{ "schema": "anvil-investigation.swift-host-cache-arm/v1", "arm": "expanded", "server_instance_id": "[operator identity redacted]", "request_accounting": { "preflight_observations": 25, "request_start_count": 43, "request_done_count": 43, "expected_total": 43, "unexpected_or_unmatched": false, "preflight_prompt_completion_multiset_matches": true, "crosspass_prompt_token_sequence_matches": true, "pairing": "IDs1..25 match preflight; IDs26..43 exact ordered prompt/completion counts match3x6 pilot, and all prompt counts/control fields match control." }, "engine": { "kv_capacity": 282112, "kv_cache": "k8v4", "context_cache": { "device_state_slots": 4, "enabled": true, "host_kv_capacity_bytes": 8589934592, "host_state_slots": 8, "max_long_anchors_per_continuation": 2, "max_private_continuations": 8, "max_shared_prefixes": 4, "total_device_state_slots": 8 }, "max_context": 262144, "max_concurrency": 4, "prefill_chunk": 4096, "speculative_backend": "dflash2", "speculative_draft_window": 7 }, "full_engine_equivalence_except_host_capacity_bundle": true, "passes": [ { "pass": "cold", "requests": [ { "request_index": 0, "native_request_id": 26, "prompt_tokens": 62327, "completion_tokens": 217, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62327, "prefix_reuse_path": "root", "visible_ttft_ms": 8828.89989999967, "e2e_ms": 8975.71170000083 }, { "request_index": 1, "native_request_id": 27, "prompt_tokens": 62328, "completion_tokens": 394, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62328, "prefix_reuse_path": "root", "visible_ttft_ms": 9190.386399997806, "e2e_ms": 9298.808199997438 }, { "request_index": 2, "native_request_id": 28, "prompt_tokens": 62328, "completion_tokens": 344, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62328, "prefix_reuse_path": "root", "visible_ttft_ms": 9079.026100000192, "e2e_ms": 9259.790000000066 }, { "request_index": 3, "native_request_id": 29, "prompt_tokens": 62330, "completion_tokens": 413, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62330, "prefix_reuse_path": "root", "visible_ttft_ms": 9589.440100000502, "e2e_ms": 9662.036300000182 }, { "request_index": 4, "native_request_id": 30, "prompt_tokens": 62333, "completion_tokens": 244, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62333, "prefix_reuse_path": "root", "visible_ttft_ms": 8781.176599997707, "e2e_ms": 8925.67009999766 }, { "request_index": 5, "native_request_id": 31, "prompt_tokens": 62335, "completion_tokens": 225, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62335, "prefix_reuse_path": "root", "visible_ttft_ms": 8540.200900002674, "e2e_ms": 8684.946299999865 } ], "reused_prompt_tokens": 0, "new_throughput_intervals": 14, "interval_counter_sums": { "actual_transfer_seconds": 0.21674060700000006, "backend_kv_transfers.d2d.bytes": 0, "backend_kv_transfers.d2d.pages": 0, "backend_kv_transfers.d2d.seconds": 0.0, "backend_kv_transfers.d2h.bytes": 0, "backend_kv_transfers.d2h.pages": 0, "backend_kv_transfers.d2h.seconds": 0.0, "backend_kv_transfers.h2d.bytes": 0, "backend_kv_transfers.h2d.pages": 0, "backend_kv_transfers.h2d.seconds": 0.0, "captures.aborted": 0, "captures.completed": 14, "main_kv_transfers.d2d.bytes": 14819328, "main_kv_transfers.d2d.pages": 9, "main_kv_transfers.d2d.seconds": 0.0032357120000000004, "main_kv_transfers.d2h.bytes": 2789326848, "main_kv_transfers.d2h.pages": 1694, "main_kv_transfers.d2h.seconds": 0.049124993000000006, "main_kv_transfers.h2d.bytes": 0, "main_kv_transfers.h2d.pages": 0, "main_kv_transfers.h2d.seconds": 0.0, "pressure.checkpoints_dropped": 13, "pressure.historical_fork_hits": 3, "pressure.maximal_fallback_selections": 0, "pressure.partial_tail_cow_pages": 9, "pressure.private_owners_degraded": 6, "pressure.private_owners_evicted": 7, "pressure.search_budget_exhaustions": 0, "pressure.searches": 28, "pressure.shared_owners_degraded": 4, "pressure.shared_owners_evicted": 2, "pressure.spill_pages": 1694, "selections.private_endpoint": 0, "selections.private_long_anchor": 0, "selections.private_response_replay": 16, "selections.private_turn_closure": 0, "selections.reused_prompt_tokens": 104817, "selections.root": 12, "selections.shared_stable_prefix": 3, "state_operations.forks": 8, "state_operations.moves": 0, "state_operations.restores": 16, "state_transfers.d2d.bytes": 5, "state_transfers.d2d.count": 5, "state_transfers.d2d.seconds": 0.000369248, "state_transfers.d2h.bytes": 16, "state_transfers.d2h.count": 16, "state_transfers.d2h.seconds": 0.108031102, "state_transfers.h2d.bytes": 16, "state_transfers.h2d.count": 16, "state_transfers.h2d.seconds": 0.055979552 }, "observed_occupancy_peaks": { "occupancy.device_backend_kv_pages": 0, "occupancy.device_main_kv_pages": 4406, "occupancy.device_state_slots": 8, "occupancy.host_kv_bytes": 2789326848, "occupancy.host_state_slots": 8, "occupancy.shared_active_references": 0 }, "end_occupancy": { "device_backend_kv_pages": 0, "device_main_kv_pages": 4282, "device_state_slots": 8, "host_kv_bytes": 2789326848, "host_state_slots": 8, "shared_active_references": 0 }, "median_visible_ttft_ms": 8953.96299999993, "nearest_rank_p50_visible_ttft_ms": 8828.89989999967, "median_e2e_ms": 9117.750850000448, "nearest_rank_p50_e2e_ms": 8975.71170000083 }, { "pass": "revisit1", "requests": [ { "request_index": 0, "native_request_id": 32, "prompt_tokens": 62327, "completion_tokens": 281, "prefix_cache_hit_tokens": 62320, "computed_prefill_tokens": 7, "prefix_reuse_path": "shared_stable_prefix", "visible_ttft_ms": 1418.298299999151, "e2e_ms": 1562.8556000010576 }, { "request_index": 1, "native_request_id": 33, "prompt_tokens": 62328, "completion_tokens": 519, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62328, "prefix_reuse_path": "root", "visible_ttft_ms": 10017.000400002871, "e2e_ms": 10089.705200000026 }, { "request_index": 2, "native_request_id": 34, "prompt_tokens": 62328, "completion_tokens": 274, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62328, "prefix_reuse_path": "root", "visible_ttft_ms": 8953.168000000005, "e2e_ms": 9133.986500000901 }, { "request_index": 3, "native_request_id": 35, "prompt_tokens": 62330, "completion_tokens": 321, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62330, "prefix_reuse_path": "root", "visible_ttft_ms": 9484.868100000313, "e2e_ms": 9593.297499999608 }, { "request_index": 4, "native_request_id": 36, "prompt_tokens": 62333, "completion_tokens": 433, "prefix_cache_hit_tokens": 62328, "computed_prefill_tokens": 5, "prefix_reuse_path": "private_response_replay", "visible_ttft_ms": 1752.577299997938, "e2e_ms": 1861.3442999994732 }, { "request_index": 5, "native_request_id": 37, "prompt_tokens": 62335, "completion_tokens": 455, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62335, "prefix_reuse_path": "root", "visible_ttft_ms": 9974.789100000635, "e2e_ms": 10137.64810000066 } ], "reused_prompt_tokens": 124648, "new_throughput_intervals": 9, "interval_counter_sums": { "actual_transfer_seconds": 0.31810087000000004, "backend_kv_transfers.d2d.bytes": 0, "backend_kv_transfers.d2d.pages": 0, "backend_kv_transfers.d2d.seconds": 0.0, "backend_kv_transfers.d2h.bytes": 0, "backend_kv_transfers.d2h.pages": 0, "backend_kv_transfers.d2h.seconds": 0.0, "backend_kv_transfers.h2d.bytes": 0, "backend_kv_transfers.h2d.pages": 0, "backend_kv_transfers.h2d.seconds": 0.0, "captures.aborted": 0, "captures.completed": 8, "main_kv_transfers.d2d.bytes": 8232960, "main_kv_transfers.d2d.pages": 5, "main_kv_transfers.d2d.seconds": 0.0018738880000000002, "main_kv_transfers.d2h.bytes": 3080773632, "main_kv_transfers.d2h.pages": 1871, "main_kv_transfers.d2h.seconds": 0.25551827800000004, "main_kv_transfers.h2d.bytes": 1603780608, "main_kv_transfers.h2d.pages": 974, "main_kv_transfers.h2d.seconds": 0.028522272, "pressure.checkpoints_dropped": 9, "pressure.historical_fork_hits": 0, "pressure.maximal_fallback_selections": 0, "pressure.partial_tail_cow_pages": 5, "pressure.private_owners_degraded": 1, "pressure.private_owners_evicted": 5, "pressure.search_budget_exhaustions": 0, "pressure.searches": 6, "pressure.shared_owners_degraded": 2, "pressure.shared_owners_evicted": 0, "pressure.spill_pages": 1871, "selections.private_endpoint": 0, "selections.private_long_anchor": 0, "selections.private_response_replay": 1, "selections.private_turn_closure": 0, "selections.reused_prompt_tokens": 124648, "selections.root": 4, "selections.shared_stable_prefix": 1, "state_operations.forks": 1, "state_operations.moves": 0, "state_operations.restores": 2, "state_transfers.d2d.bytes": 1, "state_transfers.d2d.count": 1, "state_transfers.d2d.seconds": 0.00010604800000000003, "state_transfers.d2h.bytes": 7, "state_transfers.d2h.count": 7, "state_transfers.d2h.seconds": 0.025059871999999997, "state_transfers.h2d.bytes": 2, "state_transfers.h2d.count": 2, "state_transfers.h2d.seconds": 0.007020511999999993 }, "observed_occupancy_peaks": { "occupancy.device_backend_kv_pages": 0, "occupancy.device_main_kv_pages": 4407, "occupancy.device_state_slots": 8, "occupancy.host_kv_bytes": 5870100480, "occupancy.host_state_slots": 8, "occupancy.shared_active_references": 0 }, "end_occupancy": { "device_backend_kv_pages": 0, "device_main_kv_pages": 4283, "device_state_slots": 8, "host_kv_bytes": 5868453888, "host_state_slots": 8, "shared_active_references": 0 }, "median_visible_ttft_ms": 9219.018050000159, "nearest_rank_p50_visible_ttft_ms": 8953.168000000005, "median_e2e_ms": 9363.642000000254, "nearest_rank_p50_e2e_ms": 9133.986500000901 }, { "pass": "revisit2", "requests": [ { "request_index": 0, "native_request_id": 38, "prompt_tokens": 62327, "completion_tokens": 317, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62327, "prefix_reuse_path": "root", "visible_ttft_ms": 9509.358399998746, "e2e_ms": 9618.114900000364 }, { "request_index": 1, "native_request_id": 39, "prompt_tokens": 62328, "completion_tokens": 254, "prefix_cache_hit_tokens": 62323, "computed_prefill_tokens": 5, "prefix_reuse_path": "private_response_replay", "visible_ttft_ms": 1049.1322000016226, "e2e_ms": 1211.8689000017184 }, { "request_index": 2, "native_request_id": 40, "prompt_tokens": 62328, "completion_tokens": 421, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62328, "prefix_reuse_path": "root", "visible_ttft_ms": 9787.702499997977, "e2e_ms": 9968.378499997925 }, { "request_index": 3, "native_request_id": 41, "prompt_tokens": 62330, "completion_tokens": 250, "prefix_cache_hit_tokens": 62325, "computed_prefill_tokens": 5, "prefix_reuse_path": "private_response_replay", "visible_ttft_ms": 1079.828300000372, "e2e_ms": 1260.367300001235 }, { "request_index": 4, "native_request_id": 42, "prompt_tokens": 62333, "completion_tokens": 341, "prefix_cache_hit_tokens": 62328, "computed_prefill_tokens": 5, "prefix_reuse_path": "private_response_replay", "visible_ttft_ms": 1530.1660999975866, "e2e_ms": 1656.939099997544 }, { "request_index": 5, "native_request_id": 43, "prompt_tokens": 62335, "completion_tokens": 383, "prefix_cache_hit_tokens": 62330, "computed_prefill_tokens": 5, "prefix_reuse_path": "private_response_replay", "visible_ttft_ms": 1533.4326999982295, "e2e_ms": 1678.4001999985776 } ], "reused_prompt_tokens": 249306, "new_throughput_intervals": 6, "interval_counter_sums": { "actual_transfer_seconds": 0.4016291740000001, "backend_kv_transfers.d2d.bytes": 0, "backend_kv_transfers.d2d.pages": 0, "backend_kv_transfers.d2d.seconds": 0.0, "backend_kv_transfers.d2h.bytes": 0, "backend_kv_transfers.d2h.pages": 0, "backend_kv_transfers.d2h.seconds": 0.0, "backend_kv_transfers.h2d.bytes": 0, "backend_kv_transfers.h2d.pages": 0, "backend_kv_transfers.h2d.seconds": 0.0, "captures.aborted": 0, "captures.completed": 3, "main_kv_transfers.d2d.bytes": 3293184, "main_kv_transfers.d2d.pages": 2, "main_kv_transfers.d2d.seconds": 0.000753376, "main_kv_transfers.d2h.bytes": 3079127040, "main_kv_transfers.d2h.pages": 1870, "main_kv_transfers.d2h.seconds": 0.32479560600000007, "main_kv_transfers.h2d.bytes": 3070894080, "main_kv_transfers.h2d.pages": 1865, "main_kv_transfers.h2d.seconds": 0.054784544000000004, "pressure.checkpoints_dropped": 4, "pressure.historical_fork_hits": 0, "pressure.maximal_fallback_selections": 0, "pressure.partial_tail_cow_pages": 2, "pressure.private_owners_degraded": 1, "pressure.private_owners_evicted": 2, "pressure.search_budget_exhaustions": 0, "pressure.searches": 6, "pressure.shared_owners_degraded": 3, "pressure.shared_owners_evicted": 1, "pressure.spill_pages": 1870, "selections.private_endpoint": 0, "selections.private_long_anchor": 0, "selections.private_response_replay": 4, "selections.private_turn_closure": 0, "selections.reused_prompt_tokens": 249306, "selections.root": 2, "selections.shared_stable_prefix": 0, "state_operations.forks": 1, "state_operations.moves": 0, "state_operations.restores": 4, "state_transfers.d2d.bytes": 1, "state_transfers.d2d.count": 1, "state_transfers.d2d.seconds": 9.027199999999997e-05, "state_transfers.d2h.bytes": 2, "state_transfers.d2h.count": 2, "state_transfers.d2h.seconds": 0.007133312000000003, "state_transfers.h2d.bytes": 4, "state_transfers.h2d.count": 4, "state_transfers.h2d.seconds": 0.014072063999999995 }, "observed_occupancy_peaks": { "occupancy.device_backend_kv_pages": 0, "occupancy.device_main_kv_pages": 4403, "occupancy.device_state_slots": 8, "occupancy.host_kv_bytes": 7477174272, "occupancy.host_state_slots": 8, "occupancy.shared_active_references": 0 }, "end_occupancy": { "device_backend_kv_pages": 0, "device_main_kv_pages": 4286, "device_state_slots": 8, "host_kv_bytes": 7468941312, "host_state_slots": 7, "shared_active_references": 0 }, "median_visible_ttft_ms": 1531.799399997908, "nearest_rank_p50_visible_ttft_ms": 1530.1660999975866, "median_e2e_ms": 1667.6696499980608, "nearest_rank_p50_e2e_ms": 1656.939099997544 } ], "revisit_population": { "requests": 12, "reused_prompt_tokens": 373954, "median_visible_ttft_ms": 5352.872649998972, "nearest_rank_p50_visible_ttft_ms": 1752.577299997938, "median_e2e_ms": 5497.665400000187, "nearest_rank_p50_e2e_ms": 1861.3442999994732 }, "host_memory": { "current_bytes": 32072245248, "effective_limit_bytes": 42949672960, "effective_swap_limit_bytes": 0, "events": { "high": 0, "low": 0, "max": 0, "oom": 0, "oom_group_kill": 0, "oom_kill": 0 }, "exit_code": 0, "limit_bytes": 42949672960, "oom_killed": false, "peak_bytes": 32072785920, "ram_plus_swap_limit_bytes": 42949672960 }, "limits": [ "First pass interval sums include preflight intervals; later passes sum only newly captured native interval records. Occupancy is a sampled gauge, not an exact continuous peak.", "Throughput interval attribution assumes root-reported quiet>=6s capture boundaries; native IDs prove no intervening inference.", "All normal records are private until sanitized; native errors/startup can contain identifying free text.", "Observed state_transfers.*.bytes numerically equal operation counts in control logs. Preserve raw field names but do not use these as physical state-byte volumes without producer-unit verification; main_kv_transfers bytes/pages are separately reported." ], "median_convention": "Statistical median averages the two middle observations for even N; native nearest-rank p50 is also retained. Gate evaluation must report both, never choose the favorable one.", "sources": { "swift-cache-expanded-cold.json": "094a85b938acb0ce5142d4ff59a0bec905fa873b91517fc6f12e6be290231e89", "swift-cache-expanded-revisit1.json": "0977858b36654e1a962edc71f67435b3771c3f564ded9edd6f792c775f1991ac", "swift-cache-expanded-revisit2.json": "8f46440d5270bb48a2a7a56d504a21852e3755e95e3cde8852811936eb5f0e08", "swift-cache-expanded-cold-native.json": "a5d66a9e49f0d9d69064d3cde5422b8498d32eb5dcb58b6398fc799fdf3790e8", "swift-cache-expanded-revisit1-native.json": "0dbe2ca4a763be1825e4339c10c667d7d81e834b0996b92a9cea07fb213da986", "swift-cache-expanded-revisit2-native.json": "01e8f5c53e11221cb1e96c4ced8bd057b26ca42f026cf2b724cea138b7199210", "swift-cache-expanded-preflight.json": "8fae706599c26d9ed5b1cf29cf11cab21fe51c32f8cb4e44aa9487afaf3a6b0f", "swift-cache-expanded-pilot-final-status.json": "45105fed6a6365a684e17804ab4a483e556d8a750ba4edcde7da620b6693d5d6" } }