Update embeddinggemma-300m-memory-ft-v2 documentation

README.md

@@ -129,9 +129,9 @@ passage, which does not prove none exists.
 
 At 50, v2 finds useful evidence for **99.03%** of these queries and retrieves
 **57.92%** of their known useful passages, versus 97.73% and 41.12% upstream.
-The hybrid pool below has lower coverage on the same subset, despite stronger
-early precision after reranking. Adding BM25 and capping the fused list at 50
-can displace dense candidates.
+The capped hybrid pool below covers fewer queries at 50. Adding BM25 and
+capping the fused list can displace dense candidates, while reranking can
+improve early precision.
 
 Grades 2–3 count as useful. nDCG uses gains 0, 1, 3 and 7 against the same
 reviewed reference for both models. Rankings from query forms are combined
@@ -178,44 +178,47 @@ larger quality cost. These byte counts exclude index overhead and compression;
 shorter vectors do not reduce encoder inference cost. Daecore uses 768, and
 smaller widths have not been qualified through the full hybrid pipeline.
 
-### Full pipeline: Gemma v2 + BM25 + Ettin
+### Full pipeline: the effect of Gemma v2
 
-This replay measures semantic and lexical search, reciprocal-rank fusion, then
-Ettin reranking over **82,719 passages**. The table uses the same **925 queries
-with known useful evidence** as Gemma's dense comparison, selected from the
-970-query source panel. A query remains included when this pipeline misses its
-answer. Both cards share this CUDA FP16 table; it is not either model's
-standalone score.
+Only the embedder changes: **upstream Gemma → Daecore Gemma v2**. BM25,
+reciprocal-rank fusion and the fine-tuned Ettin reranker stay fixed. Both
+pipelines search the same **82,719 passages**. Each cell reads before →
+**after**, so the change measures Gemma's contribution inside hybrid retrieval.
 
-**Reference:** 122,231 graded query–passage pairs for these 925 queries, within
-the shared 127,665-pair top-50 reference (September 2026). Abstentions are excluded.
+**Reference:** the same **925 known-answerable queries**, frozen before this
+comparison · **122,783 graded query–passage pairs** · October 2026. Retrieval
+misses remain included. Both cards share the same fully fine-tuned endpoint.
 
 | Return depth | Hit | Precision | Known-positive recall | nDCG |
 |---|---:|---:|---:|---:|
-| 1 | 75.87% | 75.87% | 2.12% | 0.6513 |
-| 3 | 89.51% | 75.07% | 5.89% | 0.6423 |
-| 5 | 92.76% | 73.71% | 9.35% | 0.6376 |
-| 10 | 94.27% | 70.46% | 17.04% | 0.6378 |
-| 20 | 95.46% | 64.81% | 29.72% | 0.6491 |
-| 50 | 97.62% | 42.40% | 45.98% | 0.6130 |
-
-**The hybrid candidate ceiling is 97.62% Hit@50 and 45.98% known-positive
-recall@50** on this shared answerable subset. These coverage measures describe
-the actual 50-candidate pool; reranking cannot change them. Daecore returns a
-selected prefix of 3–20, so fixed-depth scores do not evaluate the selector's
-choice.
-
-Grades 2–3 count as useful. At depth 1, both providers use the same **920
-queries**; five are omitted because a first result was ungradable. Depths
-3–50 use all 925. Precision pools judged positions. Abstentions remove 52
-positions at rank 20 and 93 at rank 50 per provider, without drawing in deeper
-passages. Known-positive recall counts judged passages, not every useful
-passage in the corpus.
-
-The [evaluation companion](evaluation/README.md) retains
-coverage across all 970 source queries and explains the shared reference,
-exclusions and CUDA/Vulkan comparison. The [pipeline overview](https://huggingface.co/Daecore)
-explains retrieval and selection.
+| 1 | 73.99% → **75.95%** | 73.99% → **75.95%** | 2.01% → **2.12%** | 0.6295 → **0.6527** |
+| 3 | 86.81% → **89.41%** | 71.50% → **75.11%** | 5.49% → **5.89%** | 0.6121 → **0.6424** |
+| 5 | 89.95% → **92.76%** | 70.10% → **73.67%** | 8.53% → **9.32%** | 0.6054 → **0.6373** |
+| 10 | 93.08% → **94.27%** | 65.20% → **70.49%** | 15.25% → **16.99%** | 0.5959 → **0.6377** |
+| 20 | 95.68% → **95.46%** | 56.86% → **64.80%** | 25.18% → **29.63%** | 0.5873 → **0.6489** |
+| 50 | 96.97% → **97.62%** | 35.45% → **42.40%** | 36.85% → **45.83%** | 0.5374 → **0.6125** |
+
+At three results, useful-passage precision changes from **71.50% to
+75.11%**; nDCG@10 changes from **0.5959 to 0.6377**.
+At 50, the hybrid pool's known-positive recall changes from **36.85%
+to 45.83%**. This measures evidence available to Ettin after
+fusion; reranking cannot recover passages outside those 50 slots.
+
+The gain is not uniform: Hit@20 falls from 95.68% to 95.46%, a difference of
+two queries, while precision, known-positive recall and nDCG improve at that depth.
+
+All three variants use matched **PyTorch FP16** reranking. Depth 1 uses **919 shared judged
+queries** across the three pipeline variants; depths 3–50 use all 925.
+Abstentions are excluded inside the original cutoff without backfill.
+Grades 2–3 count as useful. Precision pools retained positions; recall counts
+known useful passages.
+The expanded reference and matched runtime distinguish this table from the
+earlier CUDA/Vulkan serving check. Daecore returns 3–20 results; these fixed
+cutoffs do not evaluate its selector.
+
+The [evaluation companion](evaluation/README.md) provides
+the paired records, methods and separate serving checks. The
+[pipeline overview](https://huggingface.co/Daecore) explains retrieval and selection.
 
 ## Training data and objective
 

evaluation/README.md

@@ -11,7 +11,8 @@ private evaluations.
 | `retriever.json` | Upstream and v2 reviewed top-50 dense rankings for all 970 queries over 82,719 passages |
 | `dimensions.json` | V2 public nDCG@10 and private top-10 graded rankings at four embedding widths |
 | `promotion.json` | Ranked grades for the predecessor-to-v2 pipeline comparison and public nDCG@10 for upstream and v2 |
-| `serving.json` | The 970-query Gemma v2 + BM25 + Ettin replay used by both model cards, including CUDA/Vulkan results |
+| `pipeline.json` | Matched upstream-to-fine-tuned component changes on the frozen 925-query hybrid panel |
+| `serving.json` | Separate 970-query CUDA/Vulkan replay with the September reference |
 | `*-summary.json` | Summaries that `metrics.py` reproduces |
 | `serving-qualification.json` | CPU, CUDA and Vulkan serving checks for this model |
 | `metrics.py` | Metric code |
@@ -26,7 +27,7 @@ python metrics.py --directory .
 python figures.py --output ../figures
 ```
 
-No packages or network access are needed. Each result matches its corresponding summary. The `answerable` section of the dense and serving summaries supplies the card tables; `private.answerable` supplies the private width column. The outer `models` sections retain all-query results. This recomputes metrics from saved results; running private retrieval again would require the private corpus.
+No packages or network access are needed. Each result matches its corresponding summary. The dense card table uses `retriever-summary.json`'s `answerable` section; the pipeline table uses `pipeline-summary.json`, comparing `upstream_gemma` with `finetuned`. `private.answerable` supplies the private width column. The dense and serving records retain all-query results in their outer `models` sections. This recomputes metrics from saved results; running private retrieval again would require the private corpus.
 
 ## Dense retrieval
 
@@ -70,8 +71,9 @@ reference contained 87,434 graded pairs.
 
 The subsequent decision to headline the same 925 known-answerable queries
 changes the averaging population, not the labels or rankings. On that subset,
-nDCG@10 is **0.4418 upstream and 0.5621 for v2**. The shared reference supports
-dense, smaller-width and current pipeline results. The historical predecessor
+nDCG@10 is **0.4418 upstream and 0.5621 for v2**. The September reference supports
+dense, smaller-width and provider results. The new component comparison adds
+its own reviewed overlay, described below. The historical predecessor
 comparison below keeps its original reference and population.
 
 **Public data.**
@@ -96,25 +98,44 @@ shorten the encoder's forward pass.
 
 ## Effect in the retrieval pipeline
 
-The identical table on both cards comes from `serving.json`: Gemma v2 +
-BM25 + fusion + unchanged Ettin. It uses the **same 925 known-answerable
-queries and 122,231 graded pairs** as the dense table. CUDA FP16 and Vulkan
-FP32 use the same 50 candidates for each query. At 50, Hit is **97.62%** and
-known-positive recall is **45.98%**; reranking cannot change candidate
-coverage. Dense v2 reaches 99.03% and 57.92% on this subset. Fusion with a
-fixed 50-slot budget can displace dense candidates even while reranking
-improves early precision.
-
-Depths 3–50 use all 925 queries. Depth 1 uses 920 shared judged queries; five
-are omitted because a first result was ungradable. Each provider excludes 52
-positions at rank 20 and 93 at rank 50 without backfill. Fixed cutoffs are
-separate from the selector's choice of a 3–20-item prefix. Vulkan matches
-CUDA at Hit@5, Hit@10, Hit@20 and Hit@50; Hit@3 differs by one query and
-nDCG@10 by 0.0004.
+`pipeline.json` holds three matched arms. For this card, compare
+`upstream_gemma` with `finetuned`: only Gemma changes, while BM25, fusion and
+the fine-tuned Ettin remain fixed. The Ettin card uses the other baseline
+against the same fully fine-tuned endpoint.
+
+The comparison keeps the **925-query cohort frozen before scoring** and uses
+**122,783 graded query–passage pairs** for those queries. It adds 552 grades
+and 10 abstentions to their September reference of 122,231 pairs; old
+grades are unchanged. Two independent GPT-6.1 Sol judges and a fresh blind
+adjudicator resolved the new pairs, with 84 ordinal disagreements.
+The primary assistant reviewed 84 complete pairs, including every final quotation
+flag and abstention plus a stratified sample. These are model judgments,
+without a human reference panel.
+
+All three variants were scored anew in **PyTorch FP16**, using Transformers
+5.2.0, Sentence Transformers 5.5.1, 128 query tokens and 1,024 passage tokens.
+Gemma uses cached FP32 embeddings at 768 dimensions with matched input limits.
+The actual first stage uses the frozen query forms, dense retrieval, BM25,
+reciprocal-rank fusion with constant 60, exact-text deduplication and 50 slots.
+Before substituting upstream Gemma's vectors, replay reproduced all 970 saved
+v2 candidate lists exactly. Ties preserve candidate order.
+
+Depth 1 uses **919 common judged queries across all three arms**; depths
+3–50 retain all 925. Original prefixes are cut before abstentions are removed,
+with no backfill. Retrieval misses remain included. The reference supplies
+known-positive recall and ideal DCG; it is not exhaustive corpus labeling.
+Fixed cutoffs do not test the calibrated selector's choice of 3–20 results.
+The earlier ONNX provider replay retains its own runtime and September
+reference, so its absolute scores should not be substituted into this comparison.
+
+`serving.json` separately compares ONNX CUDA FP16 with Vulkan FP32. It keeps
+the September reference of 122,231 grades for the same 925 queries, with 920
+eligible at top 1. Both providers use identical candidate pools. This checks
+serving paths, not the effect of fine-tuning.
 
 ### All-query coverage
 
-The records preserve all 970 source queries. This audit view includes the 45
+The September dense and serving records preserve all 970 source queries. This audit view includes the 45
 without known useful evidence and is separate from the card's answerable-only
 quality table. Hybrid top 1 uses 965 queries after shared abstention
 exclusions; dense top 1 retains all 970. The counts below use the full

evaluation/metrics.py

@@ -267,10 +267,22 @@ def summarize_serving(data: dict) -> dict:
             'answerable': _summarize_ranked_grades(data, include_selected=True, answerable_only=True)}
 
 
+def summarize_pipeline(data: dict) -> dict:
+    """Matched component changes on a frozen, known-answerable query cohort."""
+    if type(data['corpus_passages']) is not int or data['corpus_passages'] <= 0:
+        raise ValueError('Pipeline comparison requires a positive corpus size')
+    output = _summarize_ranked_grades(data, include_selected=False)
+    if any(row['reference_grade_counts']['2'] + row['reference_grade_counts']['3'] == 0
+           for row in data['rows']):
+        raise ValueError('Pipeline cohort requires known useful evidence for every query')
+    return {**output, 'corpus_passages': data['corpus_passages'],
+            'reference_pairs': sum(sum(row['reference_grade_counts'].values()) for row in data['rows'])}
+
+
 SUMMARIZERS = {
     'classifier': summarize_classifier, 'reranker': summarize_reranker,
     'retriever': summarize_retriever, 'dimensions': summarize_dimensions,
-    'promotion': summarize_promotion, 'serving': summarize_serving,
+    'promotion': summarize_promotion, 'serving': summarize_serving, 'pipeline': summarize_pipeline,
 }
 
 
@@ -283,6 +295,7 @@ def summarize_record(name: str, data: dict) -> dict:
         'dimensions': {'id', 'dataset', 'ndcg@10'},
         'promotion': {'id', 'group', 'reference_grade_counts', 'ranked_grades', 'excluded', 'selected_depth'},
         'serving': {'id', 'group', 'reference_grade_counts', 'ranked_grades', 'excluded', 'selected_depth'},
+        'pipeline': {'id', 'group', 'reference_grade_counts', 'ranked_grades', 'excluded'},
     }[name]
     rows = data['rows']
     if not rows or any(set(row) != fields for row in rows):
@@ -336,7 +349,7 @@ def main() -> None:
     parser = argparse.ArgumentParser(description=__doc__)
     parser.add_argument("--directory", type=Path, default=Path(__file__).resolve().parent)
     parser.add_argument("--output", type=Path)
-    parser.add_argument("--model", choices=("classifier", "reranker", "retriever", "dimensions", "promotion", "serving"), help="Recompute one comparison; default: every record in this package")
+    parser.add_argument("--model", choices=tuple(SUMMARIZERS), help="Recompute one comparison; default: every record in this package")
     args = parser.parse_args()
     names = [args.model] if args.model else [
         name for name in SUMMARIZERS if (args.directory / f"{name}.json").is_file()

evaluation/pipeline-summary.json

@@ -0,0 +1,283 @@
+{
+  "queries": 925,
+  "models": {
+    "upstream_gemma": {
+      "1": {
+        "scored_queries": 919,
+        "excluded_queries": 6,
+        "hit": 0.7399347116430903,
+        "precision": 0.7399347116430903,
+        "macro_precision": 0.7399347116430903,
+        "ndcg": 0.6295144826156795,
+        "known_positive_recall": 0.02009148065967933,
+        "recall_queries": 919,
+        "useful": 680,
+        "retained": 919,
+        "excluded_positions": 0,
+        "mean_useful": 0.7399347116430903,
+        "mean_retained": 1.0
+      },
+      "3": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.8681081081081081,
+        "precision": 0.7149566473988439,
+        "macro_precision": 0.7153153153153153,
+        "ndcg": 0.6120552491562434,
+        "known_positive_recall": 0.05489711276806684,
+        "recall_queries": 925,
+        "useful": 1979,
+        "retained": 2768,
+        "excluded_positions": 7,
+        "mean_useful": 2.1394594594594594,
+        "mean_retained": 2.9924324324324325
+      },
+      "5": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.8994594594594595,
+        "precision": 0.7009527934170636,
+        "macro_precision": 0.7011351351351351,
+        "ndcg": 0.60538850988017,
+        "known_positive_recall": 0.08526196518888282,
+        "recall_queries": 925,
+        "useful": 3237,
+        "retained": 4618,
+        "excluded_positions": 7,
+        "mean_useful": 3.4994594594594592,
+        "mean_retained": 4.992432432432432
+      },
+      "10": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.9308108108108109,
+        "precision": 0.6519761775852734,
+        "macro_precision": 0.6521381381381381,
+        "ndcg": 0.5958896253275177,
+        "known_positive_recall": 0.15250767729827347,
+        "recall_queries": 925,
+        "useful": 6021,
+        "retained": 9235,
+        "excluded_positions": 15,
+        "mean_useful": 6.509189189189189,
+        "mean_retained": 9.983783783783784
+      },
+      "20": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.9567567567567568,
+        "precision": 0.5685743678596568,
+        "macro_precision": 0.5687161650814901,
+        "ndcg": 0.5873176954516321,
+        "known_positive_recall": 0.2518432399920873,
+        "recall_queries": 925,
+        "useful": 10501,
+        "retained": 18469,
+        "excluded_positions": 31,
+        "mean_useful": 11.352432432432433,
+        "mean_retained": 19.966486486486488
+      },
+      "50": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.9697297297297297,
+        "precision": 0.35454801464376234,
+        "macro_precision": 0.35468757089595926,
+        "ndcg": 0.5373826502849983,
+        "known_positive_recall": 0.36848413838619143,
+        "recall_queries": 925,
+        "useful": 16367,
+        "retained": 46163,
+        "excluded_positions": 87,
+        "mean_useful": 17.694054054054053,
+        "mean_retained": 49.905945945945945
+      }
+    },
+    "upstream_ettin": {
+      "1": {
+        "scored_queries": 919,
+        "excluded_queries": 6,
+        "hit": 0.5669205658324266,
+        "precision": 0.5669205658324266,
+        "macro_precision": 0.5669205658324266,
+        "ndcg": 0.4731851391263796,
+        "known_positive_recall": 0.01443636631579192,
+        "recall_queries": 919,
+        "useful": 521,
+        "retained": 919,
+        "excluded_positions": 0,
+        "mean_useful": 0.5669205658324266,
+        "mean_retained": 1.0
+      },
+      "3": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.7902702702702703,
+        "precision": 0.5603603603603604,
+        "macro_precision": 0.5603603603603603,
+        "ndcg": 0.4727968405371722,
+        "known_positive_recall": 0.04228504528013372,
+        "recall_queries": 925,
+        "useful": 1555,
+        "retained": 2775,
+        "excluded_positions": 0,
+        "mean_useful": 1.681081081081081,
+        "mean_retained": 3.0
+      },
+      "5": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.8562162162162162,
+        "precision": 0.5582954791261086,
+        "macro_precision": 0.5582162162162162,
+        "ndcg": 0.47778481822398317,
+        "known_positive_recall": 0.06742273722612116,
+        "recall_queries": 925,
+        "useful": 2581,
+        "retained": 4623,
+        "excluded_positions": 2,
+        "mean_useful": 2.79027027027027,
+        "mean_retained": 4.997837837837838
+      },
+      "10": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.9113513513513514,
+        "precision": 0.5447462395844606,
+        "macro_precision": 0.5447507507507507,
+        "ndcg": 0.49027329580696355,
+        "known_positive_recall": 0.12832495703473729,
+        "recall_queries": 925,
+        "useful": 5034,
+        "retained": 9241,
+        "excluded_positions": 9,
+        "mean_useful": 5.4421621621621625,
+        "mean_retained": 9.99027027027027
+      },
+      "20": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.9502702702702702,
+        "precision": 0.5220568335588633,
+        "macro_precision": 0.5220833774951422,
+        "ndcg": 0.5201039924104931,
+        "known_positive_recall": 0.23627621886671868,
+        "recall_queries": 925,
+        "useful": 9645,
+        "retained": 18475,
+        "excluded_positions": 25,
+        "mean_useful": 10.427027027027027,
+        "mean_retained": 19.972972972972972
+      },
+      "50": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.9762162162162162,
+        "precision": 0.424031024546656,
+        "macro_precision": 0.42417201849319736,
+        "ndcg": 0.5598208131173884,
+        "known_positive_recall": 0.4583210768438434,
+        "recall_queries": 925,
+        "useful": 19572,
+        "retained": 46157,
+        "excluded_positions": 93,
+        "mean_useful": 21.158918918918918,
+        "mean_retained": 49.89945945945946
+      }
+    },
+    "finetuned": {
+      "1": {
+        "scored_queries": 919,
+        "excluded_queries": 6,
+        "hit": 0.7595212187159956,
+        "precision": 0.7595212187159956,
+        "macro_precision": 0.7595212187159956,
+        "ndcg": 0.6526763044717343,
+        "known_positive_recall": 0.021181970033131128,
+        "recall_queries": 919,
+        "useful": 698,
+        "retained": 919,
+        "excluded_positions": 0,
+        "mean_useful": 0.7595212187159956,
+        "mean_retained": 1.0
+      },
+      "3": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.894054054054054,
+        "precision": 0.7510853835021708,
+        "macro_precision": 0.7518918918918919,
+        "ndcg": 0.6424089559634334,
+        "known_positive_recall": 0.058858902936843926,
+        "recall_queries": 925,
+        "useful": 2076,
+        "retained": 2764,
+        "excluded_positions": 11,
+        "mean_useful": 2.2443243243243245,
+        "mean_retained": 2.9881081081081082
+      },
+      "5": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.9275675675675675,
+        "precision": 0.7366710013003901,
+        "macro_precision": 0.736954954954955,
+        "ndcg": 0.6373095878906122,
+        "known_positive_recall": 0.0932034345948595,
+        "recall_queries": 925,
+        "useful": 3399,
+        "retained": 4614,
+        "excluded_positions": 11,
+        "mean_useful": 3.674594594594595,
+        "mean_retained": 4.988108108108108
+      },
+      "10": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.9427027027027027,
+        "precision": 0.7049002601908065,
+        "macro_precision": 0.7052921492921493,
+        "ndcg": 0.6377369246501603,
+        "known_positive_recall": 0.16987439814587393,
+        "recall_queries": 925,
+        "useful": 6502,
+        "retained": 9224,
+        "excluded_positions": 26,
+        "mean_useful": 7.029189189189189,
+        "mean_retained": 9.971891891891891
+      },
+      "20": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.9545945945945946,
+        "precision": 0.6479835212489159,
+        "macro_precision": 0.6484765048449259,
+        "ndcg": 0.6488792776497584,
+        "known_positive_recall": 0.2962743880804353,
+        "recall_queries": 925,
+        "useful": 11954,
+        "retained": 18448,
+        "excluded_positions": 52,
+        "mean_useful": 12.923243243243244,
+        "mean_retained": 19.943783783783783
+      },
+      "50": {
+        "scored_queries": 925,
+        "excluded_queries": 0,
+        "hit": 0.9762162162162162,
+        "precision": 0.424031024546656,
+        "macro_precision": 0.42417201849319736,
+        "ndcg": 0.612510329000538,
+        "known_positive_recall": 0.4583210768438434,
+        "recall_queries": 925,
+        "useful": 19572,
+        "retained": 46157,
+        "excluded_positions": 93,
+        "mean_useful": 21.158918918918918,
+        "mean_retained": 49.89945945945946
+      }
+    }
+  },
+  "corpus_passages": 82719,
+  "reference_pairs": 122783
+}

publication-manifest.json

@@ -1,6 +1,6 @@
 {
   "schema": "daecore.retrieval-publication-manifest",
-  "tool_sha256": "46dcb4b17750d5a4ba3d44b1236f33e9e5657673b648ae6cad48fe4ac4f147f8",
+  "tool_sha256": "60165b669f2cce17d8830fe9b712fb60a8e82280c54b224d71a1ce8fd82142ac",
   "model_id": "embeddinggemma-300m-memory-ft-v2",
   "public_repo": "Daecore/embeddinggemma-300m-memory-ft-v2",
   "role": "dense-retriever",
@@ -12,7 +12,7 @@
   "export_receipt_sha256": "e24e1692f7e03c9af8040bb3f6e310a061d305f485861007f1f37d36808b20d8",
   "serving_sha256": "8c088b1635875d5406e148782b303cfc5a961138600783fb4842b413b98a98c3",
   "source_model_tree_sha256": "30a7b9f1c42b1fcee29a6729742685d5b744e8fa87d13cefab2fd2d49ea9150e",
-  "staged_at": "2026-09-30T02:12:24+00:00",
+  "staged_at": "2026-10-01T16:22:42+00:00",
   "files": {
     "added_tokens.json": {
       "sha256": "50b2f405ba56a26d4913fd772089992252d7f942123cc0a034d96424221ba946",
@@ -80,8 +80,8 @@
       "binding": "packaging record"
     },
     "evaluation/README.md": {
-      "sha256": "bf96501863235b426b679d84e0bba96bc5b5d7fdba76571966af92993ce4150e",
-      "size": 11100,
+      "sha256": "eeda3234f34b38ccd765f67f0f11e71e125e09a383173cafcee49018884bf4a2",
+      "size": 12608,
       "binding": "packaging record"
     },
     "evaluation/serving-qualification.json": {
@@ -90,8 +90,8 @@
       "binding": "packaging record"
     },
     "evaluation/metrics.py": {
-      "sha256": "22536c4437be0b8c678df0dee4c45e31c5413e91f1fecf104750486f3e372d46",
-      "size": 19120,
+      "sha256": "8123f75da6edb158db74c62c1c65b6eefbc2713572409ac8a9bfb5424497aaf1",
+      "size": 19927,
       "binding": "packaging record"
     },
     "evaluation/chunking.md": {
@@ -119,6 +119,11 @@
       "size": 14521,
       "binding": "packaging record"
     },
+    "evaluation/pipeline-summary.json": {
+      "sha256": "5f28ae25b5db2f4c0a8d11891df7117f95f644063092d58ba45e27a59747e3a2",
+      "size": 9373,
+      "binding": "packaging record"
+    },
     "evaluation/retriever.json": {
       "sha256": "c209c96fd17144a6f7d2c7174bc27db4b346288910d085e2b614286b80ae4559",
       "size": 1156315,
@@ -139,6 +144,11 @@
       "size": 1184937,
       "binding": "packaging record"
     },
+    "evaluation/pipeline.json": {
+      "sha256": "8f140901a6b151d9db97fd109a3774bf1c046b849d9a01c874cb0efdea297300",
+      "size": 1614738,
+      "binding": "packaging record"
+    },
     "evaluation/figures.py": {
       "sha256": "f13945b350e007deb18a2eafbe1c5708a71b9df62ee12b6833d07f3e2a06094e",
       "size": 5490,
@@ -155,10 +165,10 @@
       "binding": "packaging record"
     },
     "README.md": {
-      "sha256": "7c9add0392d4d28e8e048a2308ac4cf88b5ac0d9ee2e5e49160625d3731975cb",
-      "size": 16352,
+      "sha256": "46843f44d3ca441406180c2f410a106601449ad092b99269f8c10ea751c1cfa3",
+      "size": 16894,
       "binding": "model card with upload-relative links",
-      "source_sha256": "38f074c37bde704f74971f782fa291850fdd325a4030671a344ef0ba4152a93d"
+      "source_sha256": "9e4e473e9a7479006bc9d944d70f81a27152819a708cb66d78c2c5d4915e9a5e"
     }
   }
 }