{ "schema": "qorx.evaluation.v1", "generated_at": "2026-05-02T07:38:40+00:00", "suite": "qorx-self", "target_path": ".", "qorx_home": "target/qorx-benchmark-home-qorx-self", "git_commit": "78469bd", "qorx_version": "qorx 1.0.4", "budget_tokens": 600, "squeeze_budget_tokens": 450, "queries": [ "context fault proof pages resolver boundary", "qorx carriers .qorx .qorxb qorx handle", "strict answer refusal unsupported claims" ], "supported_question": "context fault proof pages resolver boundary", "unsupported_question": "galactic banana escrow treaty", "agent_objective": "prove context fault proof pages resolver boundary", "output_json": "docs/benchmarks/2026-05-02-qorx-self.json", "output_md": "docs/benchmarks/2026-05-02-qorx-self.md", "summary": { "indexed_tokens": 219838, "strict_tasks": 2, "strict_task_passes": 2, "strict_task_pass_rate": 1.0, "expected_refusals": 1, "expected_refusal_passes": 1, "expected_refusal_pass_rate": 1.0, "agent_provider_calls": 0, "boundary": "Local benchmark. Token counts are Qorx deterministic estimates, not provider billing records." }, "commands": { "version": { "command": [ "/target/release/qorx.exe", "--version" ], "exit_code": 0, "elapsed_ms": 19.007, "stderr": "", "stdout": "qorx 1.0.4" }, "index": { "command": [ "/target/release/qorx.exe", "index", "" ], "exit_code": 0, "elapsed_ms": 238.123, "stderr": "", "stdout": "Indexed 423 quarks from into /target/qorx-benchmark-home-qorx-self/repo_index.pb (219838 estimated tokens, 2349 symbols, 1254 signals, 31271 sparse vector terms)" } }, "session": { "command": [ "/target/release/qorx.exe", "session" ], "exit_code": 0, "elapsed_ms": 56.005, "stderr": "", "json": { "handle": "qorx://s/45c14e058b31e35b", "root": "", "root_fingerprint": "45c14e058b31e35b968eece04536e627d5496df968436bf87bb68fa9ace355fd", "updated_at": "2026-05-02T07:38:40.435930900+00:00", "quark_count": 423, "indexed_tokens": 219838, "visible_tokens": 73, "omitted_tokens": 219765, "context_reduction_x": 3011.4794520547944, "boundary": "Qorx session pointers are local resolver handles; exact context stays in the local Qorx cosmos and is resolved by the Qorx proxy. Cosmos is the Qorx name for local protobuf state, not a physics claim.", "prompt_block": "QORX_SESSION qorx://s/45c14e058b31e35b\nr=45c14e058b31e35b q=423 local_idx=219838\ncosmos=local redshift=ready; local_idx stays local; resolve with Qorx.\nproof at=2026-05-02T07:38:40Z ctx=219838t vis=73t omitted=219765t redshift=3011.48x est=char4 b2c=accounting cosmos=core redshift=core_b2c" } }, "bench": { "command": [ "/target/release/qorx.exe", "bench", "--budget-tokens", "600", "context fault proof pages resolver boundary", "qorx carriers .qorx .qorxb qorx handle", "strict answer refusal unsupported claims" ], "exit_code": 0, "elapsed_ms": 96.844, "stderr": "", "json": { "budget_tokens": 600, "indexed_tokens": 219838, "rows": [ { "query": "context fault proof pages resolver boundary", "used_tokens": 484, "omitted_tokens": 219354, "context_reduction_x": 454.2107438016529, "quarks_used": 3 }, { "query": "qorx carriers .qorx .qorxb qorx handle", "used_tokens": 450, "omitted_tokens": 219388, "context_reduction_x": 488.5288888888889, "quarks_used": 3 }, { "query": "strict answer refusal unsupported claims", "used_tokens": 532, "omitted_tokens": 219306, "context_reduction_x": 413.2293233082707, "quarks_used": 4 } ], "average_reduction_x": 451.9896519996042 } }, "pack": { "command": [ "/target/release/qorx.exe", "pack", "context fault proof pages resolver boundary", "--budget-tokens", "600" ], "exit_code": 0, "elapsed_ms": 74.409, "stderr": "", "json": { "query": "context fault proof pages resolver boundary", "budget_tokens": 600, "indexed_tokens": 219838, "used_tokens": 484, "omitted_tokens": 219354, "context_reduction_x": 454.2107438016529, "quarks_used": 3, "text": "# Qorx B2C packed context\nquery: context fault proof pages resolver boundary\nbudget_tokens: 600\nindexed_tokens: 219838\nb2c_route=pack\nb2c_parallel_lanes=retrieval,portfolio,risk,cache,carrier\nb2c_math=bounded_knapsack,portfolio_diversification,omission_risk_cap,stable_quark_reuse_value\n\n## id=qva_dbc9a6ae4d2b path=examples/benchmark-lab/tasks/q-lcr.qorx lines=1-4 tokens=28 b2c_net=55.08 retrieval=30 risk=0.50 cache=3.37 carrier=fault_if_needed\nQORX 1\n@mode strict-answer\n@ask what proves Qorx Local Context Resolution has a resolver boundary?\n@budget 700\n\n## id=qva_6099b2a2f968 path=src/context_vm.rs lines=881-902 tokens=176 symbols=mut,fault_status b2c_net=67.68 retrieval=57 risk=0.33 cache=5.18 carrier=fault_if_needed\nlet mut hasher = Sha256::new();\n hasher.update(handle.as_bytes());\n hasher.update(query.as_bytes());\n for page in pages {\n hasher.update(page.uri.as_bytes());\n hasher.update(page.content_hash.as_bytes());\n }\n format!(\"qvf_{}\", &format!(\"{:x}\", hasher.finalize())[..12])\n}\n\nfn fault_status(coverage: &str, missing_terms: &[String], pages: &[ProofPage]) -> String {\n if pages.is_empty() {\n return \"not_found\".to_string();\n }\n if coverage == \"supported\" && missing_terms.is_empty() {\n return \"resolved\".to_string();\n }\n if coverage == \"partial\" || !missing_terms.is_empty() {\n return \"partial\".to_string();\n }\n \"resolved\".to_string()\n}\n\n## id=qva_71d4ef7311b3 path=examples/benchmark-lab/src/resolver.rs lines=1-23 tokens=182 symbols=Carrier,ProofPage,resolve_carrier b2c_net=71.19 retrieval=97 risk=0.17 cache=5.21 carrier=pack\npub struct Carrier<'a> {\n pub mode: &'a str,\n pub objective: &'a str,\n pub budget_tokens: u64,\n}\n\npub struct ProofPage<'a> {\n pub citation: &'a str,\n pub excerpt: &'a str,\n}\n\n// Qorx Local Context Resolution keeps a resolver boundary between model-visible\n// carrier text and local evidence. The runtime resolves .qorx, .qorxb, and\n// qorx:// carriers into proof pages when local evidence supports the request.\npub fn resolve_carrier<'a>(carrier: &Carrier<'a>) -> Option> {\n if carrier.mode == \"strict-answer\" && carrier.budget_tokens >= 128 {\n return Some(ProofPage {\n citation: \"docs/runtime-boundary.md\",\n excerpt: carrier.objective,\n });\n }\n None\n}\n" } }, "squeeze": { "command": [ "/target/release/qorx.exe", "squeeze", "context fault proof pages resolver boundary", "--budget-tokens", "450", "--limit", "4" ], "exit_code": 0, "elapsed_ms": 70.084, "stderr": "", "json": { "schema": "qorx.squeeze.v1", "query": "context fault proof pages resolver boundary", "mode": "extractive_query_squeeze", "local_only": true, "provider_calls": 0, "budget_tokens": 450, "indexed_tokens": 219838, "source_tokens": 2308, "squeezed_tokens": 315, "used_tokens": 419, "omitted_tokens": 219419, "context_reduction_x": 524.6730310262529, "quarks_used": 3, "evidence": [ { "id": "qva_65bb29dee093", "path": "src/context_vm.rs", "start_line": 321, "end_line": 400, "source_tokens": 757, "excerpt_tokens": 113, "matched_terms": [ "boundary", "context", "fault", "pages", "proof", "resolver" ], "excerpt_hash": "f14f144ca5ff80a6", "excerpt": "proof_pages,\nboundary: \"Qorx Context VM is a local context-memory manager, not a Linux virtual machine. It keeps Qorx-known context in Cosmos, exposes scoped handles, resolves proof pages on demand, and reports fresh model-visible context avoided instead of claiming that tokens disappear.\".to_string(),\nlet status = fault_status(&strict.coverage, &strict.missing_terms, &proof_pages);\nlet fault_id = fault_id(&resolved_handle, query, &proof_pages);" }, { "id": "qva_560738bd6299", "path": "src/context_vm.rs", "start_line": 241, "end_line": 320, "source_tokens": 761, "excerpt_tokens": 58, "matched_terms": [ "boundary", "context", "fault", "pages", "proof" ], "excerpt_hash": "90449f24ef565241", "excerpt": "let fault = resolve_context_fault(index, &session.handle, objective, options);\nlet proof_pages = fault.proof_pages.clone();\n\"QORX_CONTEXT_VM version={CONTEXT_VM_VERSION} handle={} objective={} budget={} proofs={}\",\nproof_pages.len()" }, { "id": "qva_305349a55da9", "path": "src/context_vm.rs", "start_line": 401, "end_line": 480, "source_tokens": 790, "excerpt_tokens": 144, "matched_terms": [ "boundary", "context", "fault", "pages", "proof", "resolver" ], "excerpt_hash": "1fa03e40a5792068", "excerpt": "proof_pages,\nboundary: \"A context fault returns small cited proof pages from the local Qorx index. It cannot reveal data that was not indexed or authorized by the handle.\".to_string(),\nboundary: \"Context injection is a compact agent contract. It carries a handle and resolver path, not raw local files.\".to_string(),\nboundary: \"A Qorx nano carrier is a pointer to local Cosmos context. Cosmos means local Qorx state, not astrophysics. The carrier does not contain the local context; Qorx must expand or fault proof pages locally before any quality claim is made.\".to_string()," } ], "text": "# Qorx squeezed context\nquery: context fault proof pages resolver boundary\nbudget_tokens: 450\nindexed_tokens: 219838\n\n## src/context_vm.rs:321-400 id=qva_65bb29dee093 source_tokens=757 excerpt_tokens=113 score=310\nproof_pages,\nboundary: \"Qorx Context VM is a local context-memory manager, not a Linux virtual machine. It keeps Qorx-known context in Cosmos, exposes scoped handles, resolves proof pages on demand, and reports fresh model-visible context avoided instead of claiming that tokens disappear.\".to_string(),\nlet status = fault_status(&strict.coverage, &strict.missing_terms, &proof_pages);\nlet fault_id = fault_id(&resolved_handle, query, &proof_pages);\n\n## src/context_vm.rs:241-320 id=qva_560738bd6299 source_tokens=761 excerpt_tokens=58 score=276\nlet fault = resolve_context_fault(index, &session.handle, objective, options);\nlet proof_pages = fault.proof_pages.clone();\n\"QORX_CONTEXT_VM version={CONTEXT_VM_VERSION} handle={} objective={} budget={} proofs={}\",\nproof_pages.len()\n\n## src/context_vm.rs:401-480 id=qva_305349a55da9 source_tokens=790 excerpt_tokens=144 score=261\nproof_pages,\nboundary: \"A context fault returns small cited proof pages from the local Qorx index. It cannot reveal data that was not indexed or authorized by the handle.\".to_string(),\nboundary: \"Context injection is a compact agent contract. It carries a handle and resolver path, not raw local files.\".to_string(),\nboundary: \"A Qorx nano carrier is a pointer to local Cosmos context. Cosmos means local Qorx state, not astrophysics. The carrier does not contain the local context; Qorx must expand or fault proof pages locally before any quality claim is made.\".to_string(),\n", "boundary": "Squeeze is extractive and local: it keeps only query-relevant lines from indexed quarks, preserves citations and hashes, and uses no model calls or generated missing context." } }, "agent": { "command": [ "/target/release/qorx.exe", "agent", "prove context fault proof pages resolver boundary", "--budget-tokens", "600" ], "exit_code": 0, "elapsed_ms": 82.75, "stderr": "", "json": { "schema": "qorx.agent.v1", "agent_name": "Marvin", "objective": "prove context fault proof pages resolver boundary", "mode": "deterministic_subatomic", "local_only": true, "provider_calls": 0, "contract": { "hallucination_policy": "refuse_unsupported_indexed_context", "error_policy": "strict_extract_then_refuse", "compression_policy": "subatomic_context_budget", "b2c_policy": "account_then_claim", "boundary": "Marvin enforces strict indexed-context mode for local evidence and keeps the working set subatomic: small enough for the requested token budget. Subatomic is a Qorx context-size label, not a physics claim. Downstream model correctness, external tool behavior, and facts absent from the local index are outside this local planner contract." }, "steps": [ { "action": "session", "status": "completed", "reason": "emit tiny local session pointer without bulk context" }, { "action": "strict-answer", "status": "supported", "reason": "ground objective against indexed quarks only" }, { "action": "pack", "status": "completed", "reason": "return the smallest relevant working set under budget" } ], "session": { "handle": "qorx://s/45c14e058b31e35b", "root": "", "root_fingerprint": "45c14e058b31e35b968eece04536e627d5496df968436bf87bb68fa9ace355fd", "updated_at": "2026-05-02T07:38:40.435930900+00:00", "quark_count": 423, "indexed_tokens": 219838, "visible_tokens": 73, "omitted_tokens": 219765, "context_reduction_x": 3011.4794520547944, "boundary": "Qorx session pointers are local resolver handles; exact context stays in the local Qorx cosmos and is resolved by the Qorx proxy. Cosmos is the Qorx name for local protobuf state, not a physics claim.", "prompt_block": "QORX_SESSION qorx://s/45c14e058b31e35b\nr=45c14e058b31e35b q=423 local_idx=219838\ncosmos=local redshift=ready; local_idx stays local; resolve with Qorx.\nproof at=2026-05-02T07:38:40Z ctx=219838t vis=73t omitted=219765t redshift=3011.48x est=char4 b2c=accounting cosmos=core redshift=core_b2c" }, "strict_answer": { "schema": "qorx.strict-answer.v1", "question": "prove context fault proof pages resolver boundary", "coverage": "supported", "answer": "proof_pages,\nboundary: \"Qorx Context VM is a local context-memory manager, not a Linux virtual machine. It keeps Qorx-known context in Cosmos, exposes scoped handles, resolves proof pages on demand, and reports fresh model-visible context avoided instead of claiming that tokens disappear.\".to_string(),\nlet status = fault_status(&strict.coverage, &strict.missing_terms, &proof_pages);\nlet fault_id = fault_id(&resolved_handle, query, &proof_pages);\nlet fault = resolve_context_fault(index, &session.handle, objective, options);\nlet proof_pages = fault.proof_pages.clone();\n\"QORX_CONTEXT_VM version={CONTEXT_VM_VERSION} handle={} objective={} budget={} proofs={}\",\nproof_pages.len()", "evidence": [ { "id": "qva_65bb29dee093", "path": "src/context_vm.rs", "start_line": 321, "end_line": 400, "score": 310, "matched_terms": [ "boundary", "context", "fault", "pages", "proof", "resolver" ], "excerpt_hash": "f14f144ca5ff80a6", "excerpt": "proof_pages,\nboundary: \"Qorx Context VM is a local context-memory manager, not a Linux virtual machine. It keeps Qorx-known context in Cosmos, exposes scoped handles, resolves proof pages on demand, and reports fresh model-visible context avoided instead of claiming that tokens disappear.\".to_string(),\nlet status = fault_status(&strict.coverage, &strict.missing_terms, &proof_pages);\nlet fault_id = fault_id(&resolved_handle, query, &proof_pages);" }, { "id": "qva_560738bd6299", "path": "src/context_vm.rs", "start_line": 241, "end_line": 320, "score": 276, "matched_terms": [ "boundary", "context", "fault", "pages", "proof" ], "excerpt_hash": "90449f24ef565241", "excerpt": "let fault = resolve_context_fault(index, &session.handle, objective, options);\nlet proof_pages = fault.proof_pages.clone();\n\"QORX_CONTEXT_VM version={CONTEXT_VM_VERSION} handle={} objective={} budget={} proofs={}\",\nproof_pages.len()" } ], "supported_terms": [ "boundary", "context", "fault", "pages", "proof", "resolver" ], "missing_terms": [], "indexed_tokens": 219838, "used_tokens": 220, "omitted_tokens": 219618, "boundary": "Strict answers are extractive: every answer byte comes from indexed quark excerpts. Unsupported questions return not_found instead of model guesses." }, "packed_context": { "query": "prove context fault proof pages resolver boundary", "budget_tokens": 600, "indexed_tokens": 219838, "used_tokens": 486, "omitted_tokens": 219352, "context_reduction_x": 452.34156378600824, "quarks_used": 3, "text": "# Qorx B2C packed context\nquery: prove context fault proof pages resolver boundary\nbudget_tokens: 600\nindexed_tokens: 219838\nb2c_route=pack\nb2c_parallel_lanes=retrieval,portfolio,risk,cache,carrier\nb2c_math=bounded_knapsack,portfolio_diversification,omission_risk_cap,stable_quark_reuse_value\n\n## id=qva_dbc9a6ae4d2b path=examples/benchmark-lab/tasks/q-lcr.qorx lines=1-4 tokens=28 b2c_net=60.26 retrieval=33 risk=0.43 cache=3.37 carrier=fault_if_needed\nQORX 1\n@mode strict-answer\n@ask what proves Qorx Local Context Resolution has a resolver boundary?\n@budget 700\n\n## id=qva_6099b2a2f968 path=src/context_vm.rs lines=881-902 tokens=176 symbols=mut,fault_status b2c_net=62.63 retrieval=57 risk=0.43 cache=5.18 carrier=fault_if_needed\nlet mut hasher = Sha256::new();\n hasher.update(handle.as_bytes());\n hasher.update(query.as_bytes());\n for page in pages {\n hasher.update(page.uri.as_bytes());\n hasher.update(page.content_hash.as_bytes());\n }\n format!(\"qvf_{}\", &format!(\"{:x}\", hasher.finalize())[..12])\n}\n\nfn fault_status(coverage: &str, missing_terms: &[String], pages: &[ProofPage]) -> String {\n if pages.is_empty() {\n return \"not_found\".to_string();\n }\n if coverage == \"supported\" && missing_terms.is_empty() {\n return \"resolved\".to_string();\n }\n if coverage == \"partial\" || !missing_terms.is_empty() {\n return \"partial\".to_string();\n }\n \"resolved\".to_string()\n}\n\n## id=qva_71d4ef7311b3 path=examples/benchmark-lab/src/resolver.rs lines=1-23 tokens=182 symbols=Carrier,ProofPage,resolve_carrier b2c_net=66.01 retrieval=97 risk=0.29 cache=5.21 carrier=fault_if_needed\npub struct Carrier<'a> {\n pub mode: &'a str,\n pub objective: &'a str,\n pub budget_tokens: u64,\n}\n\npub struct ProofPage<'a> {\n pub citation: &'a str,\n pub excerpt: &'a str,\n}\n\n// Qorx Local Context Resolution keeps a resolver boundary between model-visible\n// carrier text and local evidence. The runtime resolves .qorx, .qorxb, and\n// qorx:// carriers into proof pages when local evidence supports the request.\npub fn resolve_carrier<'a>(carrier: &Carrier<'a>) -> Option> {\n if carrier.mode == \"strict-answer\" && carrier.budget_tokens >= 128 {\n return Some(ProofPage {\n citation: \"docs/runtime-boundary.md\",\n excerpt: carrier.objective,\n });\n }\n None\n}\n" }, "boundary": "Marvin is Qorx's deterministic local planner. The planner uses local indexed evidence, zero provider calls, and explicit refusal for unsupported indexed context." } }, "strict_answer_tasks": [ { "command": [ "/target/release/qorx.exe", "strict-answer", "context fault proof pages resolver boundary", "--limit", "3" ], "exit_code": 0, "elapsed_ms": 65.729, "stderr": "", "json": { "schema": "qorx.strict-answer.v1", "question": "context fault proof pages resolver boundary", "coverage": "supported", "answer": "proof_pages,\nboundary: \"Qorx Context VM is a local context-memory manager, not a Linux virtual machine. It keeps Qorx-known context in Cosmos, exposes scoped handles, resolves proof pages on demand, and reports fresh model-visible context avoided instead of claiming that tokens disappear.\".to_string(),\nlet status = fault_status(&strict.coverage, &strict.missing_terms, &proof_pages);\nlet fault_id = fault_id(&resolved_handle, query, &proof_pages);\nlet fault = resolve_context_fault(index, &session.handle, objective, options);\nlet proof_pages = fault.proof_pages.clone();\n\"QORX_CONTEXT_VM version={CONTEXT_VM_VERSION} handle={} objective={} budget={} proofs={}\",\nproof_pages.len()\nproof_pages,\nboundary: \"A context fault returns small cited proof pages from the local Qorx index. It cannot reveal data that was not indexed or authorized by the handle.\".to_string(),\nboundary: \"Context injection is a compact agent contract. It carries a handle and resolver path, not raw local files.\".to_string(),\nboundary: \"A Qorx nano carrier is a pointer to local Cosmos context. Cosmos means local Qorx state, not astrophysics. The carrier does not contain the local context; Qorx must expand or fault proof pages locally before any quality claim is made.\".to_string(),", "evidence": [ { "id": "qva_65bb29dee093", "path": "src/context_vm.rs", "start_line": 321, "end_line": 400, "score": 310, "matched_terms": [ "boundary", "context", "fault", "pages", "proof", "resolver" ], "excerpt_hash": "f14f144ca5ff80a6", "excerpt": "proof_pages,\nboundary: \"Qorx Context VM is a local context-memory manager, not a Linux virtual machine. It keeps Qorx-known context in Cosmos, exposes scoped handles, resolves proof pages on demand, and reports fresh model-visible context avoided instead of claiming that tokens disappear.\".to_string(),\nlet status = fault_status(&strict.coverage, &strict.missing_terms, &proof_pages);\nlet fault_id = fault_id(&resolved_handle, query, &proof_pages);" }, { "id": "qva_560738bd6299", "path": "src/context_vm.rs", "start_line": 241, "end_line": 320, "score": 276, "matched_terms": [ "boundary", "context", "fault", "pages", "proof" ], "excerpt_hash": "90449f24ef565241", "excerpt": "let fault = resolve_context_fault(index, &session.handle, objective, options);\nlet proof_pages = fault.proof_pages.clone();\n\"QORX_CONTEXT_VM version={CONTEXT_VM_VERSION} handle={} objective={} budget={} proofs={}\",\nproof_pages.len()" }, { "id": "qva_305349a55da9", "path": "src/context_vm.rs", "start_line": 401, "end_line": 480, "score": 261, "matched_terms": [ "boundary", "context", "fault", "pages", "proof" ], "excerpt_hash": "1fa03e40a5792068", "excerpt": "proof_pages,\nboundary: \"A context fault returns small cited proof pages from the local Qorx index. It cannot reveal data that was not indexed or authorized by the handle.\".to_string(),\nboundary: \"Context injection is a compact agent contract. It carries a handle and resolver path, not raw local files.\".to_string(),\nboundary: \"A Qorx nano carrier is a pointer to local Cosmos context. Cosmos means local Qorx state, not astrophysics. The carrier does not contain the local context; Qorx must expand or fault proof pages locally before any quality claim is made.\".to_string()," } ], "supported_terms": [ "boundary", "context", "fault", "pages", "proof", "resolver" ], "missing_terms": [], "indexed_tokens": 219838, "used_tokens": 380, "omitted_tokens": 219458, "boundary": "Strict answers are extractive: every answer byte comes from indexed quark excerpts. Unsupported questions return not_found instead of model guesses." }, "question": "context fault proof pages resolver boundary", "expected_coverage": "supported", "passed": true }, { "command": [ "/target/release/qorx.exe", "strict-answer", "galactic banana escrow treaty", "--limit", "3" ], "exit_code": 0, "elapsed_ms": 63.352, "stderr": "", "json": { "schema": "qorx.strict-answer.v1", "question": "galactic banana escrow treaty", "coverage": "not_found", "answer": "", "evidence": [], "supported_terms": [], "missing_terms": [ "banana", "escrow", "galactic", "treaty" ], "indexed_tokens": 219838, "used_tokens": 8, "omitted_tokens": 219830, "boundary": "Strict answers are extractive: every answer byte comes from indexed quark excerpts. Unsupported questions return not_found instead of model guesses." }, "question": "galactic banana escrow treaty", "expected_coverage": "not_found", "passed": true } ], "reproduce_command": "python scripts/run-benchmark.py --target . --suite qorx-self --budget-tokens 600 --squeeze-budget-tokens 450 --query \"context fault proof pages resolver boundary\" --query \"qorx carriers .qorx .qorxb qorx handle\" --query \"strict answer refusal unsupported claims\" --supported-question \"context fault proof pages resolver boundary\" --unsupported-question \"galactic banana escrow treaty\" --agent-objective \"prove context fault proof pages resolver boundary\" --output-json docs/benchmarks/2026-05-02-qorx-self.json --output-md docs/benchmarks/2026-05-02-qorx-self.md" }