Markdown-first real-world jobs. |
 | OpenViking Docker local embedding path | `incomplete` | `[ELF benchmark adapter] Pin OpenViking Docker local embedding dependency path`. |
 | claude-mem durable/progressive-disclosure adapter | `wrong_result` / `not_encoded` | Add durable local repository and progressive-disclosure job coverage before UX parity claims. |
-| RAGFlow, LightRAG, and GraphRAG adapter feasibility | `blocked` research gates | Run D1/D2 research on setup, resource envelope, corpus ingest, query output, source mapping, and Docker retry path before implementation. |
-| Graphiti/Zep, Letta, LangGraph, nanograph, llm-wiki, gbrain, and graphify adapters | `not_encoded` research gates | Implement only after a scoped Docker path can emit evidence-linked outputs for the relevant real-world suites. |
+| RAGFlow, LightRAG, GraphRAG, Graphiti/Zep, and graphify adapters | `research_gate` adapter candidates | Follow-up issues [XY-885](https://linear.app/hack-ink/issue/XY-885/elf-benchmark-adapter-implement-ragflow-docker-evidence-smoke-adapter), [XY-886](https://linear.app/hack-ink/issue/XY-886/elf-benchmark-adapter-implement-lightrag-docker-context-export-adapter), [XY-887](https://linear.app/hack-ink/issue/XY-887/elf-benchmark-adapter-implement-graphrag-cost-bounded-docker-adapter), [XY-888](https://linear.app/hack-ink/issue/XY-888/elf-benchmark-adapter-implement-graphitizep-temporal-graph-adapter), and [XY-889](https://linear.app/hack-ink/issue/XY-889/elf-benchmark-adapter-implement-graphify-docker-graph-report-adapter) must run only Docker-contained adapter smokes that emit evidence-linked outputs before any live result claim. |
+| Letta, LangGraph, nanograph, and llm-wiki adapters | `research_only` research gates | Keep as architecture or workflow references until a contained output contract is selected. |
+| gbrain adapter | `blocked` research gate | Revisit only after a Docker-local brain repo and database path can be proven without operator-owned state. |
 
 ## Adoption Implications
 
diff --git a/docs/guide/research/comparison_external_projects.md b/docs/guide/research/comparison_external_projects.md
index 8e549544..0d297ec2 100644
--- a/docs/guide/research/comparison_external_projects.md
+++ b/docs/guide/research/comparison_external_projects.md
@@ -71,6 +71,11 @@ fixture-backed, live-baseline-only, or live-real-world evidence. Other external
 projects remain live-baseline-only, incomplete, blocked, or not encoded until their
 own `real_world_job` adapters run.
 
+XY-882 adds D1/D2 feasibility verdicts for the RAG and graph-memory research gates.
+`adapter_candidate` means an implementation follow-up is justified because a scoped
+Docker boundary and evidence-linked output contract exist. It does not mean a Docker
+adapter has run, and it does not change the `research_gate` evidence class.
+
 Benchmark suite labels:
 
 | Suite | Real-world job shape |
@@ -106,15 +111,20 @@ Project-to-suite map:
 | Graphiti / Zep | `rw.graph-temporal`, `rw.resume-evidence` | Temporal entities, relations, fact triples, validity windows, and graph search directly target stale/contradictory factual memory. | Add fact triples with validity changes, query current and historical answers, and score invalidation/append behavior under contradiction traps. | Docs-grounded D1; no benchmark adapter evidence. Confidence: medium-high for temporal-graph dimension. | ELF graph-lite covers evidence-linked validity windows and current/historical relation context; Graphiti/Zep remains the reference for broader temporal graph workflows. |
 | nanograph | `rw.graph-temporal`, `rw.retrieval-debug` | Typed schema and typed query ergonomics are relevant to making ELF graph-lite interactions inspectable and hard to misuse. | Define typed graph schemas and queries for the same fact set, then score developer-visible validation, query shape, and explainability rather than retrieval quality alone. | Docs-grounded D1; no benchmark adapter evidence. Confidence: medium for DX reference, low for memory-system comparison. | ELF should borrow typed graph ergonomics without treating nanograph as a full memory backend. |
 
-Pending watch items remain D0 even when they have checked-in `research_gate` adapter
-records. Keep them out of benchmark strength claims until current D1/D2 evidence is
-gathered and a Docker-isolated adapter actually runs:
-
-| Watch item | Candidate suite if promoted | Minimum evidence needed before adapter or quality claims |
-| ---------- | --------------------------- | ------------------------------------------------------- |
-| RAGFlow | `rw.resume-evidence`, `rw.graph-navigation`, `rw.retrieval-debug` | D1/D2 deep dive on deployability, corpus ingestion, graph/RAG retrieval path, API/CLI outputs, and Docker resource envelope. |
-| LightRAG | `rw.graph-navigation`, `rw.graph-temporal`, `rw.retrieval-debug` | D1/D2 deep dive on graph extraction/update semantics, local persistence, query output, and whether stale/corrected facts can be tested fairly. |
-| GraphRAG | `rw.graph-navigation`, `rw.knowledge-synthesis`, `rw.retrieval-debug` | D1/D2 deep dive on indexing cost, graph summaries, update/rebuild behavior, source citation guarantees, and task-level output inspectability. |
+XY-882 feasibility verdicts for RAG and graph-memory gates:
+
+| Project | Verdict | Docker boundary | Evidence-linked output contract | Follow-up |
+| ------- | ------- | --------------- | ------------------------------- | --------- |
+| RAGFlow | `adapter_candidate` | Official Docker Compose path, but the first adapter must use a tiny CPU corpus and record the 4 CPU / 16 GB RAM / 50 GB disk envelope, image size, `vm.max_map_count`, provider needs, and retry behavior. | OpenAI-compatible and agent completion responses can include `reference.chunks` with chunk id, document id/name, metadata, dataset id, positions, and similarity fields. | [XY-885](https://linear.app/hack-ink/issue/XY-885/elf-benchmark-adapter-implement-ragflow-docker-evidence-smoke-adapter); no live pass claim. |
+| LightRAG | `adapter_candidate` | Docker Compose server with explicit LLM, embedding, rerank, storage, workspace, and data-volume configuration. | Context-only query modes can return the context prepared for the LLM; core APIs can insert documents with ids and source file paths. | [XY-886](https://linear.app/hack-ink/issue/XY-886/elf-benchmark-adapter-implement-lightrag-docker-context-export-adapter); no live pass claim. |
+| GraphRAG | `adapter_candidate` | Cost-bounded Docker Python CLI/API run over a generated tiny corpus with container-local parquet artifacts. | Output tables contain generated UUIDs, human-readable ids, source documents, text units, community reports, and text-unit links for graph summaries and relationships. | [XY-887](https://linear.app/hack-ink/issue/XY-887/elf-benchmark-adapter-implement-graphrag-cost-bounded-docker-adapter); no live pass claim. |
+| Graphiti / Zep | `adapter_candidate` | Docker-local FalkorDB or Neo4j plus Python SDK runner with provider config captured under benchmark artifacts. | Search results and fact triples expose UUIDs, fact text, and validity windows (`valid_at` / `invalid_at`) that map to memory-evolution scoring. | [XY-888](https://linear.app/hack-ink/issue/XY-888/elf-benchmark-adapter-implement-graphitizep-temporal-graph-adapter); no live pass claim. |
+| graphify | `adapter_candidate` | Docker-only CLI/materializer using `pip install graphifyy` over a mounted corpus; host-global assistant hooks are out of scope. | `graph.json`, `GRAPH_REPORT.md`, and graph query output include edge types, confidence tags, source files, and source locations. | [XY-889](https://linear.app/hack-ink/issue/XY-889/elf-benchmark-adapter-implement-graphify-docker-graph-report-adapter); no live pass claim. |
+| Letta | `research_only` | Docker server exists, but current docs require explicit embedding configuration and steer Letta Code evaluation toward non-Docker local/frontier-model exploration. | Core/archival memory and shared blocks remain useful semantics, but no contained evidence export is selected for this adapter batch. | No implementation issue. |
+| LangGraph | `research_only` | A Docker harness is possible, but the project is an agent-state/checkpoint framework rather than a standalone memory adapter. | Store search and checkpoints are references for replay-regression jobs, not a direct external memory output contract here. | No implementation issue. |
+| nanograph | `research_only` | Official positioning is one CLI / one folder / no server / no Docker. | Typed schema, query, CDC, and search ergonomics remain graph-lite DX inspiration. | No implementation issue. |
+| llm-wiki | `research_only` | Plugin or instruction-file workflow would require a contained harness before scoring; host-global plugin installs are not proof. | Wiki compile/query/lint/audit workflows are derived-knowledge references, not current adapter outputs. | No implementation issue. |
+| gbrain | `blocked` | A Docker-local brain repo and database setup path was not proven in this lane. | Compiled truth, timeline, and source attribution are strong, but not enough for implementation without contained setup proof. | No implementation issue until Docker setup is proven. |
 
 ## Where ELF Is Not Yet The Reference
 
@@ -129,7 +139,7 @@ gathered and a Docker-isolated adapter actually runs:
 | Agent replay and forkable regression debugging | LangGraph | ELF traces are replay evidence for retrieval, not full persisted agent-state replay with side-effect boundaries. |
 | Derived knowledge pages and lint/repair loops | llm-wiki, gbrain | ELF does not yet ship rebuildable entity/project pages with unsupported-claim lint as a first-class workflow. |
 | Scheduled consolidation as a product surface | Always-On Memory Agent | ELF's target should be reviewable derived consolidation, but the scheduling/operator-control workflow is not implemented. |
-| Graph-compressed navigation over large corpora | graphify, GraphRAG/LightRAG watch items | ELF relation context is bounded and evidence-linked, but broader graph report/navigation workflows remain future work. |
+| Graph-compressed navigation over large corpora | graphify, GraphRAG/LightRAG adapter candidates | ELF relation context is bounded and evidence-linked, but broader graph report/navigation workflows remain future work. |
 
 ## June 2026 Agentmemory And Dreaming Refresh
 
@@ -400,6 +410,20 @@ Snapshot date for this subsection: February 17, 2026.
 
 ## Extended Source Map
 
+- RAGFlow:
+  - https://ragflow.io/docs/
+  - https://github.com/infiniflow/ragflow/blob/main/docker/README.md
+  - https://raw.githubusercontent.com/infiniflow/ragflow/main/docs/references/http_api_reference.md
+- LightRAG:
+  - https://github.com/HKUDS/LightRAG
+  - https://raw.githubusercontent.com/HKUDS/LightRAG/main/docs/DockerDeployment.md
+  - https://raw.githubusercontent.com/HKUDS/LightRAG/main/docs/LightRAG-API-Server.md
+  - https://raw.githubusercontent.com/HKUDS/LightRAG/main/docs/ProgramingWithCore.md
+- GraphRAG:
+  - https://microsoft.github.io/graphrag/
+  - https://microsoft.github.io/graphrag/index/inputs/
+  - https://microsoft.github.io/graphrag/index/outputs/
+  - https://microsoft.github.io/graphrag/query/local_search/
 - mem0:
   - https://docs.mem0.ai/platform/features/entity-scoped-memory
   - https://docs.mem0.ai/platform/features/graph-memory
diff --git a/docs/guide/research/research_projects_inventory.md b/docs/guide/research/research_projects_inventory.md
index 23c6f565..960fcfec 100644
--- a/docs/guide/research/research_projects_inventory.md
+++ b/docs/guide/research/research_projects_inventory.md
@@ -27,17 +27,36 @@ Last updated: June 10, 2026.
 | [qmd](https://github.com/tobi/qmd) | D2 | Reviewed | `rw.retrieval-debug`, `rw.lifecycle-staleness`, `rw.resume-evidence` | Retrieval routing, weighted fusion, and local-first explainability | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json` |
 | [claude-mem](https://github.com/thedotmack/claude-mem) | D2 | Reviewed | `rw.operator-continuity`, `rw.resume-evidence`, `rw.retrieval-debug` | Progressive disclosure and strong operator workflow | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json` |
 | [OpenViking](https://github.com/volcengine/OpenViking) | D2 | Reviewed | `rw.context-trajectory`, `rw.resume-evidence`, `rw.retrieval-debug` | Filesystem context paradigm, hierarchical retrieval, trajectory observability | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json` |
-| [llm-wiki](https://github.com/nvk/llm-wiki) | D1 | Reviewed | `rw.knowledge-synthesis`, `rw.resume-evidence` | LLM-maintained wiki pattern, topic-scoped knowledge bases, query-save and lint workflows | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json` |
-| [gbrain](https://github.com/garrytan/gbrain) | D1 | Reviewed | `rw.knowledge-synthesis`, `rw.operator-continuity` | Operational knowledge brain, `compiled_truth` + timeline pages, enrichment and maintenance loops | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json` |
+| [llm-wiki](https://github.com/nvk/llm-wiki) | D1 | Reviewed; XY-882 verdict `research_only` | `rw.knowledge-synthesis`, `rw.resume-evidence` | LLM-maintained wiki pattern, topic-scoped knowledge bases, query-save and lint workflows | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json`; `docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json` |
+| [gbrain](https://github.com/garrytan/gbrain) | D1 | Reviewed; XY-882 verdict `blocked` | `rw.knowledge-synthesis`, `rw.operator-continuity` | Operational knowledge brain, `compiled_truth` + timeline pages, enrichment and maintenance loops; blocked on Docker-local brain repo and database proof | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json`; `docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json` |
 | [Always-On Memory Agent](https://github.com/GoogleCloudPlatform/generative-ai/tree/main/gemini/agents/always-on-memory-agent) | D1 | Reviewed | `rw.consolidation-review`, `rw.operator-continuity` | Always-on multimodal ingest + scheduled consolidation loop with simple local ops surface | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json` |
-| [graphify](https://github.com/safishamsi/graphify) | D1 | Reviewed | `rw.graph-navigation`, `rw.knowledge-synthesis`, `rw.resume-evidence` | Multimodal graph compression, deterministic code extraction, and always-on graph-guided assistant workflow | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json` |
-| [Letta](https://github.com/letta-ai/letta) | D1 | Reviewed | `rw.core-archival`, `rw.operator-continuity` | Core vs archival memory split, shared blocks | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json` |
-| [LangGraph](https://docs.langchain.com/oss/python/langgraph/persistence) | D1 | Reviewed | `rw.replay-regression`, `rw.resume-evidence` | Checkpoint/replay mindset for quality regression workflows | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json` |
-| [Graphiti / Zep](https://help.getzep.com/graphiti/core-concepts/temporal-awareness) | D1 | Reviewed | `rw.graph-temporal`, `rw.resume-evidence` | Temporal fact validity model for graph-like memory evolution | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json` |
-| [nanograph](https://github.com/nanograph/nanograph) | D1 | Reviewed; research gate added | `rw.graph-temporal`, `rw.retrieval-debug` | Typed schema + typed query ergonomics for graph-lite developer experience | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json`; `apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json` |
-| [RAGFlow](https://github.com/infiniflow/ragflow) | D0 | Research gate added; D1/D2 still required before adapter | Candidate `rw.resume-evidence`, `rw.graph-navigation`, `rw.retrieval-debug`; no strength claim | Potential framework integration discussion; not yet audited to adoption level | `apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json`; see watch-item evidence requirements in `docs/guide/research/comparison_external_projects.md` |
-| [LightRAG](https://github.com/HKUDS/LightRAG) | D0 | Research gate added; D1/D2 still required before adapter | Candidate `rw.graph-navigation`, `rw.graph-temporal`, `rw.retrieval-debug`; no strength claim | Graph-augmented RAG strategy relevance; not yet audited to adoption level | `apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json`; see watch-item evidence requirements in `docs/guide/research/comparison_external_projects.md` |
-| [GraphRAG](https://github.com/microsoft/graphrag) | D0 | Research gate added; D1/D2 still required before adapter | Candidate `rw.graph-navigation`, `rw.knowledge-synthesis`, `rw.retrieval-debug`; no strength claim | Graph-based retrieval concepts; not yet audited to implementation decision level | `apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json`; see watch-item evidence requirements in `docs/guide/research/comparison_external_projects.md` |
+| [graphify](https://github.com/safishamsi/graphify) | D1 | Reviewed; XY-882 verdict `adapter_candidate` | `rw.graph-navigation`, `rw.knowledge-synthesis`, `rw.resume-evidence` | Multimodal graph compression, deterministic code extraction, and graph/report outputs with source-file/source-location references | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json`; `docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json` |
+| [Letta](https://github.com/letta-ai/letta) | D1 | Reviewed; XY-882 verdict `research_only` | `rw.core-archival`, `rw.operator-continuity` | Core vs archival memory split, shared blocks; not an implementation candidate until a supported contained server path can export evidence | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json`; `docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json` |
+| [LangGraph](https://docs.langchain.com/oss/python/langgraph/persistence) | D1 | Reviewed; XY-882 verdict `research_only` | `rw.replay-regression`, `rw.resume-evidence` | Checkpoint/replay mindset for quality regression workflows; not a standalone memory backend adapter | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json`; `docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json` |
+| [Graphiti / Zep](https://help.getzep.com/graphiti/core-concepts/temporal-awareness) | D1 | Reviewed; XY-882 verdict `adapter_candidate` | `rw.graph-temporal`, `rw.resume-evidence` | Temporal fact validity model with Docker-local graph-store options and UUID/fact/validity-window output | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json`; `docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json` |
+| [nanograph](https://github.com/nanograph/nanograph) | D1 | Reviewed; XY-882 verdict `research_only` | `rw.graph-temporal`, `rw.retrieval-debug` | Typed schema + typed query ergonomics for graph-lite developer experience; official shape is no server/no Docker | `docs/guide/research/comparison_external_projects.md`; `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json`; `apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json`; `docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json` |
+| [RAGFlow](https://github.com/infiniflow/ragflow) | D2 feasibility gate | Research gate remains; XY-882 verdict `adapter_candidate` | Candidate `rw.resume-evidence`, `rw.graph-navigation`, `rw.retrieval-debug`; no live strength claim | Docker setup is resource-heavy but documented; API references expose document/chunk evidence handles for a tiny-corpus adapter smoke | `apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json`; `docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json` |
+| [LightRAG](https://github.com/HKUDS/LightRAG) | D2 feasibility gate | Research gate remains; XY-882 verdict `adapter_candidate` | Candidate `rw.graph-navigation`, `rw.graph-temporal`, `rw.retrieval-debug`; no live strength claim | Docker compose path, context-only query modes, and source file-path citation shape support an implementation follow-up | `apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json`; `docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json` |
+| [GraphRAG](https://github.com/microsoft/graphrag) | D2 feasibility gate | Research gate remains; XY-882 verdict `adapter_candidate` | Candidate `rw.graph-navigation`, `rw.knowledge-synthesis`, `rw.retrieval-debug`; no live strength claim | Cost-bounded CLI/API path and parquet output tables expose document, text-unit, and graph-summary handles for evidence mapping | `apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json`; `docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json` |
+
+## June 10, 2026 Adapter Feasibility Verdicts
+
+XY-882 resolved the D1/D2 feasibility gate for the RAG and graph-memory
+`research_gate` records. These verdicts do not change any project into live adapter
+evidence; they only decide whether an implementation follow-up is justified.
+
+| Project | Verdict | Follow-up rule |
+| ------- | ------- | -------------- |
+| RAGFlow | `adapter_candidate` | Follow-up issue: [XY-885](https://linear.app/hack-ink/issue/XY-885/elf-benchmark-adapter-implement-ragflow-docker-evidence-smoke-adapter), a tiny Docker evidence-smoke adapter that records the resource envelope and maps `reference.chunks` to benchmark evidence. |
+| LightRAG | `adapter_candidate` | Follow-up issue: [XY-886](https://linear.app/hack-ink/issue/XY-886/elf-benchmark-adapter-implement-lightrag-docker-context-export-adapter), a Docker context-export adapter using explicit LLM/embedding config and source file-path citations. |
+| GraphRAG | `adapter_candidate` | Follow-up issue: [XY-887](https://linear.app/hack-ink/issue/XY-887/elf-benchmark-adapter-implement-graphrag-cost-bounded-docker-adapter), a cost-bounded Docker CLI/API adapter over a tiny corpus and parquet output tables. |
+| Graphiti / Zep | `adapter_candidate` | Follow-up issue: [XY-888](https://linear.app/hack-ink/issue/XY-888/elf-benchmark-adapter-implement-graphitizep-temporal-graph-adapter), a Docker-local temporal graph adapter that scores current/historical fact validity. |
+| graphify | `adapter_candidate` | Follow-up issue: [XY-889](https://linear.app/hack-ink/issue/XY-889/elf-benchmark-adapter-implement-graphify-docker-graph-report-adapter), a Docker-only CLI/materializer adapter over `graph.json` and `GRAPH_REPORT.md`; host-global assistant hooks remain out of scope. |
+| Letta | `research_only` | Keep as a core/archival memory reference until a supported contained path can export archival-memory evidence for scoring. |
+| LangGraph | `research_only` | Keep as a checkpoint/replay regression reference, not a standalone external memory adapter. |
+| nanograph | `research_only` | Keep as typed graph DX inspiration; official shape is no server/no Docker. |
+| llm-wiki | `research_only` | Keep as a derived knowledge-page workflow reference; host-global plugin installs are not adapter proof. |
+| gbrain | `blocked` | Revisit only after a Docker-local brain repo and database path can be proven without operator-owned state. |
 
 ## June 2026 Activity Snapshot
 
@@ -73,6 +92,7 @@ replacing ELF's evidence-bound service contract.
 - Current June 2026 research runs:
   - `docs/research/2026-06-08-agent-memory-selection.json`
   - `docs/research/2026-06-09-xy-841-external-memory-benchmark-dimensions.json`
+  - `docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json`
 
 ## Notes
 
diff --git a/docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json b/docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json
new file mode 100644
index 00000000..9f42812b
--- /dev/null
+++ b/docs/research/2026-06-10-xy-882-rag-graph-adapter-feasibility.json
@@ -0,0 +1,348 @@
+{
+  "schema": "research-run/2",
+  "run_id": "2026-06-10-xy-882-rag-graph-adapter-feasibility",
+  "question": "Which RAG and graph-memory research gates should become Docker-bounded adapter implementation candidates for ELF real-world benchmarks?",
+  "success_criteria": [
+    "Give RAGFlow, LightRAG, GraphRAG, Graphiti/Zep, Letta, LangGraph, nanograph, llm-wiki, gbrain, and graphify one explicit verdict: adapter_candidate, research_only, blocked, or reject.",
+    "Separate setup/resource feasibility from product quality; heavy setup is not treated as a quality failure.",
+    "Require adapter_candidate projects to have both a Docker-contained path and an evidence-linked output contract.",
+    "Keep all researched projects in the research_gate evidence class until a Docker adapter executes real_world_job scoring."
+  ],
+  "constraints": [
+    "Do not implement adapters in this issue.",
+    "Do not use host-global installs as proof.",
+    "Do not claim live adapter pass evidence from source or docs review.",
+    "Create implementation follow-ups only for adapter candidates with a scoped Docker boundary and evidence-linked output."
+  ],
+  "stop_rule": "Stop when every target project has a verdict, adapter candidates have scoped follow-up issue titles, and the docs/manifest still label these records as research gates rather than live evidence.",
+  "primary_hypothesis": "RAGFlow, LightRAG, GraphRAG, Graphiti/Zep, and graphify have enough Docker-bounded setup and evidence-output shape to justify implementation follow-ups; Letta, LangGraph, nanograph, and llm-wiki remain research-only references; gbrain remains blocked until a Docker-local brain repo/database path is proven.",
+  "rival_hypotheses": [
+    "All projects should remain research-only because none has executed in the benchmark runner.",
+    "All projects with official Docker or CLI instructions should become adapter candidates.",
+    "RAGFlow should be rejected because its official resource envelope is large."
+  ],
+  "falsifiers": [
+    "If a candidate cannot run without host-global state, it is not an adapter implementation candidate for this benchmark lane.",
+    "If a candidate cannot emit source IDs, document IDs, file locations, citations, or equivalent evidence handles, it cannot support real_world_job scoring.",
+    "If a project is a useful architecture reference but not a standalone memory/retrieval output path, it should remain research_only."
+  ],
+  "coverage": {
+    "mode": "primary_source_docs_and_existing_repo_contracts",
+    "min_source_families": 4
+  },
+  "events": [
+    {
+      "seq": 1,
+      "type": "probe_completed",
+      "remaining_option_count": 4,
+      "independent_option_questions": [
+        "Does the project expose a Docker-contained setup path?",
+        "Does the project expose corpus ingest and query output that can map back to source evidence?",
+        "Is the project a direct adapter candidate, a reference-only design input, blocked by missing Docker proof, or rejected?"
+      ],
+      "external_slices": [
+        "RAGFlow",
+        "LightRAG",
+        "GraphRAG",
+        "Graphiti/Zep",
+        "Letta",
+        "LangGraph",
+        "nanograph",
+        "llm-wiki",
+        "gbrain",
+        "graphify"
+      ]
+    },
+    {
+      "seq": 2,
+      "type": "evidence_recorded",
+      "evidence": [
+        {
+          "id": "E1",
+          "kind": "contract",
+          "summary": "The real-world benchmark spec defines research_gate records as source/setup/runtime/resource/retry metadata for future implementation; research gates must not count as fixture-backed, live-baseline, or live-real-world evidence.",
+          "source_family": "repo_spec",
+          "source_locator": "docs/spec/real_world_agent_memory_benchmark_v1.md"
+        },
+        {
+          "id": "E2",
+          "kind": "setup",
+          "summary": "RAGFlow official quickstart documents Docker startup, 4 CPU / 16 GB RAM / 50 GB disk prerequisites, x86/Nvidia support, image-size caveats, dataset creation, chunk visibility, and citation-backed retrieval testing.",
+          "source_family": "upstream_docs",
+          "source_locator": "https://ragflow.io/docs/"
+        },
+        {
+          "id": "E3",
+          "kind": "output_contract",
+          "summary": "RAGFlow HTTP API can include reference metadata and returns reference chunks containing chunk id, content, document id, document name, document metadata, dataset id, positions, and similarity scores.",
+          "source_family": "upstream_docs",
+          "source_locator": "https://raw.githubusercontent.com/infiniflow/ragflow/main/docs/references/http_api_reference.md"
+        },
+        {
+          "id": "E4",
+          "kind": "setup",
+          "summary": "LightRAG Docker docs describe docker compose startup, generated compose files, persistent data paths, environment-driven LLM and embedding configuration, and optional Docker-local vLLM embedding/rerank services.",
+          "source_family": "upstream_docs",
+          "source_locator": "https://raw.githubusercontent.com/HKUDS/LightRAG/main/docs/DockerDeployment.md"
+        },
+        {
+          "id": "E5",
+          "kind": "output_contract",
+          "summary": "LightRAG supports query prefixes including context-only modes, can return the context prepared for the LLM, supports inserting documents with stable ids, and traces sources through file_paths.",
+          "source_family": "upstream_docs",
+          "source_locator": "https://raw.githubusercontent.com/HKUDS/LightRAG/main/docs/LightRAG-API-Server.md"
+        },
+        {
+          "id": "E6",
+          "kind": "output_contract",
+          "summary": "GraphRAG writes parquet output tables with UUIDs and human-readable ids; communities and reports carry text_unit_ids, and text_units carry raw text plus document ids and relationship/entity ids.",
+          "source_family": "upstream_docs",
+          "source_locator": "https://microsoft.github.io/graphrag/index/outputs/"
+        },
+        {
+          "id": "E7",
+          "kind": "setup",
+          "summary": "GraphRAG input and query docs describe a CLI/API indexing and local-search path over structured documents, raw text chunks, graph data, and query context builders.",
+          "source_family": "upstream_docs",
+          "source_locator": "https://microsoft.github.io/graphrag/"
+        },
+        {
+          "id": "E8",
+          "kind": "output_contract",
+          "summary": "Graphiti/Zep requires Python plus Neo4j or FalkorDB, supports Docker-local FalkorDB, adds episodes or fact triples, and search results include UUID, fact text, valid_at, and invalid_at fields.",
+          "source_family": "upstream_docs",
+          "source_locator": "https://help.getzep.com/graphiti/getting-started/quick-start"
+        },
+        {
+          "id": "E9",
+          "kind": "boundary",
+          "summary": "Letta remains a strong core/archival memory reference, but Docker use needs explicit embedding configuration and the current docs steer new Letta Code users away from Docker-first evaluation.",
+          "source_family": "upstream_docs",
+          "source_locator": "https://docs.letta.com/guides/docker/"
+        },
+        {
+          "id": "E10",
+          "kind": "boundary",
+          "summary": "LangGraph persistence provides checkpoints, replay, stores, and semantic memory search, but it is an agent-state framework rather than a standalone external memory service adapter.",
+          "source_family": "upstream_docs",
+          "source_locator": "https://docs.langchain.com/oss/python/langgraph/persistence"
+        },
+        {
+          "id": "E11",
+          "kind": "boundary",
+          "summary": "nanograph documents one CLI, one folder, schema-as-code, no server, no cloud, and no Docker; this makes it a graph-lite DX reference rather than a Docker adapter candidate for this lane.",
+          "source_family": "upstream_docs",
+          "source_locator": "https://www.nanograph.io/"
+        },
+        {
+          "id": "E12",
+          "kind": "boundary",
+          "summary": "llm-wiki ships as agent plugins or portable instructions with wiki query, compile, lint, audit, and output workflows; it is a derived knowledge workflow reference, not a service adapter candidate without a contained plugin harness.",
+          "source_family": "upstream_docs",
+          "source_locator": "https://github.com/nvk/llm-wiki"
+        },
+        {
+          "id": "E13",
+          "kind": "boundary",
+          "summary": "gbrain has strong compiled-truth, append-only timeline, and source attribution contracts, but this lane did not prove a Docker-local brain repository and database setup path.",
+          "source_family": "upstream_docs",
+          "source_locator": "https://raw.githubusercontent.com/garrytan/gbrain/master/docs/guides/compiled-truth.md"
+        },
+        {
+          "id": "E14",
+          "kind": "output_contract",
+          "summary": "graphify can run over a folder, produces graph.html, GRAPH_REPORT.md, graph.json, and cache artifacts, and query output includes node labels, edge types, confidence tags, source files, and source locations.",
+          "source_family": "upstream_docs",
+          "source_locator": "https://raw.githubusercontent.com/safishamsi/graphify/v3/README.md"
+        }
+      ]
+    },
+    {
+      "seq": 3,
+      "type": "project_verdicts_recorded",
+      "verdicts": [
+        {
+          "project": "RAGFlow",
+          "verdict": "adapter_candidate",
+          "supporting_evidence_ids": [
+            "E2",
+            "E3"
+          ],
+          "docker_boundary": "Nested Docker service profile or baseline compose service using official RAGFlow Docker Compose, capped to a tiny corpus and CPU mode first.",
+          "output_contract": "Map RAGFlow reference.chunks fields to real_world_job expected evidence ids.",
+          "follow_up_title": "[ELF benchmark adapter] Implement RAGFlow Docker evidence-smoke adapter",
+          "follow_up_issue": "XY-885",
+          "follow_up_url": "https://linear.app/hack-ink/issue/XY-885/elf-benchmark-adapter-implement-ragflow-docker-evidence-smoke-adapter"
+        },
+        {
+          "project": "LightRAG",
+          "verdict": "adapter_candidate",
+          "supporting_evidence_ids": [
+            "E4",
+            "E5"
+          ],
+          "docker_boundary": "Docker Compose LightRAG server with explicit LLM, embedding, rerank, and data-volume configuration.",
+          "output_contract": "Use context-only query modes and file_paths-backed citations for evidence scoring.",
+          "follow_up_title": "[ELF benchmark adapter] Implement LightRAG Docker context-export adapter",
+          "follow_up_issue": "XY-886",
+          "follow_up_url": "https://linear.app/hack-ink/issue/XY-886/elf-benchmark-adapter-implement-lightrag-docker-context-export-adapter"
+        },
+        {
+          "project": "GraphRAG",
+          "verdict": "adapter_candidate",
+          "supporting_evidence_ids": [
+            "E6",
+            "E7"
+          ],
+          "docker_boundary": "Cost-bounded Docker Python CLI/API run over a generated tiny corpus with container-local parquet artifacts.",
+          "output_contract": "Map documents, text_units, communities, and community_reports output tables back to source evidence ids.",
+          "follow_up_title": "[ELF benchmark adapter] Implement GraphRAG cost-bounded Docker adapter",
+          "follow_up_issue": "XY-887",
+          "follow_up_url": "https://linear.app/hack-ink/issue/XY-887/elf-benchmark-adapter-implement-graphrag-cost-bounded-docker-adapter"
+        },
+        {
+          "project": "Graphiti/Zep",
+          "verdict": "adapter_candidate",
+          "supporting_evidence_ids": [
+            "E8"
+          ],
+          "docker_boundary": "Docker-local FalkorDB or Neo4j plus Python SDK runner with provider configuration explicit in benchmark artifacts.",
+          "output_contract": "Score UUID, fact, valid_at, and invalid_at search output against memory_evolution current/historical evidence.",
+          "follow_up_title": "[ELF benchmark adapter] Implement Graphiti/Zep temporal graph adapter",
+          "follow_up_issue": "XY-888",
+          "follow_up_url": "https://linear.app/hack-ink/issue/XY-888/elf-benchmark-adapter-implement-graphitizep-temporal-graph-adapter"
+        },
+        {
+          "project": "Letta",
+          "verdict": "research_only",
+          "supporting_evidence_ids": [
+            "E9"
+          ],
+          "reason": "Keep as core/archival memory semantics reference; do not create an implementation issue until a supported, contained server path can export archival evidence for scoring."
+        },
+        {
+          "project": "LangGraph",
+          "verdict": "research_only",
+          "supporting_evidence_ids": [
+            "E10"
+          ],
+          "reason": "Keep as checkpoint/replay regression reference; it is not a standalone external memory adapter candidate in this benchmark lane."
+        },
+        {
+          "project": "nanograph",
+          "verdict": "research_only",
+          "supporting_evidence_ids": [
+            "E11"
+          ],
+          "reason": "Keep as typed graph DX inspiration; official positioning is no server/no Docker and no real_world_job evidence contract is proven."
+        },
+        {
+          "project": "llm-wiki",
+          "verdict": "research_only",
+          "supporting_evidence_ids": [
+            "E12"
+          ],
+          "reason": "Keep as derived knowledge-page workflow inspiration; no host-global plugin install may be used as adapter proof."
+        },
+        {
+          "project": "gbrain",
+          "verdict": "blocked",
+          "supporting_evidence_ids": [
+            "E13"
+          ],
+          "reason": "The evidence contract is strong, but a Docker-local brain repo and database path must be proven before an implementation issue is safe."
+        },
+        {
+          "project": "graphify",
+          "verdict": "adapter_candidate",
+          "supporting_evidence_ids": [
+            "E14"
+          ],
+          "docker_boundary": "Docker-only CLI/materializer run using pip-installed graphifyy over mounted benchmark corpus, with no assistant global hook install.",
+          "output_contract": "Score graph.json query output and GRAPH_REPORT.md source-file/source-location references against expected evidence.",
+          "follow_up_title": "[ELF benchmark adapter] Implement graphify Docker graph-report adapter",
+          "follow_up_issue": "XY-889",
+          "follow_up_url": "https://linear.app/hack-ink/issue/XY-889/elf-benchmark-adapter-implement-graphify-docker-graph-report-adapter"
+        }
+      ]
+    },
+    {
+      "seq": 4,
+      "type": "tradeoffs_recorded",
+      "tradeoffs": [
+        {
+          "id": "T1",
+          "summary": "RAGFlow is resource-heavy, but the official Docker and reference chunk output make it an adapter candidate as long as the follow-up starts with a tiny corpus and records resource bounds instead of making a quality claim.",
+          "supporting_evidence_ids": [
+            "E2",
+            "E3"
+          ],
+          "disconfirming_evidence_ids": []
+        },
+        {
+          "id": "T2",
+          "summary": "LightRAG and GraphRAG can become adapter candidates because both expose bounded ingest/query paths and source mapping, but their first adapter issues must remain cost-bounded.",
+          "supporting_evidence_ids": [
+            "E4",
+            "E5",
+            "E6",
+            "E7"
+          ],
+          "disconfirming_evidence_ids": []
+        },
+        {
+          "id": "T3",
+          "summary": "Graphiti/Zep is a stronger adapter candidate than generic graph-memory references because it can emit temporal facts with validity windows and run against Docker-local graph stores.",
+          "supporting_evidence_ids": [
+            "E8"
+          ],
+          "disconfirming_evidence_ids": []
+        },
+        {
+          "id": "T4",
+          "summary": "Letta, LangGraph, nanograph, and llm-wiki should still inform ELF design, but creating adapter implementation issues now would blur reference workflows with executable memory-service evidence.",
+          "supporting_evidence_ids": [
+            "E9",
+            "E10",
+            "E11",
+            "E12"
+          ],
+          "disconfirming_evidence_ids": []
+        },
+        {
+          "id": "T5",
+          "summary": "gbrain has a good citation and current-truth/timeline contract, but the missing Docker-local brain repo/database setup keeps it blocked rather than adapter_candidate.",
+          "supporting_evidence_ids": [
+            "E13"
+          ],
+          "disconfirming_evidence_ids": []
+        },
+        {
+          "id": "T6",
+          "summary": "graphify is an adapter candidate only if implemented as an isolated CLI/materializer over generated corpus artifacts, not as a host-global assistant hook install.",
+          "supporting_evidence_ids": [
+            "E14"
+          ],
+          "disconfirming_evidence_ids": []
+        }
+      ]
+    },
+    {
+      "seq": 5,
+      "type": "challenge_recorded",
+      "summary": "The main risk is that adapter_candidate could be read as benchmark evidence. The mitigation is to keep evidence_class=research_gate, keep overall status non-pass, and state that follow-up implementation issues must still run Docker and real_world_job scoring before any live evidence claim.",
+      "resolved": true
+    },
+    {
+      "seq": 6,
+      "type": "finalized_decision_ready",
+      "confidence": "medium",
+      "decision": "Create implementation follow-ups only for RAGFlow, LightRAG, GraphRAG, Graphiti/Zep, and graphify. Keep Letta, LangGraph, nanograph, and llm-wiki as research_only references. Keep gbrain blocked pending a Docker-local brain repo/database proof. Do not change any research_gate record into live evidence until an adapter executes inside Docker and emits evidence-linked outputs.",
+      "missing_evidence": [
+        "No Docker adapter was implemented or executed in this lane.",
+        "No host-global install was used as proof.",
+        "Provider credentials and private corpora remain out of scope."
+      ]
+    }
+  ]
+}