loudink-v1.1 1.1.1: P0 lean/mid prompts + correction micro (writing neural 100%)

README.md

@@ -15,7 +15,7 @@ tags:
 library_name: mlx-lm
 ---
 
-# loudink-v1.1 v1.1.0 (canary)
+# loudink-v1.1 v1.1.1 (canary)
 
 On-device dual-stack for **LoudInk / Flowcast** on Apple Silicon, plus **user skills**:
 
@@ -30,7 +30,7 @@ This package is the **v1.1 canary** (PR-A→E). Prefer dual-package client routi
 
 | Slice | Shipped | Neural | Latency |
 |-------|---------|--------|---------|
-| Writing daily | **100%** | **92.0%** | p50 **171 ms** |
+| Writing daily | **100%** | **100.0%** | p50 **184 ms** |
 | Computer daily | **100%** | — | mostly fast-path |
 | Multi-step | **100%** | — | mostly fast-path |
 | **Daily-Mac overall (n=69)** | **100%** | — | — |

SHA256SUMS

@@ -1,19 +1,19 @@
-c20806b8e4dfc28e162efd1bc3d4cae42e8e5224aaeda7722562bf2b9539acf5  README.md
-cf88d3c382ec0deaa9a89bb682da29bbf8822191c97d5da784de38bd3e817344  STATUS.md
+e79d4a3ff4bb2a05b7dcc13afd816abcea782913f0a7552b89e1d6083388e54b  README.md
+9cb14787b48df0d0271729464ea2b032562c941d54bb701d3f6c154201b01157  STATUS.md
 887cad558e8fde9d4c357f2cdae4507e4e614b1f0dbad190d1fa9d21714ad92f  V1_1_COMPLETE.md
 3965991f7c62bfb29001f9b19f704deefbc890dc85649e4d72fd57243ab317f0  client_contract.md
-4710b258140647aee5399ce3158da7f1f29bea5cb8bb6ed3541be122d5e53064  inference_config.json
+7b0558b82dcc6016ba38f17ecee0073e0f6a03c9f399ac133f6ae7925a82cf6c  inference_config.json
 4159a5a3cd7ed71c4da7a6712b2ec5049d7c1ec2e84cbe538178136e5e99d4f3  ir/adapter_config.json
 868d11b4360748cb3c493e26e2ee4cf4eef937eb0c9755988c364a741e5d3a82  ir/adapters.safetensors
 69abbc73e77287e6646bbc323a3fd6869218124b12512cd262a40d0b034ddda2  ir_honesty_fallback/adapter_config.json
 02b521c2348477cf03f667c7b500f0be82bfb937277806e3f4e562b9593a1bbe  ir_honesty_fallback/adapters.safetensors
-4290d51b5ad2ffc89ff22086d2d6aede91485ea4c4f1ecbe2ecd122520a05e4a  manifest.json
-b3b4a8eaa9bef08c32c608f538c1e176acc41f49d18d66aa7c36673123746af9  product_scorecard.json
+62aae5e5d5e8e1c1a2914b9d3e11427e8636826981bc69fb67522fdbefb72c4f  manifest.json
+314f9d8f56364e2726458768848aab60ba1dda02ad28b4823edb142cc97c4719  product_scorecard.json
 8856981bbea72862e2d44a6d5f4de9e540e1793ed1674fcc0dd7ace2364be5f0  router.json
 12db4143f37da30858061eb64bf3fb645d15e195bb18b9274f3620640eb63f37  skill_store/index.json
-f18e2b92a8264142df82780fa0757c5c6aa60569e364a824a1e2029787a81378  skill_store/skills/invoice_chase.json
-55715a59fbdc06b0656d20ace72181dda4142d6c8edbe791b3752c71f4510566  skill_store/skills/morning_focus.json
-a025338ad42f4fee0725f6e8bbd5bf2384a42973ab847838bd9a54526f64ac22  skill_store/skills/standup_post.json
+a7f482c3bae12b95d8fae950642701280e881796de2e18b8044e92d9f883a7dd  skill_store/skills/invoice_chase.json
+bcadd34579d6b57d4d584158d58aba534b072ac13603564c824c36addd4ad8de  skill_store/skills/morning_focus.json
+d3313b56bbc1571e2af3ee7e1c3be8398d37b657fa90a351abed811d9f95ee26  skill_store/skills/standup_post.json
 ceb709ccf7a40ea608c5cc74c247071bbcb0980017f1811f7c55604b4241ab7c  writer/promoted_core_100.safetensors
-734e75c575696d88ec67283d23f8dbeb87a03c3c0e45e0da6b4cfa531036cb4e  writer_unified/adapter_config.json
-2db8a19242ab84fb4b64820528509c3eb842c998f53247e6cf13999b0651b955  writer_unified/adapters.safetensors
+4a1802e622de9412ca8f675639ecc6f7cc0f91ba3ed7c09662f1d8b7b6ec8c9a  writer_unified/adapter_config.json
+fce889b02639a5ad3b99b8184a714a84e84cb469a68084b142f242aa7ce6a8a9  writer_unified/adapters.safetensors

STATUS.md

@@ -7,7 +7,7 @@ Last updated: 2026-07-08
 | Line | Repo | Role |
 |------|------|------|
 | **Stable GA** | [nsalerni/loudink-v1](https://huggingface.co/nsalerni/loudink-v1) | **0.4.0** freeze |
-| **Canary** | [nsalerni/loudink-v1.1](https://huggingface.co/nsalerni/loudink-v1.1) | **1.1.0** skills + expanded daily-mac |
+| **Canary** | [nsalerni/loudink-v1.1](https://huggingface.co/nsalerni/loudink-v1.1) | **1.1.1** skills + P0 speed/accuracy |
 
 Rebuild canary:
 
@@ -33,19 +33,26 @@ python -m gemmaflow_tune.cli.publish_sota --model loudink-v1.1 \
 |--------|-----|--------|
 | Skill invoke (n=19) | ≥95% named / ≥85% paraphrase | **100%** shipped |
 | Daily-Mac overall (n=69) | no drop vs 0.4 | **100%** shipped |
-| Writing neural | hold ≥80% | **92%** |
-| Writing p50 | ≤250 ms | **171 ms** |
+| Writing neural | hold ≥80% | **100%** (P0) |
+| Writing p50 | ≤250 ms | **184 ms** |
 | Computer / multi-step | hold 100% | **100%** |
 | IR honesty (n=112) | ≥45% neural | **46.4%** (v0.4 lineage) |
 
-### Daily-Mac scorecard (v1.1 expanded)
+### Daily-Mac scorecard (v1.1.1 P0)
 
 | Slice | Shipped | Neural | p50 |
 |-------|---------|--------|-----|
-| Writing (n=25) | **100%** | **92%** | **171 ms** |
+| Writing (n=25) | **100%** | **100%** | **184 ms** |
 | Computer (n=44) | **100%** | — | ~0 ms FP |
 | Multi-step (n=25) | **100%** | — | ~0 ms FP |
-| **Overall (n=69)** | **100%** | — | mean ~60 ms |
+| **Overall (n=69)** | **100%** | — | mean ~63 ms |
+
+### P0 speed/accuracy (1.1.1)
+
+- Lean + mid dictation prompts (cut full ~260-token prefill on short/mid cases)
+- `writing_daily` in dictation early-stop suites + **coverage-gated** sanitized stop (no multi-clause truncation)
+- Wait-no / money / time correction micro SFT → residual neural fails closed
+- Writer: `artifacts/sft_loudink_v1_1_correction_micro/adapters`
 
 Scorecard: `artifacts/loudink_v1/product_scorecard.json`  
 Skill bench: `artifacts/benchmarks/loudink_v1_1/skill_invoke/`
@@ -53,7 +60,7 @@ Skill bench: `artifacts/benchmarks/loudink_v1_1/skill_invoke/`
 ## Production stack (v1.1 canary)
 
 ```
-writer:       artifacts/sft_loudink_v1_rft_daily/adapters
+writer:       artifacts/sft_loudink_v1_1_correction_micro/adapters
 polish:       structural (+ daily rules)
 ir:           artifacts/sft_loudink_v1_1_skill_invoke_ir/adapters
 ir_fallback:  artifacts/sft_loudink_v1_ir_honesty/adapters

inference_config.json

@@ -2,7 +2,7 @@
   "model_name": "loudink-v1.1",
   "model_tag": "loudink-v1.1",
   "product": "loudink-v1.1",
-  "version": "1.1.0",
+  "version": "1.1.1",
   "runner_kind": "loudink_v1",
   "hf_repo": "nsalerni/loudink-v1.1",
   "predecessor": "nsalerni/loudink-v1",
@@ -38,8 +38,8 @@
   "metrics_v11": {
     "daily_mac_overall_shipped": 1.0,
     "daily_mac_writing_shipped": 1.0,
-    "daily_mac_writing_neural": 0.92,
-    "daily_mac_writing_p50_ms": 171,
+    "daily_mac_writing_neural": 1.0,
+    "daily_mac_writing_p50_ms": 184,
     "daily_mac_computer_shipped": 1.0,
     "daily_mac_multi_step_shipped": 1.0,
     "daily_mac_n": 69,

manifest.json

@@ -1,6 +1,6 @@
 {
   "product": "loudink-v1.1",
-  "version": "1.1.0",
+  "version": "1.1.1",
   "pipeline_namespace": "loudink_v1_1",
   "runner_kind": "loudink_v1",
   "gate_mode": "warn",
@@ -50,8 +50,8 @@
   "metrics_v11": {
     "daily_mac_overall_shipped": 1.0,
     "daily_mac_writing_shipped": 1.0,
-    "daily_mac_writing_neural": 0.92,
-    "daily_mac_writing_p50_ms": 171,
+    "daily_mac_writing_neural": 1.0,
+    "daily_mac_writing_p50_ms": 184,
     "daily_mac_computer_shipped": 1.0,
     "daily_mac_multi_step_shipped": 1.0,
     "daily_mac_n": 69,
@@ -59,5 +59,5 @@
     "skill_invoke_n": 19,
     "ir_honesty_neural_n112": 0.4643
   },
-  "notes": "1.1.0 canary: PR-A skill store/recorder/compiler; PR-B paraphrase invoke + IR labels; PR-C slot fill; PR-D expanded daily-mac (Focus/DND, Reminders, reply, tabs, Notion/Linear, writing reply/code); PR-E vision fallback scaffold (disabled by default). Holds v0.4 bars and exceeds writing neural on expanded suite."
+  "notes": "1.1.1 canary P0: lean/mid dictation prompts + writing_daily early-stop coverage fix + wait-no correction micro SFT. Writing neural 100% (was 92%), overall 69/69, p50 writing ~184ms. Skills PR-A\u2013E unchanged."
 }

product_scorecard.json

@@ -2,7 +2,7 @@
   "reports": [
     {
       "source": "artifacts/benchmarks/loudink_v1/daily_mac/loudink_v1_daily_mac_measurement.json",
-      "label": "loudink-v1 v0.4 daily-mac product",
+      "label": "loudink-v1.1 daily-mac product (n=69)",
       "writing": {
         "shipped": {
           "total": 25,
@@ -11,19 +11,19 @@
         },
         "neural": {
           "total": 25,
-          "passed": 23,
-          "accuracy": 0.92,
+          "passed": 25,
+          "accuracy": 1.0,
           "missing": 0
         },
-        "polish_lift": 0.07999999999999996,
+        "polish_lift": 0.0,
         "latency": {
           "n": 25,
-          "mean": 166.6401715839654,
-          "p50": 171.05649993754923,
-          "p95": 332.83937512896955,
-          "max": 383.80437484011054,
-          "model_invoked_p50": 176.41600011847913,
-          "model_invoked_p95": 332.83937512896955,
+          "mean": 175.00264003708958,
+          "p50": 183.71283309534192,
+          "p95": 322.333584073931,
+          "max": 329.85158381052315,
+          "model_invoked_p50": 196.95341680198908,
+          "model_invoked_p95": 322.333584073931,
           "model_invoked_n": 20,
           "fast_path_rate": 0.19999999999999996
         }
@@ -79,12 +79,12 @@
         },
         "latency": {
           "n": 69,
-          "mean": 60.40875782027732,
+          "mean": 63.43863769459767,
           "p50": 0.05,
-          "p95": 323.94058303907514,
-          "max": 383.80437484011054,
-          "model_invoked_p50": 176.41600011847913,
-          "model_invoked_p95": 332.83937512896955,
+          "p95": 291.4309168700129,
+          "max": 329.85158381052315,
+          "model_invoked_p50": 196.95341680198908,
+          "model_invoked_p95": 322.333584073931,
           "model_invoked_n": 20,
           "fast_path_rate": 0.7101449275362319
         }

publish_report.json

@@ -1,7 +1,7 @@
 {
   "hf_repo": "nsalerni/loudink-v1.1",
   "output_dir": "artifacts/hf_publish/loudink-v1.1",
-  "version": "1.1.0",
+  "version": "1.1.1",
   "measured_adapter_bundle_mib": 70.14,
   "files": [
     "README.md",

skill_store/skills/invoice_chase.json

@@ -6,7 +6,7 @@
     "chase invoices"
   ],
   "source": "user_recording",
-  "created_at": "2026-07-09T05:42:10+00:00",
+  "created_at": "2026-07-09T05:57:44+00:00",
   "skill_version": 1,
   "schema_version": "1.1.0-pra",
   "steps": [

skill_store/skills/morning_focus.json

@@ -6,7 +6,7 @@
     "morning focus workflow"
   ],
   "source": "user_recording",
-  "created_at": "2026-07-09T05:42:10+00:00",
+  "created_at": "2026-07-09T05:57:44+00:00",
   "skill_version": 1,
   "schema_version": "1.1.0-pra",
   "steps": [

skill_store/skills/standup_post.json

@@ -6,7 +6,7 @@
     "daily standup slack"
   ],
   "source": "user_recording",
-  "created_at": "2026-07-09T05:42:10+00:00",
+  "created_at": "2026-07-09T05:57:44+00:00",
   "skill_version": 1,
   "schema_version": "1.1.0-pra",
   "steps": [

writer_unified/adapter_config.json

@@ -1,15 +1,15 @@
 {
-    "adapter_path": "artifacts/sft_loudink_v1_rft_daily/adapters",
+    "adapter_path": "artifacts/sft_loudink_v1_1_correction_micro/adapters",
     "batch_size": 1,
     "clear_cache_threshold": 0,
     "config": null,
-    "data": "data/sft_loudink_v1_rft_daily_micro",
+    "data": "data/sft_loudink_v1_1_correction_micro",
     "fine_tune_type": "lora",
     "gpu_mode": "shared",
     "grad_accumulation_steps": 8,
     "grad_checkpoint": true,
-    "iters": 50,
-    "learning_rate": 5e-07,
+    "iters": 60,
+    "learning_rate": 6e-07,
     "lora_parameters": {
         "dropout": 0.05,
         "keys": [
@@ -45,9 +45,9 @@
     "project_name": null,
     "report_to": null,
     "resume_adapter_file": "artifacts/sft_loudink_v1_rft_daily/adapters/adapters.safetensors",
-    "save_every": 25,
-    "seed": 35,
-    "steps_per_eval": 25,
+    "save_every": 20,
+    "seed": 41,
+    "steps_per_eval": 20,
     "steps_per_report": 10,
     "test": false,
     "test_batches": 500,

writer_unified/adapters.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2db8a19242ab84fb4b64820528509c3eb842c998f53247e6cf13999b0651b955
+oid sha256:fce889b02639a5ad3b99b8184a714a84e84cb469a68084b142f242aa7ce6a8a9
 size 22292714