diff --git a/dirigo-results/0.2.13-port/intercept/conf-base.json b/dirigo-results/0.2.13-port/intercept/conf-base.json
new file mode 100644
index 00000000000..522f818e182
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/conf-base.json
@@ -0,0 +1,63 @@
+{
+ "exe": "/home/bob/bin/strata-sycl-host.sh",
+ "args": [
+  "--pack",
+  "/home/bob/XeStrata-data/packs/coder-iq1_m",
+  "--native",
+  "/home/bob/strata-models/IQ1_M/Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00001-of-00002.gguf",
+  "--ple-gguf",
+  "/home/bob/strata-models/IQ1_M/Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00002-of-00002.gguf",
+  "--expert-profile",
+  "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/data/expert-profile-mine.bin",
+  "--expert-cache",
+  "auto",
+  "--stream-experts",
+  "--prefill",
+  "auto",
+  "--prompt-cache",
+  "6",
+  "--prompt-cache-every",
+  "4096",
+  "--batch",
+  "0",
+  "--spec",
+  "6",
+  "--spec-min-p",
+  "0.5",
+  "--mtp",
+  "/home/bob/XeStrata-data/mtp/rt",
+  "--mtp-draft-vocab",
+  "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/data/draft_vocab_en.bin",
+  "--mtp-window",
+  "8192",
+  "--max-context",
+  "131072",
+  "--kv",
+  "int8",
+  "--kv-resident",
+  "32768",
+  "--vram-reserve-mib",
+  "700",
+  "--conversation-cache-mib",
+  "2048"
+ ],
+ "cwd": "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13",
+ "tokenizer": "/home/bob/XeStrata-data/packs/coder-iq1_m/tokenizer",
+ "model_name": "qwen3.8-flash-next-coder-iq1_m",
+ "log": "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/strata-coder-iq1_m-sycl-tuned.log",
+ "port": 8095,
+ "backend": "sycl",
+ "open_browser": false,
+ "thinking_policy": "on",
+ "conv_route": "on",
+ "conv_route_batch_after": 4,
+ "short_answer": {
+  "spec": 0,
+  "max_tokens": 16
+ },
+ "spec_window": 0,
+ "spec_window_needle_min": 6000,
+ "code_tune": {
+  "mode": "off"
+ }
+}
\ No newline at end of file
diff --git a/dirigo-results/0.2.13-port/intercept/conf-base2.json b/dirigo-results/0.2.13-port/intercept/conf-base2.json
new file mode 100644
index 00000000000..522f818e182
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/conf-base2.json
@@ -0,0 +1,63 @@
+{
+ "exe": "/home/bob/bin/strata-sycl-host.sh",
+ "args": [
+  "--pack",
+  "/home/bob/XeStrata-data/packs/coder-iq1_m",
+  "--native",
+  "/home/bob/strata-models/IQ1_M/Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00001-of-00002.gguf",
+  "--ple-gguf",
+  "/home/bob/strata-models/IQ1_M/Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00002-of-00002.gguf",
+  "--expert-profile",
+  "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/data/expert-profile-mine.bin",
+  "--expert-cache",
+  "auto",
+  "--stream-experts",
+  "--prefill",
+  "auto",
+  "--prompt-cache",
+  "6",
+  "--prompt-cache-every",
+  "4096",
+  "--batch",
+  "0",
+  "--spec",
+  "6",
+  "--spec-min-p",
+  "0.5",
+  "--mtp",
+  "/home/bob/XeStrata-data/mtp/rt",
+  "--mtp-draft-vocab",
+  "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/data/draft_vocab_en.bin",
+  "--mtp-window",
+  "8192",
+  "--max-context",
+  "131072",
+  "--kv",
+  "int8",
+  "--kv-resident",
+  "32768",
+  "--vram-reserve-mib",
+  "700",
+  "--conversation-cache-mib",
+  "2048"
+ ],
+ "cwd": "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13",
+ "tokenizer": "/home/bob/XeStrata-data/packs/coder-iq1_m/tokenizer",
+ "model_name": "qwen3.8-flash-next-coder-iq1_m",
+ "log": "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/strata-coder-iq1_m-sycl-tuned.log",
+ "port": 8095,
+ "backend": "sycl",
+ "open_browser": false,
+ "thinking_policy": "on",
+ "conv_route": "on",
+ "conv_route_batch_after": 4,
+ "short_answer": {
+  "spec": 0,
+  "max_tokens": 16
+ },
+ "spec_window": 0,
+ "spec_window_needle_min": 6000,
+ "code_tune": {
+  "mode": "off"
+ }
+}
\ No newline at end of file
diff --git a/dirigo-results/0.2.13-port/intercept/conf-nocache.json b/dirigo-results/0.2.13-port/intercept/conf-nocache.json
new file mode 100644
index 00000000000..e65b1f9139f
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/conf-nocache.json
@@ -0,0 +1,65 @@
+{
+ "exe": "/home/bob/bin/strata-sycl-host.sh",
+ "args": [
+  "--pack",
+  "/home/bob/XeStrata-data/packs/coder-iq1_m",
+  "--native",
+  "/home/bob/strata-models/IQ1_M/Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00001-of-00002.gguf",
+  "--ple-gguf",
+  "/home/bob/strata-models/IQ1_M/Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00002-of-00002.gguf",
+  "--expert-profile",
+  "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/data/expert-profile-mine.bin",
+  "--expert-cache",
+  "auto",
+  "--stream-experts",
+  "--prefill",
+  "auto",
+  "--prompt-cache",
+  "6",
+  "--prompt-cache-every",
+  "4096",
+  "--batch",
+  "0",
+  "--spec",
+  "6",
+  "--spec-min-p",
+  "0.5",
+  "--mtp",
+  "/home/bob/XeStrata-data/mtp/rt",
+  "--mtp-draft-vocab",
+  "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/data/draft_vocab_en.bin",
+  "--mtp-window",
+  "8192",
+  "--max-context",
+  "131072",
+  "--kv",
+  "int8",
+  "--kv-resident",
+  "32768",
+  "--vram-reserve-mib",
+  "700",
+  "--conversation-cache-mib",
+  "2048",
+  "--prompt-cache",
+  "0"
+ ],
+ "cwd": "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13",
+ "tokenizer": "/home/bob/XeStrata-data/packs/coder-iq1_m/tokenizer",
+ "model_name": "qwen3.8-flash-next-coder-iq1_m",
+ "log": "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/strata-coder-iq1_m-sycl-tuned.log",
+ "port": 8095,
+ "backend": "sycl",
+ "open_browser": false,
+ "thinking_policy": "on",
+ "conv_route": "on",
+ "conv_route_batch_after": 4,
+ "short_answer": {
+  "spec": 0,
+  "max_tokens": 16
+ },
+ "spec_window": 0,
+ "spec_window_needle_min": 6000,
+ "code_tune": {
+  "mode": "off"
+ }
+}
\ No newline at end of file
diff --git a/dirigo-results/0.2.13-port/intercept/conf-nospec.json b/dirigo-results/0.2.13-port/intercept/conf-nospec.json
new file mode 100644
index 00000000000..1021cd90dfe
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/conf-nospec.json
@@ -0,0 +1,65 @@
+{
+ "exe": "/home/bob/bin/strata-sycl-host.sh",
+ "args": [
+  "--pack",
+  "/home/bob/XeStrata-data/packs/coder-iq1_m",
+  "--native",
+  "/home/bob/strata-models/IQ1_M/Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00001-of-00002.gguf",
+  "--ple-gguf",
+  "/home/bob/strata-models/IQ1_M/Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00002-of-00002.gguf",
+  "--expert-profile",
+  "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/data/expert-profile-mine.bin",
+  "--expert-cache",
+  "auto",
+  "--stream-experts",
+  "--prefill",
+  "auto",
+  "--prompt-cache",
+  "6",
+  "--prompt-cache-every",
+  "4096",
+  "--batch",
+  "0",
+  "--spec",
+  "6",
+  "--spec-min-p",
+  "0.5",
+  "--mtp",
+  "/home/bob/XeStrata-data/mtp/rt",
+  "--mtp-draft-vocab",
+  "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/data/draft_vocab_en.bin",
+  "--mtp-window",
+  "8192",
+  "--max-context",
+  "131072",
+  "--kv",
+  "int8",
+  "--kv-resident",
+  "32768",
+  "--vram-reserve-mib",
+  "700",
+  "--conversation-cache-mib",
+  "2048",
+  "--spec",
+  "0"
+ ],
+ "cwd": "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13",
+ "tokenizer": "/home/bob/XeStrata-data/packs/coder-iq1_m/tokenizer",
+ "model_name": "qwen3.8-flash-next-coder-iq1_m",
+ "log": "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/strata-coder-iq1_m-sycl-tuned.log",
+ "port": 8095,
+ "backend": "sycl",
+ "open_browser": false,
+ "thinking_policy": "on",
+ "conv_route": "on",
+ "conv_route_batch_after": 4,
+ "short_answer": {
+  "spec": 0,
+  "max_tokens": 16
+ },
+ "spec_window": 0,
+ "spec_window_needle_min": 6000,
+ "code_tune": {
+  "mode": "off"
+ }
+}
\ No newline at end of file
diff --git a/dirigo-results/0.2.13-port/intercept/conf-on.json b/dirigo-results/0.2.13-port/intercept/conf-on.json
new file mode 100644
index 00000000000..522f818e182
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/conf-on.json
@@ -0,0 +1,63 @@
+{
+ "exe": "/home/bob/bin/strata-sycl-host.sh",
+ "args": [
+  "--pack",
+  "/home/bob/XeStrata-data/packs/coder-iq1_m",
+  "--native",
+  "/home/bob/strata-models/IQ1_M/Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00001-of-00002.gguf",
+  "--ple-gguf",
+  "/home/bob/strata-models/IQ1_M/Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00002-of-00002.gguf",
+  "--expert-profile",
+  "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/data/expert-profile-mine.bin",
+  "--expert-cache",
+  "auto",
+  "--stream-experts",
+  "--prefill",
+  "auto",
+  "--prompt-cache",
+  "6",
+  "--prompt-cache-every",
+  "4096",
+  "--batch",
+  "0",
+  "--spec",
+  "6",
+  "--spec-min-p",
+  "0.5",
+  "--mtp",
+  "/home/bob/XeStrata-data/mtp/rt",
+  "--mtp-draft-vocab",
+  "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/data/draft_vocab_en.bin",
+  "--mtp-window",
+  "8192",
+  "--max-context",
+  "131072",
+  "--kv",
+  "int8",
+  "--kv-resident",
+  "32768",
+  "--vram-reserve-mib",
+  "700",
+  "--conversation-cache-mib",
+  "2048"
+ ],
+ "cwd": "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13",
+ "tokenizer": "/home/bob/XeStrata-data/packs/coder-iq1_m/tokenizer",
+ "model_name": "qwen3.8-flash-next-coder-iq1_m",
+ "log": "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/strata-coder-iq1_m-sycl-tuned.log",
+ "port": 8095,
+ "backend": "sycl",
+ "open_browser": false,
+ "thinking_policy": "on",
+ "conv_route": "on",
+ "conv_route_batch_after": 4,
+ "short_answer": {
+  "spec": 0,
+  "max_tokens": 16
+ },
+ "spec_window": 0,
+ "spec_window_needle_min": 6000,
+ "code_tune": {
+  "mode": "off"
+ }
+}
\ No newline at end of file
diff --git a/dirigo-results/0.2.13-port/intercept/conf-shadow.json b/dirigo-results/0.2.13-port/intercept/conf-shadow.json
new file mode 100644
index 00000000000..7bd3fc5006f
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/conf-shadow.json
@@ -0,0 +1,63 @@
+{
+ "exe": "/home/bob/bin/strata-sycl-host.sh",
+ "args": [
+  "--pack",
+  "/home/bob/XeStrata-data/packs/coder-iq1_m",
+  "--native",
+  "/home/bob/strata-models/IQ1_M/Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00001-of-00002.gguf",
+  "--ple-gguf",
+  "/home/bob/strata-models/IQ1_M/Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00002-of-00002.gguf",
+  "--expert-profile",
+  "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/data/expert-profile-mine.bin",
+  "--expert-cache",
+  "auto",
+  "--stream-experts",
+  "--prefill",
+  "auto",
+  "--prompt-cache",
+  "6",
+  "--prompt-cache-every",
+  "4096",
+  "--batch",
+  "0",
+  "--spec",
+  "6",
+  "--spec-min-p",
+  "0.5",
+  "--mtp",
+  "/home/bob/XeStrata-data/mtp/rt",
+  "--mtp-draft-vocab",
+  "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/data/draft_vocab_en.bin",
+  "--mtp-window",
+  "8192",
+  "--max-context",
+  "131072",
+  "--kv",
+  "int8",
+  "--kv-resident",
+  "32768",
+  "--vram-reserve-mib",
+  "700",
+  "--conversation-cache-mib",
+  "2048"
+ ],
+ "cwd": "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13",
+ "tokenizer": "/home/bob/XeStrata-data/packs/coder-iq1_m/tokenizer",
+ "model_name": "qwen3.8-flash-next-coder-iq1_m",
+ "log": "/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731/strata-0.2.13/strata-coder-iq1_m-sycl-tuned.log",
+ "port": 8095,
+ "backend": "sycl",
+ "open_browser": false,
+ "thinking_policy": "on",
+ "conv_route": "shadow",
+ "conv_route_batch_after": 4,
+ "short_answer": {
+  "spec": 0,
+  "max_tokens": 16
+ },
+ "spec_window": 0,
+ "spec_window_needle_min": 6000,
+ "code_tune": {
+  "mode": "off"
+ }
+}
\ No newline at end of file
diff --git a/dirigo-results/0.2.13-port/intercept/probe-base.json b/dirigo-results/0.2.13-port/intercept/probe-base.json
new file mode 100644
index 00000000000..d7de38aded1
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/probe-base.json
@@ -0,0 +1,96 @@
+[
+ {
+  "target": 75,
+  "ttft_ms": 558.1,
+  "wall_ms": 626.8,
+  "prompt_tokens": 126,
+  "reused": 0,
+  "fresh": 126,
+  "prefill_ms": 537,
+  "prefill_tps": 234.7,
+  "decode_ms": 87,
+  "park_tokens": 96,
+  "park_ms": 100.4,
+  "kv_hit_pct": 93.1,
+  "kv_mib_from_ram": 1.2,
+  "other_ms": -79.3,
+  "label": "base"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 546.6,
+  "wall_ms": 605.7,
+  "prompt_tokens": 126,
+  "reused": 0,
+  "fresh": 126,
+  "prefill_ms": 525,
+  "prefill_tps": 239.8,
+  "decode_ms": 78,
+  "park_tokens": 133,
+  "park_ms": 101.9,
+  "kv_hit_pct": 91.21,
+  "kv_mib_from_ram": 1.6,
+  "other_ms": -80.3,
+  "label": "base"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 456.1,
+  "wall_ms": 524.3,
+  "prompt_tokens": 126,
+  "reused": 0,
+  "fresh": 126,
+  "prefill_ms": 435,
+  "prefill_tps": 289.9,
+  "decode_ms": 87,
+  "park_tokens": 133,
+  "park_ms": 101.6,
+  "other_ms": -80.5,
+  "label": "base"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 349.8,
+  "wall_ms": 465.3,
+  "prompt_tokens": 126,
+  "reused": 0,
+  "fresh": 126,
+  "prefill_ms": 328,
+  "prefill_tps": 383.7,
+  "decode_ms": 134,
+  "kv_hit_pct": 90.58,
+  "kv_mib_from_ram": 1.6,
+  "other_ms": 21.8,
+  "label": "base"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 348.2,
+  "wall_ms": 409.3,
+  "prompt_tokens": 126,
+  "reused": 0,
+  "fresh": 126,
+  "prefill_ms": 327,
+  "prefill_tps": 385.6,
+  "decode_ms": 80,
+  "kv_hit_pct": 93.0,
+  "kv_mib_from_ram": 1.7,
+  "other_ms": 21.2,
+  "label": "base"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 349.1,
+  "wall_ms": 447.9,
+  "prompt_tokens": 126,
+  "reused": 0,
+  "fresh": 126,
+  "prefill_ms": 328,
+  "prefill_tps": 384.5,
+  "decode_ms": 117,
+  "kv_hit_pct": 88.51,
+  "kv_mib_from_ram": 1.6,
+  "other_ms": 21.1,
+  "label": "base"
+ }
+]
\ No newline at end of file
diff --git a/dirigo-results/0.2.13-port/intercept/probe-base2.json b/dirigo-results/0.2.13-port/intercept/probe-base2.json
new file mode 100644
index 00000000000..54479683209
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/probe-base2.json
@@ -0,0 +1,90 @@
+[
+ {
+  "target": 75,
+  "ttft_ms": 557.8,
+  "wall_ms": 616.7,
+  "prompt_tokens": 127,
+  "reused": 0,
+  "fresh": 127,
+  "prefill_ms": 537,
+  "prefill_tps": 236.7,
+  "decode_ms": 78,
+  "park_tokens": 96,
+  "park_ms": 99.4,
+  "kv_hit_pct": 93.1,
+  "kv_mib_from_ram": 1.2,
+  "other_ms": -78.6,
+  "label": "base2"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 540.6,
+  "wall_ms": 599.3,
+  "prompt_tokens": 127,
+  "reused": 0,
+  "fresh": 127,
+  "prefill_ms": 520,
+  "prefill_tps": 244.4,
+  "decode_ms": 78,
+  "park_tokens": 134,
+  "park_ms": 97.5,
+  "other_ms": -76.9,
+  "label": "base2"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 460.8,
+  "wall_ms": 518.8,
+  "prompt_tokens": 127,
+  "reused": 0,
+  "fresh": 127,
+  "prefill_ms": 439,
+  "prefill_tps": 289.0,
+  "decode_ms": 77,
+  "park_tokens": 134,
+  "park_ms": 108.3,
+  "other_ms": -86.5,
+  "label": "base2"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 346.7,
+  "wall_ms": 423.9,
+  "prompt_tokens": 127,
+  "reused": 0,
+  "fresh": 127,
+  "prefill_ms": 325,
+  "prefill_tps": 390.5,
+  "decode_ms": 96,
+  "other_ms": 21.7,
+  "label": "base2"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 346.5,
+  "wall_ms": 422.9,
+  "prompt_tokens": 127,
+  "reused": 0,
+  "fresh": 127,
+  "prefill_ms": 325,
+  "prefill_tps": 390.5,
+  "decode_ms": 95,
+  "kv_hit_pct": 91.94,
+  "kv_mib_from_ram": 1.7,
+  "other_ms": 21.5,
+  "label": "base2"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 347.0,
+  "wall_ms": 423.0,
+  "prompt_tokens": 127,
+  "reused": 0,
+  "fresh": 127,
+  "prefill_ms": 325,
+  "prefill_tps": 390.4,
+  "decode_ms": 95,
+  "other_ms": 22.0,
+  "label": "base2"
+ }
+]
\ No newline at end of file
diff --git a/dirigo-results/0.2.13-port/intercept/probe-nocache.json b/dirigo-results/0.2.13-port/intercept/probe-nocache.json
new file mode 100644
index 00000000000..7761934c7b1
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/probe-nocache.json
@@ -0,0 +1,88 @@
+[
+ {
+  "target": 75,
+  "ttft_ms": 360.7,
+  "wall_ms": 439.2,
+  "prompt_tokens": 127,
+  "reused": 0,
+  "fresh": 127,
+  "prefill_ms": 339,
+  "prefill_tps": 374.5,
+  "decode_ms": 97,
+  "kv_hit_pct": 90.77,
+  "kv_mib_from_ram": 1.2,
+  "other_ms": 21.7,
+  "label": "nocache"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 345.0,
+  "wall_ms": 413.4,
+  "prompt_tokens": 127,
+  "reused": 0,
+  "fresh": 127,
+  "prefill_ms": 324,
+  "prefill_tps": 392.4,
+  "decode_ms": 87,
+  "other_ms": 21.0,
+  "label": "nocache"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 345.8,
+  "wall_ms": 413.4,
+  "prompt_tokens": 127,
+  "reused": 0,
+  "fresh": 127,
+  "prefill_ms": 324,
+  "prefill_tps": 391.8,
+  "decode_ms": 86,
+  "kv_hit_pct": 90.66,
+  "kv_mib_from_ram": 1.6,
+  "other_ms": 21.8,
+  "label": "nocache"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 344.9,
+  "wall_ms": 412.6,
+  "prompt_tokens": 127,
+  "reused": 0,
+  "fresh": 127,
+  "prefill_ms": 324,
+  "prefill_tps": 392.4,
+  "decode_ms": 86,
+  "kv_hit_pct": 90.66,
+  "kv_mib_from_ram": 1.6,
+  "other_ms": 20.9,
+  "label": "nocache"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 344.5,
+  "wall_ms": 411.9,
+  "prompt_tokens": 127,
+  "reused": 0,
+  "fresh": 127,
+  "prefill_ms": 323,
+  "prefill_tps": 393.3,
+  "decode_ms": 86,
+  "other_ms": 21.5,
+  "label": "nocache"
+ },
+ {
+  "target": 75,
+  "ttft_ms": 344.3,
+  "wall_ms": 411.9,
+  "prompt_tokens": 127,
+  "reused": 0,
+  "fresh": 127,
+  "prefill_ms": 323,
+  "prefill_tps": 393.2,
+  "decode_ms": 86,
+  "kv_hit_pct": 90.66,
+  "kv_mib_from_ram": 1.6,
+  "other_ms": 21.3,
+  "label": "nocache"
+ }
+]
\ No newline at end of file
diff --git a/dirigo-results/0.2.13-port/intercept/probe-on-byteprefix.json b/dirigo-results/0.2.13-port/intercept/probe-on-byteprefix.json
new file mode 100644
index 00000000000..3cb134d7a92
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/probe-on-byteprefix.json
@@ -0,0 +1,294 @@
+[
+ {
+  "target": 4,
+  "ttft_ms": 0.9,
+  "wall_ms": 606.4,
+  "prompt_tokens": 74,
+  "reused": 0,
+  "fresh": 74,
+  "prefill_ms": 495,
+  "prefill_tps": 149.5,
+  "decode_ms": 109,
+  "park_tokens": 96,
+  "park_ms": 99.8,
+  "other_ms": -593.9,
+  "label": "on"
+ },
+ {
+  "target": 4,
+  "ttft_ms": 0.8,
+  "wall_ms": 582.0,
+  "prompt_tokens": 74,
+  "reused": 0,
+  "fresh": 74,
+  "prefill_ms": 487,
+  "prefill_tps": 152.0,
+  "decode_ms": 93,
+  "park_tokens": 84,
+  "park_ms": 101.2,
+  "kv_hit_pct": 93.31,
+  "kv_mib_from_ram": 1.0,
+  "other_ms": -587.4,
+  "label": "on"
+ },
+ {
+  "target": 8,
+  "ttft_ms": 1.0,
+  "wall_ms": 495.1,
+  "prompt_tokens": 75,
+  "reused": 0,
+  "fresh": 75,
+  "prefill_ms": 400,
+  "prefill_tps": 187.7,
+  "decode_ms": 93,
+  "park_tokens": 84,
+  "park_ms": 99.8,
+  "kv_hit_pct": 93.31,
+  "kv_mib_from_ram": 1.0,
+  "other_ms": -498.8,
+  "label": "on"
+ },
+ {
+  "target": 8,
+  "ttft_ms": 0.8,
+  "wall_ms": 386.9,
+  "prompt_tokens": 75,
+  "reused": 0,
+  "fresh": 75,
+  "prefill_ms": 291,
+  "prefill_tps": 257.4,
+  "decode_ms": 93,
+  "kv_hit_pct": 90.98,
+  "kv_mib_from_ram": 1.1,
+  "other_ms": -290.2,
+  "label": "on"
+ },
+ {
+  "target": 16,
+  "ttft_ms": 1.0,
+  "wall_ms": 430.8,
+  "prompt_tokens": 83,
+  "reused": 0,
+  "fresh": 83,
+  "prefill_ms": 302,
+  "prefill_tps": 275.2,
+  "decode_ms": 126,
+  "kv_hit_pct": 90.98,
+  "kv_mib_from_ram": 1.1,
+  "other_ms": -301.0,
+  "label": "on"
+ },
+ {
+  "target": 16,
+  "ttft_ms": 0.9,
+  "wall_ms": 401.0,
+  "prompt_tokens": 83,
+  "reused": 0,
+  "fresh": 83,
+  "prefill_ms": 300,
+  "prefill_tps": 276.7,
+  "decode_ms": 99,
+  "kv_hit_pct": 93.73,
+  "kv_mib_from_ram": 1.2,
+  "other_ms": -299.1,
+  "label": "on"
+ },
+ {
+  "target": 32,
+  "ttft_ms": 0.8,
+  "wall_ms": 406.6,
+  "prompt_tokens": 95,
+  "reused": 0,
+  "fresh": 95,
+  "prefill_ms": 308,
+  "prefill_tps": 308.0,
+  "decode_ms": 95,
+  "other_ms": -307.2,
+  "label": "on"
+ },
+ {
+  "target": 32,
+  "ttft_ms": 0.9,
+  "wall_ms": 401.6,
+  "prompt_tokens": 95,
+  "reused": 0,
+  "fresh": 95,
+  "prefill_ms": 307,
+  "prefill_tps": 309.8,
+  "decode_ms": 92,
+  "other_ms": -306.1,
+  "label": "on"
+ },
+ {
+  "target": 64,
+  "ttft_ms": 0.9,
+  "wall_ms": 424.9,
+  "prompt_tokens": 120,
+  "reused": 0,
+  "fresh": 120,
+  "prefill_ms": 336,
+  "prefill_tps": 357.5,
+  "decode_ms": 87,
+  "kv_hit_pct": 91.12,
+  "kv_mib_from_ram": 1.3,
+  "other_ms": -335.1,
+  "label": "on"
+ },
+ {
+  "target": 64,
+  "ttft_ms": 1.0,
+  "wall_ms": 409.4,
+  "prompt_tokens": 120,
+  "reused": 0,
+  "fresh": 120,
+  "prefill_ms": 320,
+  "prefill_tps": 375.0,
+  "decode_ms": 87,
+  "kv_hit_pct": 90.72,
+  "kv_mib_from_ram": 1.5,
+  "other_ms": -319.0,
+  "label": "on"
+ },
+ {
+  "target": 128,
+  "ttft_ms": 1.0,
+  "wall_ms": 453.0,
+  "prompt_tokens": 167,
+  "reused": 0,
+  "fresh": 167,
+  "prefill_ms": 364,
+  "prefill_tps": 459.4,
+  "decode_ms": 87,
+  "other_ms": -363.0,
+  "label": "on"
+ },
+ {
+  "target": 128,
+  "ttft_ms": 1.0,
+  "wall_ms": 429.6,
+  "prompt_tokens": 167,
+  "reused": 0,
+  "fresh": 167,
+  "prefill_ms": 340,
+  "prefill_tps": 490.9,
+  "decode_ms": 87,
+  "other_ms": -339.0,
+  "label": "on"
+ },
+ {
+  "target": 256,
+  "ttft_ms": 1.2,
+  "wall_ms": 548.7,
+  "prompt_tokens": 264,
+  "reused": 0,
+  "fresh": 264,
+  "prefill_ms": 459,
+  "prefill_tps": 575.0,
+  "decode_ms": 87,
+  "kv_hit_pct": 90.72,
+  "kv_mib_from_ram": 2.1,
+  "other_ms": -457.8,
+  "label": "on"
+ },
+ {
+  "target": 256,
+  "ttft_ms": 1.5,
+  "wall_ms": 509.1,
+  "prompt_tokens": 264,
+  "reused": 0,
+  "fresh": 264,
+  "prefill_ms": 419,
+  "prefill_tps": 629.6,
+  "decode_ms": 87,
+  "kv_hit_pct": 90.82,
+  "kv_mib_from_ram": 3.3,
+  "other_ms": -417.5,
+  "label": "on"
+ },
+ {
+  "target": 512,
+  "ttft_ms": 1.8,
+  "wall_ms": 646.4,
+  "prompt_tokens": 456,
+  "reused": 0,
+  "fresh": 456,
+  "prefill_ms": 533,
+  "prefill_tps": 855.9,
+  "decode_ms": 110,
+  "other_ms": -531.2,
+  "label": "on"
+ },
+ {
+  "target": 512,
+  "ttft_ms": 1.6,
+  "wall_ms": 630.6,
+  "prompt_tokens": 456,
+  "reused": 0,
+  "fresh": 456,
+  "prefill_ms": 517,
+  "prefill_tps": 882.0,
+  "decode_ms": 110,
+  "kv_hit_pct": 92.27,
+  "kv_mib_from_ram": 5.6,
+  "other_ms": -515.4,
+  "label": "on"
+ },
+ {
+  "target": 1024,
+  "ttft_ms": 2.6,
+  "wall_ms": 1036.4,
+  "prompt_tokens": 842,
+  "reused": 0,
+  "fresh": 842,
+  "prefill_ms": 921,
+  "prefill_tps": 914.7,
+  "decode_ms": 111,
+  "kv_hit_pct": 92.27,
+  "kv_mib_from_ram": 5.6,
+  "other_ms": -918.4,
+  "label": "on"
+ },
+ {
+  "target": 1024,
+  "ttft_ms": 2.5,
+  "wall_ms": 1027.3,
+  "prompt_tokens": 842,
+  "reused": 0,
+  "fresh": 842,
+  "prefill_ms": 911,
+  "prefill_tps": 923.8,
+  "decode_ms": 111,
+  "kv_hit_pct": 92.27,
+  "kv_mib_from_ram": 10.3,
+  "other_ms": -908.5,
+  "label": "on"
+ },
+ {
+  "target": 2048,
+  "ttft_ms": 4.8,
+  "wall_ms": 1565.1,
+  "prompt_tokens": 1608,
+  "reused": 0,
+  "fresh": 1608,
+  "prefill_ms": 1432,
+  "prefill_tps": 1122.6,
+  "decode_ms": 127,
+  "kv_hit_pct": 90.89,
+  "kv_mib_from_ram": 19.5,
+  "other_ms": -1427.2,
+  "label": "on"
+ },
+ {
+  "target": 2048,
+  "ttft_ms": 3.9,
+  "wall_ms": 1548.3,
+  "prompt_tokens": 1608,
+  "reused": 0,
+  "fresh": 1608,
+  "prefill_ms": 1452,
+  "prefill_tps": 1107.2,
+  "decode_ms": 90,
+  "other_ms": -1448.1,
+  "label": "on"
+ }
+]
\ No newline at end of file
diff --git a/dirigo-results/0.2.13-port/intercept/probe-on.json b/dirigo-results/0.2.13-port/intercept/probe-on.json
new file mode 100644
index 00000000000..f5e1694460a
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/probe-on.json
@@ -0,0 +1,304 @@
+[
+ {
+  "target": 4,
+  "ttft_ms": 505.4,
+  "wall_ms": 596.7,
+  "prompt_tokens": 74,
+  "reused": 0,
+  "fresh": 74,
+  "prefill_ms": 484,
+  "prefill_tps": 152.8,
+  "decode_ms": 109,
+  "park_tokens": 96,
+  "park_ms": 98.1,
+  "kv_hit_pct": 93.1,
+  "kv_mib_from_ram": 1.2,
+  "other_ms": -76.7,
+  "label": "on"
+ },
+ {
+  "target": 4,
+  "ttft_ms": 506.0,
+  "wall_ms": 580.6,
+  "prompt_tokens": 74,
+  "reused": 0,
+  "fresh": 74,
+  "prefill_ms": 485,
+  "prefill_tps": 152.5,
+  "decode_ms": 93,
+  "park_tokens": 84,
+  "park_ms": 101.8,
+  "other_ms": -80.8,
+  "label": "on"
+ },
+ {
+  "target": 8,
+  "ttft_ms": 423.1,
+  "wall_ms": 497.3,
+  "prompt_tokens": 75,
+  "reused": 0,
+  "fresh": 75,
+  "prefill_ms": 402,
+  "prefill_tps": 186.6,
+  "decode_ms": 93,
+  "park_tokens": 84,
+  "park_ms": 101.9,
+  "kv_hit_pct": 93.31,
+  "kv_mib_from_ram": 1.0,
+  "other_ms": -80.8,
+  "label": "on"
+ },
+ {
+  "target": 8,
+  "ttft_ms": 311.1,
+  "wall_ms": 416.5,
+  "prompt_tokens": 75,
+  "reused": 0,
+  "fresh": 75,
+  "prefill_ms": 290,
+  "prefill_tps": 258.6,
+  "decode_ms": 124,
+  "kv_hit_pct": 90.98,
+  "kv_mib_from_ram": 1.1,
+  "other_ms": 21.1,
+  "label": "on"
+ },
+ {
+  "target": 16,
+  "ttft_ms": 320.6,
+  "wall_ms": 399.9,
+  "prompt_tokens": 83,
+  "reused": 0,
+  "fresh": 83,
+  "prefill_ms": 299,
+  "prefill_tps": 277.3,
+  "decode_ms": 98,
+  "kv_hit_pct": 93.29,
+  "kv_mib_from_ram": 1.1,
+  "other_ms": 21.6,
+  "label": "on"
+ },
+ {
+  "target": 16,
+  "ttft_ms": 319.6,
+  "wall_ms": 384.4,
+  "prompt_tokens": 83,
+  "reused": 0,
+  "fresh": 83,
+  "prefill_ms": 298,
+  "prefill_tps": 278.1,
+  "decode_ms": 83,
+  "kv_hit_pct": 88.44,
+  "kv_mib_from_ram": 1.1,
+  "other_ms": 21.6,
+  "label": "on"
+ },
+ {
+  "target": 32,
+  "ttft_ms": 329.5,
+  "wall_ms": 407.3,
+  "prompt_tokens": 95,
+  "reused": 0,
+  "fresh": 95,
+  "prefill_ms": 308,
+  "prefill_tps": 308.3,
+  "decode_ms": 97,
+  "kv_hit_pct": 89.64,
+  "kv_mib_from_ram": 1.1,
+  "other_ms": 21.5,
+  "label": "on"
+ },
+ {
+  "target": 32,
+  "ttft_ms": 326.8,
+  "wall_ms": 400.2,
+  "prompt_tokens": 95,
+  "reused": 0,
+  "fresh": 95,
+  "prefill_ms": 306,
+  "prefill_tps": 310.7,
+  "decode_ms": 92,
+  "kv_hit_pct": 88.5,
+  "kv_mib_from_ram": 1.3,
+  "other_ms": 20.8,
+  "label": "on"
+ },
+ {
+  "target": 64,
+  "ttft_ms": 356.5,
+  "wall_ms": 423.9,
+  "prompt_tokens": 120,
+  "reused": 0,
+  "fresh": 120,
+  "prefill_ms": 335,
+  "prefill_tps": 358.0,
+  "decode_ms": 86,
+  "kv_hit_pct": 91.12,
+  "kv_mib_from_ram": 1.3,
+  "other_ms": 21.5,
+  "label": "on"
+ },
+ {
+  "target": 64,
+  "ttft_ms": 342.9,
+  "wall_ms": 410.4,
+  "prompt_tokens": 120,
+  "reused": 0,
+  "fresh": 120,
+  "prefill_ms": 322,
+  "prefill_tps": 373.1,
+  "decode_ms": 86,
+  "kv_hit_pct": 90.72,
+  "kv_mib_from_ram": 1.5,
+  "other_ms": 20.9,
+  "label": "on"
+ },
+ {
+  "target": 128,
+  "ttft_ms": 385.4,
+  "wall_ms": 452.9,
+  "prompt_tokens": 167,
+  "reused": 0,
+  "fresh": 167,
+  "prefill_ms": 364,
+  "prefill_tps": 459.1,
+  "decode_ms": 86,
+  "kv_hit_pct": 90.72,
+  "kv_mib_from_ram": 1.5,
+  "other_ms": 21.4,
+  "label": "on"
+ },
+ {
+  "target": 128,
+  "ttft_ms": 361.8,
+  "wall_ms": 446.7,
+  "prompt_tokens": 167,
+  "reused": 0,
+  "fresh": 167,
+  "prefill_ms": 340,
+  "prefill_tps": 490.6,
+  "decode_ms": 104,
+  "kv_hit_pct": 90.72,
+  "kv_mib_from_ram": 2.1,
+  "other_ms": 21.8,
+  "label": "on"
+ },
+ {
+  "target": 256,
+  "ttft_ms": 481.9,
+  "wall_ms": 566.3,
+  "prompt_tokens": 264,
+  "reused": 0,
+  "fresh": 264,
+  "prefill_ms": 460,
+  "prefill_tps": 574.1,
+  "decode_ms": 104,
+  "kv_hit_pct": 90.82,
+  "kv_mib_from_ram": 3.3,
+  "other_ms": 21.9,
+  "label": "on"
+ },
+ {
+  "target": 256,
+  "ttft_ms": 441.3,
+  "wall_ms": 508.9,
+  "prompt_tokens": 264,
+  "reused": 0,
+  "fresh": 264,
+  "prefill_ms": 419,
+  "prefill_tps": 629.5,
+  "decode_ms": 86,
+  "other_ms": 22.3,
+  "label": "on"
+ },
+ {
+  "target": 512,
+  "ttft_ms": 552.8,
+  "wall_ms": 637.6,
+  "prompt_tokens": 456,
+  "reused": 0,
+  "fresh": 456,
+  "prefill_ms": 530,
+  "prefill_tps": 860.1,
+  "decode_ms": 105,
+  "kv_hit_pct": 90.82,
+  "kv_mib_from_ram": 3.3,
+  "other_ms": 22.8,
+  "label": "on"
+ },
+ {
+  "target": 512,
+  "ttft_ms": 537.8,
+  "wall_ms": 622.2,
+  "prompt_tokens": 456,
+  "reused": 0,
+  "fresh": 456,
+  "prefill_ms": 515,
+  "prefill_tps": 885.1,
+  "decode_ms": 104,
+  "kv_hit_pct": 90.86,
+  "kv_mib_from_ram": 5.6,
+  "other_ms": 22.8,
+  "label": "on"
+ },
+ {
+  "target": 1024,
+  "ttft_ms": 939.5,
+  "wall_ms": 1007.6,
+  "prompt_tokens": 842,
+  "reused": 0,
+  "fresh": 842,
+  "prefill_ms": 915,
+  "prefill_tps": 920.0,
+  "decode_ms": 88,
+  "kv_hit_pct": 90.86,
+  "kv_mib_from_ram": 5.6,
+  "other_ms": 24.5,
+  "label": "on"
+ },
+ {
+  "target": 1024,
+  "ttft_ms": 935.8,
+  "wall_ms": 1027.1,
+  "prompt_tokens": 842,
+  "reused": 0,
+  "fresh": 842,
+  "prefill_ms": 911,
+  "prefill_tps": 923.8,
+  "decode_ms": 111,
+  "kv_hit_pct": 90.86,
+  "kv_mib_from_ram": 10.3,
+  "other_ms": 24.8,
+  "label": "on"
+ },
+ {
+  "target": 2048,
+  "ttft_ms": 1469.2,
+  "wall_ms": 1537.8,
+  "prompt_tokens": 1608,
+  "reused": 0,
+  "fresh": 1608,
+  "prefill_ms": 1442,
+  "prefill_tps": 1115.2,
+  "decode_ms": 90,
+  "kv_hit_pct": 92.27,
+  "kv_mib_from_ram": 10.3,
+  "other_ms": 27.2,
+  "label": "on"
+ },
+ {
+  "target": 2048,
+  "ttft_ms": 1475.0,
+  "wall_ms": 1566.9,
+  "prompt_tokens": 1608,
+  "reused": 0,
+  "fresh": 1608,
+  "prefill_ms": 1443,
+  "prefill_tps": 1114.0,
+  "decode_ms": 113,
+  "kv_hit_pct": 90.89,
+  "kv_mib_from_ram": 19.5,
+  "other_ms": 32.0,
+  "label": "on"
+ }
+]
\ No newline at end of file
diff --git a/dirigo-results/0.2.13-port/intercept/probe-shadow-byteprefix.json b/dirigo-results/0.2.13-port/intercept/probe-shadow-byteprefix.json
new file mode 100644
index 00000000000..650ec1c971c
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/probe-shadow-byteprefix.json
@@ -0,0 +1,332 @@
+[
+ {
+  "target": 4,
+  "ttft_ms": 0.9,
+  "wall_ms": 605.6,
+  "prompt_tokens": 74,
+  "reused": 0,
+  "fresh": 74,
+  "prefill_ms": 493,
+  "prefill_tps": 150.2,
+  "decode_ms": 110,
+  "park_tokens": 97,
+  "park_ms": 98.1,
+  "other_ms": -590.2,
+  "label": "shadow"
+ },
+ {
+  "target": 4,
+  "ttft_ms": 0.8,
+  "wall_ms": 617.6,
+  "prompt_tokens": 74,
+  "reused": 0,
+  "fresh": 74,
+  "prefill_ms": 489,
+  "prefill_tps": 151.4,
+  "decode_ms": 126,
+  "park_tokens": 82,
+  "park_ms": 100.4,
+  "other_ms": -588.6,
+  "label": "shadow"
+ },
+ {
+  "target": 8,
+  "ttft_ms": 0.9,
+  "wall_ms": 624.3,
+  "prompt_tokens": 75,
+  "reused": 0,
+  "fresh": 75,
+  "prefill_ms": 498,
+  "prefill_tps": 150.7,
+  "decode_ms": 124,
+  "park_tokens": 84,
+  "park_ms": 105.3,
+  "kv_hit_pct": 94.93,
+  "kv_mib_from_ram": 1.0,
+  "other_ms": -602.4,
+  "label": "shadow"
+ },
+ {
+  "target": 8,
+  "ttft_ms": 0.8,
+  "wall_ms": 617.3,
+  "prompt_tokens": 75,
+  "reused": 0,
+  "fresh": 75,
+  "prefill_ms": 491,
+  "prefill_tps": 152.7,
+  "decode_ms": 124,
+  "park_tokens": 86,
+  "park_ms": 102.0,
+  "kv_hit_pct": 94.53,
+  "kv_mib_from_ram": 1.1,
+  "other_ms": -592.2,
+  "label": "shadow"
+ },
+ {
+  "target": 16,
+  "ttft_ms": 0.9,
+  "wall_ms": 615.4,
+  "prompt_tokens": 83,
+  "reused": 0,
+  "fresh": 83,
+  "prefill_ms": 511,
+  "prefill_tps": 162.3,
+  "decode_ms": 101,
+  "park_tokens": 86,
+  "park_ms": 101.0,
+  "kv_hit_pct": 94.53,
+  "kv_mib_from_ram": 1.1,
+  "other_ms": -611.1,
+  "label": "shadow"
+ },
+ {
+  "target": 16,
+  "ttft_ms": 0.8,
+  "wall_ms": 595.8,
+  "prompt_tokens": 83,
+  "reused": 0,
+  "fresh": 83,
+  "prefill_ms": 510,
+  "prefill_tps": 162.9,
+  "decode_ms": 84,
+  "park_tokens": 90,
+  "park_ms": 100.9,
+  "kv_hit_pct": 92.43,
+  "kv_mib_from_ram": 1.1,
+  "other_ms": -610.1,
+  "label": "shadow"
+ },
+ {
+  "target": 32,
+  "ttft_ms": 0.9,
+  "wall_ms": 617.4,
+  "prompt_tokens": 95,
+  "reused": 0,
+  "fresh": 95,
+  "prefill_ms": 519,
+  "prefill_tps": 183.0,
+  "decode_ms": 95,
+  "park_tokens": 90,
+  "park_ms": 100.0,
+  "kv_hit_pct": 92.43,
+  "kv_mib_from_ram": 1.1,
+  "other_ms": -618.1,
+  "label": "shadow"
+ },
+ {
+  "target": 32,
+  "ttft_ms": 0.8,
+  "wall_ms": 643.6,
+  "prompt_tokens": 95,
+  "reused": 0,
+  "fresh": 95,
+  "prefill_ms": 522,
+  "prefill_tps": 182.0,
+  "decode_ms": 119,
+  "park_tokens": 105,
+  "park_ms": 102.2,
+  "kv_hit_pct": 93.63,
+  "kv_mib_from_ram": 1.3,
+  "other_ms": -623.4,
+  "label": "shadow"
+ },
+ {
+  "target": 64,
+  "ttft_ms": 0.9,
+  "wall_ms": 636.0,
+  "prompt_tokens": 120,
+  "reused": 0,
+  "fresh": 120,
+  "prefill_ms": 546,
+  "prefill_tps": 219.6,
+  "decode_ms": 87,
+  "park_tokens": 107,
+  "park_ms": 101.2,
+  "kv_hit_pct": 94.35,
+  "kv_mib_from_ram": 1.3,
+  "other_ms": -646.3,
+  "label": "shadow"
+ },
+ {
+  "target": 64,
+  "ttft_ms": 1.0,
+  "wall_ms": 622.4,
+  "prompt_tokens": 120,
+  "reused": 0,
+  "fresh": 120,
+  "prefill_ms": 533,
+  "prefill_tps": 225.2,
+  "decode_ms": 87,
+  "park_tokens": 127,
+  "park_ms": 100.4,
+  "kv_hit_pct": 93.1,
+  "kv_mib_from_ram": 1.5,
+  "other_ms": -632.4,
+  "label": "shadow"
+ },
+ {
+  "target": 128,
+  "ttft_ms": 1.0,
+  "wall_ms": 666.8,
+  "prompt_tokens": 167,
+  "reused": 0,
+  "fresh": 167,
+  "prefill_ms": 577,
+  "prefill_tps": 289.4,
+  "decode_ms": 87,
+  "park_tokens": 127,
+  "park_ms": 101.7,
+  "kv_hit_pct": 93.1,
+  "kv_mib_from_ram": 1.5,
+  "other_ms": -677.7,
+  "label": "shadow"
+ },
+ {
+  "target": 128,
+  "ttft_ms": 1.0,
+  "wall_ms": 664.7,
+  "prompt_tokens": 167,
+  "reused": 0,
+  "fresh": 167,
+  "prefill_ms": 558,
+  "prefill_tps": 299.1,
+  "decode_ms": 104,
+  "park_tokens": 174,
+  "park_ms": 100.7,
+  "kv_hit_pct": 93.12,
+  "kv_mib_from_ram": 2.1,
+  "other_ms": -657.7,
+  "label": "shadow"
+ },
+ {
+  "target": 256,
+  "ttft_ms": 1.3,
+  "wall_ms": 761.2,
+  "prompt_tokens": 264,
+  "reused": 0,
+  "fresh": 264,
+  "prefill_ms": 672,
+  "prefill_tps": 393.0,
+  "decode_ms": 86,
+  "park_tokens": 174,
+  "park_ms": 102.0,
+  "other_ms": -772.7,
+  "label": "shadow"
+ },
+ {
+  "target": 256,
+  "ttft_ms": 1.3,
+  "wall_ms": 724.8,
+  "prompt_tokens": 264,
+  "reused": 0,
+  "fresh": 264,
+  "prefill_ms": 635,
+  "prefill_tps": 415.9,
+  "decode_ms": 87,
+  "park_tokens": 271,
+  "park_ms": 99.3,
+  "kv_hit_pct": 93.23,
+  "kv_mib_from_ram": 3.3,
+  "other_ms": -733.0,
+  "label": "shadow"
+ },
+ {
+  "target": 512,
+  "ttft_ms": 1.7,
+  "wall_ms": 861.4,
+  "prompt_tokens": 456,
+  "reused": 0,
+  "fresh": 456,
+  "prefill_ms": 748,
+  "prefill_tps": 609.3,
+  "decode_ms": 110,
+  "park_tokens": 271,
+  "park_ms": 101.2,
+  "other_ms": -847.5,
+  "label": "shadow"
+ },
+ {
+  "target": 512,
+  "ttft_ms": 1.7,
+  "wall_ms": 860.2,
+  "prompt_tokens": 456,
+  "reused": 0,
+  "fresh": 456,
+  "prefill_ms": 730,
+  "prefill_tps": 624.4,
+  "decode_ms": 127,
+  "park_tokens": 463,
+  "park_ms": 101.4,
+  "kv_hit_pct": 94.07,
+  "kv_mib_from_ram": 5.6,
+  "other_ms": -829.7,
+  "label": "shadow"
+ },
+ {
+  "target": 1024,
+  "ttft_ms": 2.6,
+  "wall_ms": 1273.2,
+  "prompt_tokens": 842,
+  "reused": 0,
+  "fresh": 842,
+  "prefill_ms": 1144,
+  "prefill_tps": 736.2,
+  "decode_ms": 125,
+  "park_tokens": 463,
+  "park_ms": 100.7,
+  "other_ms": -1242.1,
+  "label": "shadow"
+ },
+ {
+  "target": 1024,
+  "ttft_ms": 2.5,
+  "wall_ms": 1268.7,
+  "prompt_tokens": 842,
+  "reused": 0,
+  "fresh": 842,
+  "prefill_ms": 1139,
+  "prefill_tps": 739.2,
+  "decode_ms": 125,
+  "park_tokens": 849,
+  "park_ms": 102.1,
+  "kv_hit_pct": 95.21,
+  "kv_mib_from_ram": 10.3,
+  "other_ms": -1238.6,
+  "label": "shadow"
+ },
+ {
+  "target": 2048,
+  "ttft_ms": 3.8,
+  "wall_ms": 1769.9,
+  "prompt_tokens": 1608,
+  "reused": 0,
+  "fresh": 1608,
+  "prefill_ms": 1676,
+  "prefill_tps": 959.3,
+  "decode_ms": 88,
+  "park_tokens": 849,
+  "park_ms": 104.8,
+  "kv_hit_pct": 95.21,
+  "kv_mib_from_ram": 10.3,
+  "other_ms": -1777.0,
+  "label": "shadow"
+ },
+ {
+  "target": 2048,
+  "ttft_ms": 3.9,
+  "wall_ms": 1772.1,
+  "prompt_tokens": 1608,
+  "reused": 0,
+  "fresh": 1608,
+  "prefill_ms": 1678,
+  "prefill_tps": 958.1,
+  "decode_ms": 88,
+  "park_tokens": 1615,
+  "park_ms": 110.8,
+  "kv_hit_pct": 93.32,
+  "kv_mib_from_ram": 19.5,
+  "other_ms": -1784.9,
+  "label": "shadow"
+ }
+]
\ No newline at end of file
diff --git a/dirigo-results/0.2.13-port/intercept/probe-shadow.json b/dirigo-results/0.2.13-port/intercept/probe-shadow.json
new file mode 100644
index 00000000000..fe996de0c83
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/probe-shadow.json
@@ -0,0 +1,328 @@
+[
+ {
+  "target": 4,
+  "ttft_ms": 515.6,
+  "wall_ms": 631.9,
+  "prompt_tokens": 74,
+  "reused": 0,
+  "fresh": 74,
+  "prefill_ms": 495,
+  "prefill_tps": 149.6,
+  "decode_ms": 134,
+  "park_tokens": 97,
+  "park_ms": 101.7,
+  "kv_hit_pct": 91.69,
+  "kv_mib_from_ram": 1.2,
+  "other_ms": -81.1,
+  "label": "shadow"
+ },
+ {
+  "target": 4,
+  "ttft_ms": 512.7,
+  "wall_ms": 633.3,
+  "prompt_tokens": 74,
+  "reused": 0,
+  "fresh": 74,
+  "prefill_ms": 492,
+  "prefill_tps": 150.5,
+  "decode_ms": 139,
+  "park_tokens": 84,
+  "park_ms": 104.6,
+  "kv_hit_pct": 95.39,
+  "kv_mib_from_ram": 1.0,
+  "other_ms": -83.9,
+  "label": "shadow"
+ },
+ {
+  "target": 8,
+  "ttft_ms": 513.2,
+  "wall_ms": 588.1,
+  "prompt_tokens": 75,
+  "reused": 0,
+  "fresh": 75,
+  "prefill_ms": 492,
+  "prefill_tps": 152.4,
+  "decode_ms": 93,
+  "park_tokens": 84,
+  "park_ms": 100.6,
+  "kv_hit_pct": 94.67,
+  "kv_mib_from_ram": 1.0,
+  "other_ms": -79.4,
+  "label": "shadow"
+ },
+ {
+  "target": 8,
+  "ttft_ms": 510.3,
+  "wall_ms": 615.3,
+  "prompt_tokens": 75,
+  "reused": 0,
+  "fresh": 75,
+  "prefill_ms": 489,
+  "prefill_tps": 153.3,
+  "decode_ms": 123,
+  "park_tokens": 85,
+  "park_ms": 102.1,
+  "kv_hit_pct": 94.53,
+  "kv_mib_from_ram": 1.1,
+  "other_ms": -80.8,
+  "label": "shadow"
+ },
+ {
+  "target": 16,
+  "ttft_ms": 530.6,
+  "wall_ms": 607.5,
+  "prompt_tokens": 83,
+  "reused": 0,
+  "fresh": 83,
+  "prefill_ms": 510,
+  "prefill_tps": 162.8,
+  "decode_ms": 95,
+  "park_tokens": 86,
+  "park_ms": 101.3,
+  "other_ms": -80.7,
+  "label": "shadow"
+ },
+ {
+  "target": 16,
+  "ttft_ms": 532.6,
+  "wall_ms": 639.2,
+  "prompt_tokens": 83,
+  "reused": 0,
+  "fresh": 83,
+  "prefill_ms": 512,
+  "prefill_tps": 162.3,
+  "decode_ms": 125,
+  "park_tokens": 93,
+  "park_ms": 101.8,
+  "other_ms": -81.2,
+  "label": "shadow"
+ },
+ {
+  "target": 32,
+  "ttft_ms": 541.0,
+  "wall_ms": 618.1,
+  "prompt_tokens": 95,
+  "reused": 0,
+  "fresh": 95,
+  "prefill_ms": 520,
+  "prefill_tps": 182.7,
+  "decode_ms": 95,
+  "park_tokens": 94,
+  "park_ms": 102.3,
+  "kv_hit_pct": 94.84,
+  "kv_mib_from_ram": 1.2,
+  "other_ms": -81.3,
+  "label": "shadow"
+ },
+ {
+  "target": 32,
+  "ttft_ms": 540.4,
+  "wall_ms": 617.4,
+  "prompt_tokens": 95,
+  "reused": 0,
+  "fresh": 95,
+  "prefill_ms": 519,
+  "prefill_tps": 182.9,
+  "decode_ms": 95,
+  "park_tokens": 105,
+  "park_ms": 101.7,
+  "kv_hit_pct": 93.63,
+  "kv_mib_from_ram": 1.3,
+  "other_ms": -80.3,
+  "label": "shadow"
+ },
+ {
+  "target": 64,
+  "ttft_ms": 567.6,
+  "wall_ms": 635.8,
+  "prompt_tokens": 120,
+  "reused": 0,
+  "fresh": 120,
+  "prefill_ms": 547,
+  "prefill_tps": 219.6,
+  "decode_ms": 87,
+  "park_tokens": 105,
+  "park_ms": 100.3,
+  "kv_hit_pct": 93.63,
+  "kv_mib_from_ram": 1.3,
+  "other_ms": -79.7,
+  "label": "shadow"
+ },
+ {
+  "target": 64,
+  "ttft_ms": 554.4,
+  "wall_ms": 623.1,
+  "prompt_tokens": 120,
+  "reused": 0,
+  "fresh": 120,
+  "prefill_ms": 533,
+  "prefill_tps": 225.0,
+  "decode_ms": 87,
+  "park_tokens": 127,
+  "park_ms": 99.9,
+  "other_ms": -78.5,
+  "label": "shadow"
+ },
+ {
+  "target": 128,
+  "ttft_ms": 597.7,
+  "wall_ms": 666.2,
+  "prompt_tokens": 167,
+  "reused": 0,
+  "fresh": 167,
+  "prefill_ms": 577,
+  "prefill_tps": 289.6,
+  "decode_ms": 87,
+  "park_tokens": 127,
+  "park_ms": 101.0,
+  "kv_hit_pct": 93.12,
+  "kv_mib_from_ram": 2.1,
+  "other_ms": -80.3,
+  "label": "shadow"
+ },
+ {
+  "target": 128,
+  "ttft_ms": 580.4,
+  "wall_ms": 648.9,
+  "prompt_tokens": 167,
+  "reused": 0,
+  "fresh": 167,
+  "prefill_ms": 559,
+  "prefill_tps": 298.6,
+  "decode_ms": 87,
+  "park_tokens": 174,
+  "park_ms": 102.6,
+  "kv_hit_pct": 93.12,
+  "kv_mib_from_ram": 2.1,
+  "other_ms": -81.2,
+  "label": "shadow"
+ },
+ {
+  "target": 256,
+  "ttft_ms": 696.7,
+  "wall_ms": 764.8,
+  "prompt_tokens": 264,
+  "reused": 0,
+  "fresh": 264,
+  "prefill_ms": 675,
+  "prefill_tps": 390.9,
+  "decode_ms": 86,
+  "park_tokens": 174,
+  "park_ms": 101.2,
+  "other_ms": -79.5,
+  "label": "shadow"
+ },
+ {
+  "target": 256,
+  "ttft_ms": 656.4,
+  "wall_ms": 754.5,
+  "prompt_tokens": 264,
+  "reused": 0,
+  "fresh": 264,
+  "prefill_ms": 635,
+  "prefill_tps": 415.8,
+  "decode_ms": 117,
+  "park_tokens": 271,
+  "park_ms": 100.5,
+  "other_ms": -79.1,
+  "label": "shadow"
+ },
+ {
+  "target": 512,
+  "ttft_ms": 776.9,
+  "wall_ms": 845.2,
+  "prompt_tokens": 456,
+  "reused": 0,
+  "fresh": 456,
+  "prefill_ms": 755,
+  "prefill_tps": 603.9,
+  "decode_ms": 87,
+  "park_tokens": 271,
+  "park_ms": 100.7,
+  "other_ms": -78.8,
+  "label": "shadow"
+ },
+ {
+  "target": 512,
+  "ttft_ms": 755.8,
+  "wall_ms": 847.1,
+  "prompt_tokens": 456,
+  "reused": 0,
+  "fresh": 456,
+  "prefill_ms": 734,
+  "prefill_tps": 621.4,
+  "decode_ms": 110,
+  "park_tokens": 463,
+  "park_ms": 102.8,
+  "kv_hit_pct": 93.27,
+  "kv_mib_from_ram": 5.6,
+  "other_ms": -81.0,
+  "label": "shadow"
+ },
+ {
+  "target": 1024,
+  "ttft_ms": 1169.0,
+  "wall_ms": 1237.8,
+  "prompt_tokens": 842,
+  "reused": 0,
+  "fresh": 842,
+  "prefill_ms": 1146,
+  "prefill_tps": 734.6,
+  "decode_ms": 87,
+  "park_tokens": 463,
+  "park_ms": 101.2,
+  "kv_hit_pct": 94.07,
+  "kv_mib_from_ram": 5.6,
+  "other_ms": -78.2,
+  "label": "shadow"
+ },
+ {
+  "target": 1024,
+  "ttft_ms": 1165.1,
+  "wall_ms": 1242.0,
+  "prompt_tokens": 842,
+  "reused": 0,
+  "fresh": 842,
+  "prefill_ms": 1143,
+  "prefill_tps": 736.9,
+  "decode_ms": 95,
+  "park_tokens": 849,
+  "park_ms": 102.1,
+  "kv_hit_pct": 93.28,
+  "kv_mib_from_ram": 10.3,
+  "other_ms": -80.0,
+  "label": "shadow"
+ },
+ {
+  "target": 2048,
+  "ttft_ms": 1701.7,
+  "wall_ms": 1771.1,
+  "prompt_tokens": 1608,
+  "reused": 0,
+  "fresh": 1608,
+  "prefill_ms": 1678,
+  "prefill_tps": 958.6,
+  "decode_ms": 88,
+  "park_tokens": 849,
+  "park_ms": 104.3,
+  "other_ms": -80.6,
+  "label": "shadow"
+ },
+ {
+  "target": 2048,
+  "ttft_ms": 1723.2,
+  "wall_ms": 1815.8,
+  "prompt_tokens": 1608,
+  "reused": 0,
+  "fresh": 1608,
+  "prefill_ms": 1699,
+  "prefill_tps": 946.4,
+  "decode_ms": 111,
+  "park_tokens": 1615,
+  "park_ms": 110.6,
+  "kv_hit_pct": 93.32,
+  "kv_mib_from_ram": 19.5,
+  "other_ms": -86.4,
+  "label": "shadow"
+ }
+]
\ No newline at end of file
diff --git a/dirigo-results/0.2.13-port/intercept/strata-intercept-cause.sh b/dirigo-results/0.2.13-port/intercept/strata-intercept-cause.sh
new file mode 100755
index 00000000000..c785da58ba9
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/strata-intercept-cause.sh
@@ -0,0 +1,51 @@
+#!/usr/bin/env bash
+# strata-intercept-cause.sh - which engine path owns the ~220 ms that remains with the cache off?
+#
+# The ladder probe put the fixed cost at 222 ms (cache off) / 434 ms (cache on) inside the engine's own prefill
+# accounting, with the PLE read path already excluded by the row-cache sweep. This fixes the prompt at ~75 tokens
+# and toggles one engine subsystem per card window, six samples each:
+#
+#   base    deployed config
+#   nospec  --spec 0        (no MTP/suffix drafting: the verify geometry at request start)
+#   nocache --prompt-cache 0 (no conversation cache slots at all)
+#   base2   deployed config again (drift check)
+#
+# UNDO: rm this file and ~/strata-intercept-cause
+set -uo pipefail
+S=/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731
+TREE=$S/strata-0.2.13
+BASE=$TREE/strata-coder-iq1_m-sycl-tuned.json
+OUT=$HOME/strata-intercept-cause
+PROBE=$HOME/bin/strata-intercept-probe.py
+mkdir -p "$OUT"
+
+variant() {   # name extra-args...
+    local name=$1; shift
+    python3 - "$BASE" "$OUT/conf-$name.json" "$@" <<'PY'
+import json, sys
+src, dst, extra = sys.argv[1], sys.argv[2], sys.argv[3:]
+d = json.load(open(src)); d["args"] = list(d.get("args", [])) + extra
+json.dump(d, open(dst, "w"), indent=1); print(dst, "extra:", extra)
+PY
+}
+
+arm() {   # tag
+    local tag=$1 conf="$OUT/conf-$1.json"
+    echo "[$(date +%H:%M:%S)] ===== cause-$tag"
+    ~/bin/strata-card-window.sh "$TREE" 8097 --conf "$conf" -- bash -c "
+        EL=\$(python3 -c \"import json;print(json.load(open('$conf'))['log'])\")
+        curl -s -m 5 http://127.0.0.1:\$PORT/v1/status | python3 -c 'import json,sys;d=json.load(sys.stdin);print(\"  engine:\",d[\"engine\"])'
+        /usr/bin/python3 $PROBE --port \$PORT --log \"\$EL\" --out $OUT/probe-$tag.json --label $tag \
+            --sizes 75 --samples 6
+    " 2>&1 | grep -E '=====|engine:|^\s+\{|wrote|restore'
+}
+
+variant base
+variant base2
+variant nospec --spec 0
+variant nocache --prompt-cache 0
+
+for tag in base nospec nocache base2; do
+    arm "$tag"
+done
+echo "[$(date +%H:%M:%S)] cause sweep done"
diff --git a/dirigo-results/0.2.13-port/intercept/strata-intercept-probe.py b/dirigo-results/0.2.13-port/intercept/strata-intercept-probe.py
new file mode 100755
index 00000000000..e3eaad963c8
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/strata-intercept-probe.py
@@ -0,0 +1,128 @@
+#!/usr/bin/env python3
+"""Where a request's fixed ~330 ms goes: prompt read, conversation-cache park, or neither.
+
+Sends unique prompts of a ladder of sizes and, for each one, reads the engine's own accounting out of its log:
+
+    prompt N tokens = R reused + F read in X ms (Y tok/s), G generated in Z ms ...
+    conversation cache: parked N tokens in X ms; ...
+
+so three numbers can be separated per request: TTFT as the client sees it, the engine's prompt read, and the park
+copy. The difference between TTFT and those two is everything else the request pays (queue, server, first-token
+setup) — the term nobody has named.
+
+    ~/bin/strata-intercept-probe.py --port 8097 --log <engine log> --out /tmp/ip.json [--label a]
+
+Unique prompts only: an identical prefix hits the conversation cache and reports 14 000 tok/s "prefill", which
+measures the cache instead of the prompt path.
+"""
+import argparse
+import json
+import os
+import random
+import re
+import time
+import urllib.request
+
+FILLER = ("The engine streams expert weights over PCIe while the model is resident in RAM. ")
+SIZES = [4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048]
+RE_PROMPT = re.compile(r"prompt (\d+) tokens = (\d+) reused \+ (\d+) read in (\d+) ms \(([0-9.]+) tok/s\), "
+                       r"(\d+) generated in (\d+) ms")
+RE_PARK = re.compile(r"conversation cache: parked (\d+) tokens in ([0-9.]+) ms")
+RE_KV = re.compile(r"KV streaming: ([0-9.]+)% of (\d+) block reads hit VRAM, ([0-9.]+) MiB read from RAM")
+
+
+def prompt_of(tokens: int, salt: str) -> str:
+    body = (FILLER * (tokens // 15 + 2))[: max(8, tokens * 4)]
+    return salt + " " + body
+
+
+def tail_since(path: str, offset: int) -> str:
+    with open(path, "rb") as f:
+        f.seek(offset)
+        return f.read().decode("utf-8", "replace")
+
+
+def one(port: int, tokens: int, salt: str, log: str, wait_s: float = 60.0) -> dict:
+    off = os.path.getsize(log)
+    body = json.dumps({"model": "m", "max_tokens": 8, "temperature": 0.0, "stream": True,
+                       "messages": [{"role": "user", "content": prompt_of(tokens, salt)}]}).encode()
+    req = urllib.request.Request(f"http://127.0.0.1:{port}/v1/chat/completions", data=body,
+                                 headers={"Content-Type": "application/json"})
+    t0 = time.time()
+    first = None
+    with urllib.request.urlopen(req, timeout=180) as r:
+        for raw in r:
+            if first is not None:
+                continue
+            line = raw.decode("utf-8", "replace").strip()
+            if not line.startswith("data:"):
+                continue
+            try:
+                d = json.loads(line[5:].strip())
+            except Exception:
+                continue
+            delta = (d.get("choices") or [{}])[0].get("delta") or {}
+            # the first CONTENT token: the stream opens before the prompt is read, so the first byte is not TTFT
+            if delta.get("content") or delta.get("reasoning_content"):
+                first = time.time() - t0
+    wall = time.time() - t0
+    # the engine writes its accounting at the end of the request; poll for the prompt line
+    seg = ""
+    deadline = time.time() + wait_s
+    while time.time() < deadline:
+        seg = tail_since(log, off)
+        if RE_PROMPT.search(seg):
+            break
+        time.sleep(0.2)
+    m = None
+    for m in RE_PROMPT.finditer(seg):
+        pass
+    p = None
+    for p in RE_PARK.finditer(seg):
+        pass
+    kv = None
+    for kv in RE_KV.finditer(seg):
+        pass
+    row = dict(target=tokens, ttft_ms=None if first is None else round(first * 1000, 1),
+               wall_ms=round(wall * 1000, 1))
+    if m:
+        row.update(prompt_tokens=int(m.group(1)), reused=int(m.group(2)), fresh=int(m.group(3)),
+                   prefill_ms=int(m.group(4)), prefill_tps=float(m.group(5)),
+                   decode_ms=int(m.group(7)))
+    if p:
+        row.update(park_tokens=int(p.group(1)), park_ms=float(p.group(2)))
+    if kv:
+        row.update(kv_hit_pct=float(kv.group(1)), kv_mib_from_ram=float(kv.group(3)))
+    # what TTFT pays beyond the engine's own prompt read (server, queue, first-token setup).
+    # The park copy is written AFTER the answer in both modes, so it is not subtracted here.
+    if row.get("ttft_ms") is not None and "prefill_ms" in row:
+        row["server_ms"] = round(row["ttft_ms"] - row["prefill_ms"], 1)
+    return row
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--port", type=int, default=8097)
+    ap.add_argument("--log", required=True)
+    ap.add_argument("--out", required=True)
+    ap.add_argument("--label", default="arm")
+    ap.add_argument("--samples", type=int, default=2)
+    ap.add_argument("--sizes", default="", help="comma-separated ladder override, e.g. 75")
+    args = ap.parse_args()
+    rows = []
+    # warm the server so the first request's cost is not the load
+    one(args.port, 32, f"warm-{random.random():.9f}", args.log)
+    sizes = [int(s) for s in args.sizes.split(",") if s.strip()] or SIZES
+    for size in sizes:
+        for i in range(args.samples):
+            salt = f"probe {args.label} {size} {i} {random.random():.9f}"
+            r = one(args.port, size, salt, args.log)
+            r["label"] = args.label
+            rows.append(r)
+            print(f"  {json.dumps(r)}", flush=True)
+    json.dump(rows, open(args.out, "w"), indent=1)
+    print(f"wrote {args.out}: {len(rows)} samples")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/dirigo-results/0.2.13-port/intercept/strata-intercept-sweep.sh b/dirigo-results/0.2.13-port/intercept/strata-intercept-sweep.sh
new file mode 100755
index 00000000000..1b8437086c0
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/strata-intercept-sweep.sh
@@ -0,0 +1,50 @@
+#!/usr/bin/env bash
+# strata-intercept-sweep.sh - price a request's fixed cost, with and without the fork's cache routing.
+#
+# Two card windows on the 0.2.13 tree, same engine, same model:
+#   on      the deployed config (conv_route: on)  -> single-shot traffic runs with the prompt cache OFF
+#   shadow  conv_route: shadow (classify and log, never switch) -> every request parks, like stock
+#
+# Each window runs ~/bin/strata-intercept-probe.py over a 4..2048-token ladder of unique prompts. The probe reads
+# the engine's own "prompt N tokens = ... read in X ms" and "conversation cache: parked ... in Y ms" lines, so
+# TTFT splits into prompt read + park + everything else.
+#
+# UNDO: rm this file and ~/strata-intercept-sweep
+set -uo pipefail
+S=/media/bob/3d651e2c-e9a4-4758-ba77-863725fe3731
+TREE=$S/strata-0.2.13
+BASE=$TREE/strata-coder-iq1_m-sycl-tuned.json
+OUT=$HOME/strata-intercept-sweep
+PROBE=$HOME/bin/strata-intercept-probe.py
+mkdir -p "$OUT"
+
+variant() {   # name key value
+    python3 - "$BASE" "$OUT/conf-$1.json" "$2" "$3" <<'PY'
+import json, sys
+src, dst, key, val = sys.argv[1:5]
+d = json.load(open(src))
+try: val = json.loads(val)
+except Exception: pass
+d[key] = val
+json.dump(d, open(dst, "w"), indent=1)
+print(dst, key, "=", d[key])
+PY
+}
+
+arm() {   # tag
+    local tag=$1 conf="$OUT/conf-$1.json"
+    echo "[$(date +%H:%M:%S)] ===== intercept-$tag"
+    ~/bin/strata-card-window.sh "$TREE" 8097 --conf "$conf" -- bash -c "
+        EL=\$(python3 -c \"import json;print(json.load(open('$conf'))['log'])\")
+        curl -s -m 5 http://127.0.0.1:\$PORT/v1/status | python3 -c 'import json,sys;d=json.load(sys.stdin);print(\"  engine:\",d[\"engine\"])'
+        /usr/bin/python3 $PROBE --port \$PORT --log \"\$EL\" --out $OUT/probe-$tag.json --label $tag
+    " 2>&1 | grep -E '=====|engine:|^\s+\{|wrote|restore'
+}
+
+variant on conv_route on
+variant shadow conv_route shadow
+
+for tag in on shadow; do
+    arm "$tag"
+done
+echo "[$(date +%H:%M:%S)] intercept sweep done"
diff --git a/dirigo-results/0.2.13-port/intercept/summary.json b/dirigo-results/0.2.13-port/intercept/summary.json
new file mode 100644
index 00000000000..9d5eac26a6a
--- /dev/null
+++ b/dirigo-results/0.2.13-port/intercept/summary.json
@@ -0,0 +1,97 @@
+{
+ "on": {
+  "ttft_ms": {
+   "intercept_ms": 242.8,
+   "ms_per_token": 0.772
+  },
+  "prefill_ms": {
+   "intercept_ms": 222.4,
+   "ms_per_token": 0.766
+  },
+  "server_ms": [
+   21.4,
+   21.8,
+   21.9,
+   22.3,
+   22.8,
+   22.8,
+   24.5,
+   24.8,
+   27.2,
+   32.0
+  ],
+  "park_ms": [
+   98.1,
+   101.8,
+   101.9
+  ]
+ },
+ "shadow": {
+  "ttft_ms": {
+   "intercept_ms": 455.0,
+   "ms_per_token": 0.79
+  },
+  "prefill_ms": {
+   "intercept_ms": 434.1,
+   "ms_per_token": 0.788
+  },
+  "server_ms": [
+   20.7,
+   21.4,
+   21.7,
+   21.4,
+   21.9,
+   21.8,
+   23.0,
+   22.1,
+   23.7,
+   24.2
+  ],
+  "park_ms": [
+   101.7,
+   104.6,
+   100.6,
+   102.1,
+   101.3,
+   101.8,
+   102.3,
+   101.7,
+   100.3,
+   99.9,
+   101.0,
+   102.6,
+   101.2,
+   100.5,
+   100.7,
+   102.8,
+   101.2,
+   102.1,
+   104.3,
+   110.6
+  ]
+ },
+ "base": {
+  "n": 6,
+  "prefill_median_ms": 381.5,
+  "prefill_min_max": [
+   327,
+   537
+  ]
+ },
+ "nocache": {
+  "n": 6,
+  "prefill_median_ms": 324.0,
+  "prefill_min_max": [
+   323,
+   339
+  ]
+ },
+ "base2": {
+  "n": 6,
+  "prefill_median_ms": 382.0,
+  "prefill_min_max": [
+   325,
+   537
+  ]
+ }
+}
\ No newline at end of file
diff --git a/docs/PERF-NEXT.md b/docs/PERF-NEXT.md
index 14028f52985..71e2235f970 100644
--- a/docs/PERF-NEXT.md
+++ b/docs/PERF-NEXT.md
@@ -36,6 +36,7 @@ important as the open one.
 | `--ple-row-cache` 4 M / 16 M rows, `--ple-inflight` 1024 | 4 M rows and inflight 1024 are **flat** (inside the 0.2–3.6 % drift band); 16 M rows (1.44 GB) costs **−8 % on fresh 4.2k, −8…−12 % on 8.8k, −17 % on a 4.4k follow-up, −10 % on 30k**. The 94 MB default and inflight 256 stand — [the sweep](../dirigo-results/0.2.13-port/ple-sweep/) |
 | `spec_window` policy | recovered only half the retrieval loss (+1.3…4.8 % vs +3.6…7.5 %) — left off |
 | DFlash2 / DSpark drafters | **CUDA/HIP only**; the SYCL delta in 0.2.12 is 3 lines in `prefill.cpp` |
+| `--spec 0` to remove drafting entirely | the engine refuses it on a native IQ pack: *"it needs `--native` SHARD1, `--spec T (T >= 2)` and `--prefill` CHUNK"* — so drafting cannot be excluded this way |
 
 ## 3. Where the remaining time is
 
@@ -48,13 +49,14 @@ fixed term.
 
 | # | idea | evidence it could pay | cost | first test |
 | --- | --- | --- | --- | --- |
-| 1 | ship `conv-route-batch-after` 1–2 | measured **−21 % median TTFT** (0.371 → 0.293 s), summed −29 %, steady state at request 2 instead of 5; the default is still 4 | a flag + a probe | a chat-after-single-shot shape to price the flip-back |
-| 2 | attribute the ~330 ms fixed per-request cost | nobody has split it on SYCL: slot init, expert warm-up for the first layers, KV allocation. The PLE read path is now excluded by measurement (§5) | an hour, no code | `bench_prefill` at 8/16/32/64/128/256 tokens |
-| 3 | residency-biased router on SYCL | `route_prior.cu` is CUDA-only; here **12,288 of 24,576** experts are resident, so half of every token's experts cross PCIe. Upstream measured +15 % / +37 % on a 12 GB NVIDIA card | engine port, days | decode lanes, 5 interleaved pairs |
-| 4 | prefill chunk at 4k/8k/12k with intervals | 16k/32k were tested, but 8,192 was never shown to be the knee | ~20 min | paired, three prompt sizes, per-prompt medians |
-| 5 | CPU pool workers and affinity on the 7700X | 15 workers is a 16-thread default; p-core pinning was **+27 %** decode on the 13700K | 15 min | workers 8/12/15/16, `--pool-affinity` on/off |
-| 6 | dispatch overhead on SYCL | the Vulkan analysis found ~88 % of per-dispatch cost is fixed; nobody has counted dispatches per token on SYCL | a day | `unitrace` kernel count per token, then fuse the five most frequent small ops |
-| 7 | DFlash2 block drafter on SYCL | upstream's headline decode lever; 7 drafts per pass | weeks | only after 1–3 |
+| 1 | ship `conv-route-batch-after` 1–2 | measured **−21 % median TTFT** (0.371 → 0.293 s); §6 prices what it saves at **+212 ms per request** while the cache is on, and at `batch_after 4` the first three fresh requests of a run still pay it | a flag + a probe | a chat-after-single-shot shape to price the flip-back |
+| 2 | engine-side per-request timer | §6 leaves **~165 ms** inside prefill that no configuration lever explains (PLE reads excluded, cache slots ~58 ms of it); naming it needs a phase timer in the request path | code, a day | timestamp slot setup / first-layer expert reach / KV alloc per request |
+| 3 | a one-shot mode that also drops the cache slots | §6: `--prompt-cache 0` is **58 ms cheaper per request** than the router's cache-off path at 75 tokens, and far more stable (324 ms against 382, spread 323–339 against 327–537). The setting is restart-class today | engine work to make it per-request | the same ladder with a 0-slot variant |
+| 4 | residency-biased router on SYCL | `route_prior.cu` is CUDA-only; here **12,288 of 24,576** experts are resident, so half of every token's experts cross PCIe. Upstream measured +15 % / +37 % on a 12 GB NVIDIA card | engine port, days | decode lanes, 5 interleaved pairs |
+| 5 | prefill chunk at 4k/8k/12k with intervals | 16k/32k were tested, but 8,192 was never shown to be the knee | ~20 min | paired, three prompt sizes, per-prompt medians |
+| 6 | CPU pool workers and affinity on the 7700X | 15 workers is a 16-thread default; p-core pinning was **+27 %** decode on the 13700K | 15 min | workers 8/12/15/16, `--pool-affinity` on/off |
+| 7 | dispatch overhead on SYCL | the Vulkan analysis found ~88 % of per-dispatch cost is fixed; nobody has counted dispatches per token on SYCL | a day | `unitrace` kernel count per token, then fuse the five most frequent small ops |
+| 8 | DFlash2 block drafter on SYCL | upstream's headline decode lever; 7 drafts per pass | weeks | only after 1–3 |
 
 ## 5. The PLE sweep, run 2026-11-14 — closed
 
@@ -81,7 +83,36 @@ reuse the KV cache instead of re-reading rows.
 Artifacts: [`0.2.13-port/ple-sweep/`](../dirigo-results/0.2.13-port/ple-sweep/) — the six configs, the per-leg JSON, the log with each
 window's restore line, `summary.json`, and the script.
 
-## 6. Method rules these numbers were produced under
+## 6. The fixed per-request cost, measured 2026-11-14
+
+A ladder of unique prompts (74 … 1,608 real tokens), two arms — the deployed `conv_route: on` and `shadow` (cache
+on, like stock) — client TTFT from the first content delta, and the engine's own `prompt … read in X ms` and
+`conversation cache: parked … in Y ms` lines for the split:
+
+| arm | TTFT fit | prefill fit (engine) | server (TTFT − prefill) | park |
+| --- | --- | --- | --- | --- |
+| `conv_route: on` (cache off for one-shot) | **243 ms + 0.77 ms/token** | **222 ms + 0.77 ms/token** | 21–32 ms | ~100 ms on the first 3 requests only, then none |
+| `shadow` (cache on, like stock) | **455 ms + 0.79 ms/token** | **434 ms + 0.79 ms/token** | 21–24 ms | ~101 ms on every request |
+
+1. **Nothing hides outside the engine's own accounting.** TTFT is prefill plus ~21 ms of server, in both modes; the
+   unaccounted term that earlier notes called "450–500 ms" is not overhead, it is prefill.
+2. **The cache-on state costs +212 ms per request** (intercept 434 against 222) — the prize the router takes. At
+   `batch_after 4` a run's first three fresh requests still pay it: 484 / 485 / 402 ms at 74 / 74 / 75 tokens against
+   290 ms once the cache is off.
+3. **Parking is ~101 ms and happens after the answer**, in both modes; it is not part of TTFT, and with the cache off
+   it stops entirely.
+4. **The cache slots cost something even when unused**: `--prompt-cache 0` measured 324 ms median at 75 tokens
+   (323–339) against 382 ms on the router's cache-off path (327–537), the same prompts and the same build, with a
+   repeated default arm reproducing 382 ms exactly.
+5. **The draft layer could not be switched off to test its share**: `--spec 0` is refused on a native IQ pack, which
+   needs `--spec T (T >= 2)`.
+6. **What is left is ~165 ms** inside prefill after the cache slots' ~58 ms, unexplained by any configuration lever
+   tried so far (PLE row reads are excluded by §5). Naming it needs a phase timer inside the request path.
+
+Artifacts: [`0.2.13-port/intercept/`](../dirigo-results/0.2.13-port/intercept/) — both ladders, the three
+cause-sweep arms, `summary.json`, the probe and both drivers, and the logs with every restore line.
+
+## 7. Method rules these numbers were produced under
 
 - Paired or interleaved arms; median of repeats; the box's run-to-run spread stated next to the effect.
 - One prompt per configuration is not a measurement: every lane here has several requests, and a single-prompt