Chulinz commited on
Commit
e4eafc9
·
verified ·
1 Parent(s): 8712839

Reframe as sub-10B benchmark; drop anomaly and period change

Browse files
README.md CHANGED
@@ -6,88 +6,65 @@ tags:
6
  - text-to-sql
7
  - benchmark
8
  - postgresql
9
- - nemotron
10
- pretty_name: Nemotron 3.5 Bolt E4B app text-to-SQL baseline
11
  viewer: false
12
  ---
13
 
14
- # Nemotron 3.5 Bolt E4B: app text-to-SQL baseline
15
 
16
- Zero-shot PostgreSQL text-to-SQL results for [NVIDIA Nemotron 3.5 Bolt E4B](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16), recorded on 2026-10-27 as the reference point for a planned LoRA fine-tune. The same 2,800 questions were scored with the same execution scorer for Qwen3.5-9B as a similar-size general model (added 2026-10-29), for SLM-SQL 0.5B/1.5B and for Qwen3.5-0.8B.
17
 
18
- This repository holds results only. The questions, gold SQL and databases are kept private so the set stays usable as a held-out test, which also means the aggregate numbers cannot be reproduced from this repository alone.
19
 
20
  **Key results**
21
 
22
- - On ordinary lookups (`select` and `detail`) Qwen3.5-9B with thinking off answers 57.9% correctly against 51.4% for Nemotron. Qwen is ahead on six of the seven databases (HR is the exception) and the paired difference is 119 questions only Qwen gets right against 46 only Nemotron gets right (exact McNemar p < 1e-7). Both are ahead of SLM-SQL-1.5B at 41.9%.
23
- - On unanswerable questions Nemotron returns no runnable SQL 86.8% of the time against 72.7% for Qwen3.5-9B (97 against 18 questions handled by only one of the two, p < 1e-13). Nemotron emits the explicit `-- UNSUPPORTED` marker on 387 of 560 and Qwen on 287.
24
- - Every model scores zero on `anomaly` and `period_change`. The baseline prompt omits the reference date and the application's formulas. For Nemotron, stating the date and conventions lifts period change to 11.8%; adding SQL templates lifts anomaly to 18.6%. Even with templates, 431 of the 959 failed answers do not run; 110 of those call `strftime`, a SQLite function PostgreSQL lacks.
25
 
26
  ## Results
27
 
28
- Execution accuracy on 2,800 prompts over seven PostgreSQL databases. Each database contributes 400 prompts: 80 each of `select`, `detail`, `anomaly`, `period_change` and unanswerable questions.
29
 
30
- ![Execution accuracy by question category](charts/categories.png)
31
 
32
- | Model | Select + detail | Select | Detail | Anomaly | Period change | Unanswerable handled |
33
- |---|---:|---:|---:|---:|---:|---:|
34
- | Nemotron 3.5 Bolt E4B, zero-shot | 51.4 | 60.5 | 42.3 | 0.0 | 0.0 | **86.8** |
35
- | Qwen3.5-9B, thinking off, zero-shot | **57.9** | **69.1** | **46.8** | 0.0 | 0.0 | 72.7 |
36
- | SLM-SQL-1.5B | 41.9 | 50.4 | 33.4 | 0.0 | 0.0 | 57.7 |
37
- | SLM-SQL-0.5B | 30.4 | 37.5 | 23.4 | 0.0 | 0.0 | 59.5 |
38
-
39
- Percentages. Select + detail covers 1,120 questions, each other answerable column 560 and the unanswerable column 560. Qwen3.5-0.8B is reported per database below; its hosted endpoint failed on AdventureWorks and Synthea, so it has no seven-database total.
40
 
41
- `anomaly` and `period_change` questions follow conventions the prompt does not state: relative periods such as "this month" refer to the last complete month of data. Anomalies are scored with a modified z-score. All four models score zero on both. Nemotron answered `-- UNSUPPORTED` on 305 of the 560 `period_change` questions, which accounts for most of its 402 explicit refusals on answerable prompts (`select` 20, `detail` 21, `anomaly` 56). Qwen3.5-9B shows the same pattern with 211 of its 237 refusals on answerable prompts falling on `period_change`. The ablation below supplies those conventions in the prompt.
42
 
43
  ### Select + detail by database
44
 
45
  ![Select + detail accuracy by database](charts/databases.png)
46
 
47
- | Database | Nemotron 3.5 Bolt E4B | Qwen3.5-9B | SLM-SQL-1.5B | SLM-SQL-0.5B | Qwen3.5-0.8B |
48
  |---|---:|---:|---:|---:|---:|
49
- | Chinook | 63.1 | 70.6 | 61.2 | 46.2 | 21.2 |
50
- | HR | 55.6 | 50.0 | 39.4 | 39.4 | 19.4 |
51
- | AdventureWorks | 45.0 | 54.4 | 38.1 | 14.4 | n/a |
52
- | Northwind | 60.6 | 65.6 | 50.6 | 36.2 | 20.6 |
53
- | NYC Flights | 48.1 | 53.8 | 42.5 | 32.5 | 25.0 |
54
- | Pagila | 48.1 | 56.9 | 38.1 | 30.0 | 13.8 |
55
- | Synthea | 39.4 | 54.4 | 23.1 | 14.4 | n/a |
56
 
57
  Each cell covers 160 questions. Training data for these models is not known to us. Chinook, Northwind and Pagila are widely published sample databases, so prior exposure to their schemas cannot be ruled out.
58
 
59
- ## Ablation: stating the conventions in the prompt
60
-
61
- Nemotron was rerun on the 1,120 `anomaly` and `period_change` prompts with extra context inserted before the question, in two variants. The first adds the database's complete-data window, the last complete month and the output rules for both question types. The second also adds a SQL template for each type. Both blocks are reproduced in [prompt_context.md](prompt_context.md).
62
-
63
- ![Nemotron with app conventions in the prompt](charts/ablation.png)
64
-
65
- | Prompt | Anomaly | Period change | `-- UNSUPPORTED` on these prompts | Output did not run |
66
- |---|---:|---:|---:|---:|
67
- | Baseline | 0.0 | 0.0 | 361 | 553 |
68
- | + date and conventions | 0.4 | 11.8 | 41 | 689 |
69
- | + date, conventions and SQL templates | **18.6** | 10.2 | 30 | 431 |
70
-
71
- Percentages of 560 questions per column; the last two columns count across both types. Stating the date and conventions removes most refusals. The templates lift anomaly accuracy. With templates, 431 of the 959 failed answers do not run at all; among them 110 call `strftime`, which PostgreSQL does not provide, 32 break GROUP BY rules and 49 reference columns that do not exist. The other 528 run and return a different result from the gold query. The baseline row is the same run as the main table.
72
-
73
  ## Method
74
 
75
- **Prompt.** One user message containing the engine name, the full `CREATE TABLE` schema with column comments and up to three example values per column, the question and the instruction `Write one PostgreSQL query that answers the question. Reply with the SQL only. If the schema cannot answer the question, reply with -- UNSUPPORTED.` Nemotron, Qwen3.5-9B and Qwen3.5-0.8B received this exact text. SLM-SQL received the same schema and question inside its own published prompt template, which asks for step-by-step reasoning and a fenced SQL block and has no instruction for unanswerable questions; its unanswerable column therefore measures only how often its output failed to run.
76
-
77
- **Nemotron inference.** NVIDIA API catalog, model `nvidia/nemotron-3.5-bolt-e4b`, temperature 0, `max_tokens` 1024, one sample per prompt. All 2,800 baseline requests and all 2,240 ablation requests succeeded with no reasoning tokens returned. The hosted model's precision and serving stack are NVIDIA's and were not controlled by us.
78
-
79
- **Qwen3.5-9B inference.** `Qwen/Qwen3.5-9B` on a self-hosted OpenAI-compatible endpoint with `chat_template_kwargs.enable_thinking` set to false, temperature 0, `max_tokens` 1024 and one sample per prompt. With thinking left on the model spent the whole token budget reasoning and returned no SQL, so the reported row is thinking off. All 2,800 requests succeeded after retrying dropped connections. Serving precision was not recorded.
80
-
81
- **Scoring.** Each output is executed on PostgreSQL and its result table is compared with the gold query's result. Columns are matched by their values, so column names and order do not matter, but the column count, row count and every value must agree; row order counts only for gold queries marked as ordered. An unanswerable prompt counts as handled when the output does not run as SQL, so a broken query on an unanswerable question also counts. Nemotron emitted the exact `-- UNSUPPORTED` marker on 387 of the 560 unanswerable prompts.
82
 
83
- **Sample.** Each database has a 2,000-question held-out set. The 400 prompts per database are the first 80 per category after ordering question indices by SHA-256 of `database|index`, a rule that does not look at any model output. SLM-SQL was run on the full sets earlier and is scored here on the same 2,800 indices. SHA-256 of the seven sample files concatenated in the order Chinook, HR, AdventureWorks, Northwind, NYC Flights, Pagila, Synthea: `b87f4fbcc2c63425c75030d50e8b92d7cfdc9a6516bc50afc6ce0c3c18cb6354`.
84
 
85
- Full counts per database and category, including the ablation, are in [results.json](results.json).
86
 
87
- ## Deployment note
88
 
89
- Nemotron 3.5 Bolt E4B has 8B total and 4B effective parameters. Its Q4_K_M GGUF is 6.18 GB, which fits workstation and server use but not a typical phone.
90
 
91
  ## License
92
 
93
- These results are released under CC BY 4.0. Nemotron is distributed by NVIDIA under the OpenMDW License Agreement 1.1. Qwen3.5-9B is distributed by the Qwen team under its own license.
 
6
  - text-to-sql
7
  - benchmark
8
  - postgresql
9
+ pretty_name: Sub-10B app text-to-SQL benchmark
 
10
  viewer: false
11
  ---
12
 
13
+ # Sub-10B app text-to-SQL benchmark
14
 
15
+ Zero-shot PostgreSQL text-to-SQL results for open models under 10B parameters on the kind of questions a user asks an analytics app: lookups that return an aggregate or a list of records and questions the database cannot answer, where the right reply is a refusal. All models were scored with one execution scorer on the same 1,680 questions over seven databases.
16
 
17
+ This repository holds results only. The questions, gold SQL and databases are kept private so the set stays usable as a held-out test, which also means the numbers cannot be reproduced from this repository alone.
18
 
19
  **Key results**
20
 
21
+ - [Qwen3.5-9B](https://huggingface.co/Qwen/Qwen3.5-9B) with thinking off answers 57.9% of lookups correctly. It is ahead of [Nemotron 3.5 Bolt E4B](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16) at 51.4% on six of the seven databases, HR being the exception. On the 1,120 paired questions 119 are right only for Qwen and 46 only for Nemotron (exact McNemar p < 1e-7).
22
+ - Nemotron handles unanswerable questions best, returning no runnable SQL on 86.8% of them against 72.7% for Qwen3.5-9B (97 against 18 questions handled by only one of the two, p < 1e-13). It emits the explicit `-- UNSUPPORTED` marker on 387 of 560 and Qwen on 287.
23
+ - The SQL-specialised SLM-SQL models trail both general models on lookups, at 41.9% for 1.5B and 30.4% for 0.5B.
24
 
25
  ## Results
26
 
27
+ Each database contributes 240 prompts in three categories of 80. A `select` question asks for one aggregate measure, either as an overall number (with or without filters) or as a breakdown or top-k ranking by up to two columns. A `detail` question asks for a list of individual records with named columns and a filter or ordering that keeps it to at most 100 rows. An unanswerable question asks for something the database cannot provide: a data change, a cause, a prediction or data it does not hold.
28
 
29
+ ![Execution accuracy by question type](charts/categories.png)
30
 
31
+ | Model | Parameters | Select + detail | Select | Detail | Unanswerable handled |
32
+ |---|---:|---:|---:|---:|---:|
33
+ | Qwen3.5-9B, thinking off | 9B | **57.9** | **69.1** | **46.8** | 72.7 |
34
+ | Nemotron 3.5 Bolt E4B | 8B (4B effective) | 51.4 | 60.5 | 42.3 | **86.8** |
35
+ | SLM-SQL-1.5B | 1.5B | 41.9 | 50.4 | 33.4 | 57.7 |
36
+ | SLM-SQL-0.5B | 0.5B | 30.4 | 37.5 | 23.4 | 59.5 |
 
 
37
 
38
+ Percentages. Select + detail covers 1,120 questions and each other column 560. Qwen3.5-0.8B is reported per database below; its hosted endpoint failed on AdventureWorks and Synthea, so it has no seven-database total.
39
 
40
  ### Select + detail by database
41
 
42
  ![Select + detail accuracy by database](charts/databases.png)
43
 
44
+ | Database | Qwen3.5-9B | Nemotron 3.5 Bolt E4B | SLM-SQL-1.5B | SLM-SQL-0.5B | Qwen3.5-0.8B |
45
  |---|---:|---:|---:|---:|---:|
46
+ | Chinook | 70.6 | 63.1 | 61.2 | 46.2 | 21.2 |
47
+ | HR | 50.0 | 55.6 | 39.4 | 39.4 | 19.4 |
48
+ | AdventureWorks | 54.4 | 45.0 | 38.1 | 14.4 | n/a |
49
+ | Northwind | 65.6 | 60.6 | 50.6 | 36.2 | 20.6 |
50
+ | NYC Flights | 53.8 | 48.1 | 42.5 | 32.5 | 25.0 |
51
+ | Pagila | 56.9 | 48.1 | 38.1 | 30.0 | 13.8 |
52
+ | Synthea | 54.4 | 39.4 | 23.1 | 14.4 | n/a |
53
 
54
  Each cell covers 160 questions. Training data for these models is not known to us. Chinook, Northwind and Pagila are widely published sample databases, so prior exposure to their schemas cannot be ruled out.
55
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
56
  ## Method
57
 
58
+ **Prompt.** One user message containing the engine name, the full `CREATE TABLE` schema with column comments and up to three example values per column, the question and the instruction `Write one PostgreSQL query that answers the question. Reply with the SQL only. If the schema cannot answer the question, reply with -- UNSUPPORTED.` Qwen3.5-9B, Nemotron and Qwen3.5-0.8B received this exact text. SLM-SQL received the same schema and question inside its own published prompt template, which asks for step-by-step reasoning and a fenced SQL block and has no instruction for unanswerable questions; its unanswerable column therefore measures only how often its output failed to run.
 
 
 
 
 
 
59
 
60
+ **Inference.** Temperature 0, `max_tokens` 1024 and one sample per prompt for every model. Qwen3.5-9B ran on a self-hosted OpenAI-compatible endpoint on 2026-10-29 with `chat_template_kwargs.enable_thinking` set to false; with thinking on it spent the whole token budget reasoning and returned no SQL. Its serving precision was not recorded. Nemotron ran on the NVIDIA API catalog as `nvidia/nemotron-3.5-bolt-e4b` on 2026-10-27 with no reasoning tokens returned; its precision and serving stack are NVIDIA's. Every request for both models succeeded, some after retrying a dropped connection.
61
 
62
+ **Scoring.** Each output is executed on PostgreSQL and its result table is compared with the gold query's result. Columns are matched by their values, so column names and order do not matter, but the column count, row count and every value must agree; row order counts only for gold queries marked as ordered. An unanswerable prompt counts as handled when the output does not run as SQL, so a broken query on an unanswerable question also counts.
63
 
64
+ **Sample.** Each database has a 2,000-question held-out set. The prompts per database are the first 80 per category after ordering question indices by SHA-256 of `database|index`, a rule that does not look at any model output. SLM-SQL was run on the full sets earlier and is scored here on the same indices. SHA-256 of the seven sample files concatenated in the order Chinook, HR, AdventureWorks, Northwind, NYC Flights, Pagila, Synthea: `b87f4fbcc2c63425c75030d50e8b92d7cfdc9a6516bc50afc6ce0c3c18cb6354`. Those files also hold prompts of two further categories that this release does not score.
65
 
66
+ Full counts per database and category are in [results.json](results.json).
67
 
68
  ## License
69
 
70
+ These results are released under CC BY 4.0. Each model is distributed under its own license by its authors.
charts/ablation.png DELETED

Git LFS Details

  • SHA256: 881a7580890ba098788d6277b48cc0cbd9fbf2e0c8d28fb78781f18b33523767
  • Pointer size: 130 Bytes
  • Size of remote file: 47.5 kB
charts/categories.png CHANGED

Git LFS Details

  • SHA256: dbda35d2bf31b5dfb907965dd9d5415020499f573aeb2d3909e6e4568967e31b
  • Pointer size: 130 Bytes
  • Size of remote file: 80.5 kB

Git LFS Details

  • SHA256: 5bfd37ac30becf096aa5d3fd708a7c90e4aa5c01390a4d8350a60751e4dd1f58
  • Pointer size: 130 Bytes
  • Size of remote file: 65.7 kB
charts/databases.png CHANGED

Git LFS Details

  • SHA256: b81b09d5b3677dce301510ea4385b7b17fa1f4dd4cfaab792ac46abb52b0fbbe
  • Pointer size: 130 Bytes
  • Size of remote file: 86.2 kB

Git LFS Details

  • SHA256: f589ee01ca5077073b3efd52fbaea1b7654113d02a50a55fdd6d25809588a282
  • Pointer size: 130 Bytes
  • Size of remote file: 98.9 kB
prompt_context.md DELETED
@@ -1,42 +0,0 @@
1
- # Prompt context used in the ablation
2
-
3
- Inserted before `Question:` in each anomaly and period_change prompt. Placeholders in braces are filled per database from the complete-data window below; the application computes the same window at runtime.
4
-
5
- | Database | Start | End (exclusive) |
6
- |---|---|---|
7
- | chinook | 2021-01-01 | 2026-01-01 |
8
- | hr | 2011-01-01 | 2018-05-01 |
9
- | adventureworks | 2022-05-01 | 2025-07-01 |
10
- | northwind | 1996-07-01 | 1998-06-01 |
11
- | nycflights | 2013-01-01 | 2014-01-01 |
12
- | pagila | 2022-02-01 | 2026-08-01 |
13
- | synthea | 1914-07-01 | 2021-12-01 |
14
-
15
- ## 1. Date and conventions
16
-
17
- ```
18
- Application conventions:
19
- - Complete data covers [{start}, {end}). The latest complete month is {latest_month}; treat it as "this month". "This year" is [{year_start}, {end}); "last year" and "the same stretch last year" are the same span moved back 12 months. "Last month" is the month before {latest_month}. "Whole coverage" or no period means [{start}, {end}).
20
- - Filter dates as column >= 'start'::timestamp AND column < 'end'::timestamp.
21
- - Period change: compute the measure in the current and previous windows (per group when grouped, keeping groups present in either window) and return [group,] current_value, previous_value, change = current_value - previous_value, pct_change = change / previous_value * 100 (NULL when previous_value is NULL or 0). Return at most 100 rows unless the question asks for a number; one row when ungrouped. When ranking by percentage, drop groups whose previous_value is below 5% of the largest previous_value.
22
- - Anomaly: per unit, value = the measure. Over non-null values: median, MAD = median of |value - median|, mean_ad = mean of |value - median|. score = 0.6745 * (value - median) / MAD when MAD > 0, else (value - median) / (1.253314 * mean_ad) when mean_ad > 0, else 0. Return no rows when fewer than 8 units. Output unit, value, baseline (the median), score, is_outlier (abs(score) >= 3.5; score >= 3.5 for unusually high only; score <= -3.5 for unusually low only), ordered by abs(score) DESC then unit, at most 20 rows unless the question asks for a number.
23
- ```
24
-
25
- ## 2. SQL templates (second variant only)
26
-
27
- ```
28
- SQL templates (replace <...>; drop the grp parts when ungrouped):
29
- Period change:
30
- WITH current_period AS (SELECT <group> AS grp, (<measure>)::numeric AS value FROM <tables> WHERE <date> >= '<current_start>'::timestamp AND <date> < '<current_end>'::timestamp GROUP BY <group>),
31
- previous_period AS (SELECT <group> AS grp, (<measure>)::numeric AS value FROM <tables> WHERE <date> >= '<previous_start>'::timestamp AND <date> < '<previous_end>'::timestamp GROUP BY <group>),
32
- keys AS (SELECT grp FROM current_period UNION SELECT grp FROM previous_period),
33
- combined AS (SELECT k.grp, c.value AS current_value, p.value AS previous_value FROM keys k LEFT JOIN current_period c ON c.grp IS NOT DISTINCT FROM k.grp LEFT JOIN previous_period p ON p.grp IS NOT DISTINCT FROM k.grp)
34
- SELECT grp, current_value, previous_value, current_value - previous_value AS change, CASE WHEN previous_value IS NULL OR previous_value = 0 THEN NULL ELSE (current_value - previous_value) / previous_value * 100 END AS pct_change FROM combined ORDER BY <order> NULLS LAST, grp ASC NULLS LAST LIMIT <rows>
35
- Anomaly:
36
- WITH units AS (SELECT <unit> AS unit, (<measure>)::numeric AS value FROM <tables> WHERE <date> >= '<start>'::timestamp AND <date> < '<end>'::timestamp GROUP BY <unit>),
37
- population AS (SELECT unit, value FROM units WHERE value IS NOT NULL),
38
- center AS (SELECT percentile_cont(0.5) WITHIN GROUP (ORDER BY value) AS median, count(*) AS units FROM population),
39
- spread AS (SELECT percentile_cont(0.5) WITHIN GROUP (ORDER BY abs(p.value - c.median)) AS mad, avg(abs(p.value - c.median)) AS mean_ad FROM population p CROSS JOIN center c),
40
- scored AS (SELECT p.unit, p.value, c.median AS baseline, CASE WHEN s.mad > 0 THEN 0.6745 * (p.value - c.median) / s.mad WHEN s.mean_ad > 0 THEN (p.value - c.median) / (1.253314 * s.mean_ad) ELSE 0 END AS score FROM population p CROSS JOIN center c CROSS JOIN spread s WHERE c.units >= 8)
41
- SELECT unit, value, baseline, score, abs(score) >= 3.5 AS is_outlier FROM scored ORDER BY abs(score) DESC, unit ASC NULLS LAST LIMIT <rows>
42
- ```
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
results.json CHANGED
@@ -1,7 +1,6 @@
1
  {
2
  "meta": {
3
- "benchmark": "App text-to-SQL held-out sample, PostgreSQL execution accuracy",
4
- "date": "2026-10-27",
5
  "sets": [
6
  "chinook",
7
  "hr",
@@ -11,23 +10,38 @@
11
  "pagila",
12
  "synthea"
13
  ],
14
- "per_set": "400 prompts: 80 each of select, detail, anomaly, period_change and unanswerable",
15
- "sample_prompts_sha256": "b87f4fbcc2c63425c75030d50e8b92d7cfdc9a6516bc50afc6ce0c3c18cb6354",
16
- "nemotron_inference": {
17
- "endpoint": "NVIDIA API catalog integrate.api.nvidia.com/v1/chat/completions",
18
- "model": "nvidia/nemotron-3.5-bolt-e4b",
19
- "temperature": 0,
20
- "max_tokens": 1024,
21
- "samples": 1,
22
- "reasoning_tokens_returned": 0,
23
- "failed_requests": 0
24
- },
25
  "scoring": "Generated SQL is executed on PostgreSQL and its result table is compared with the gold query result; row order counts only for gold queries marked as ordered. An unanswerable prompt counts as handled when the output does not run as SQL.",
26
  "prompts": "Nemotron, Qwen3.5-9B and Qwen3.5-0.8B share one prompt text. SLM-SQL used its own published template with the same schema and question and no unanswerable instruction.",
27
- "sample_rule": "Per database, first 80 indices per category ordered by sha256(\"database|index\"); independent of model outputs.",
28
- "prompt_context_ablation": "Nemotron only, the 1,120 anomaly and period_change prompts. date_and_spec inserts prompt_context.md section 1 before the question; date_spec_templates adds section 2.",
29
- "qwen35_9b_inference": "Qwen/Qwen3.5-9B on a self-hosted OpenAI-compatible endpoint, chat_template_kwargs.enable_thinking=false, temperature 0, max_tokens 1024, one sample per prompt, same prompt text as Nemotron. All 2,800 requests succeeded after retries. Serving precision was not recorded.",
30
- "paired_tests": {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
31
  "select_plus_detail": {
32
  "qwen35_9b_only_correct": 119,
33
  "nemotron_only_correct": 46,
@@ -38,11 +52,10 @@
38
  "nemotron_only": 97,
39
  "exact_mcnemar_p": 2.8e-14
40
  }
41
- },
42
- "qwen35_9b_date": "2026-10-29"
43
  },
44
  "results": {
45
- "nvidia/nemotron-3.5-bolt-e4b (zero-shot)": {
46
  "sets_scored": [
47
  "chinook",
48
  "hr",
@@ -54,77 +67,51 @@
54
  ],
55
  "overall": {
56
  "select": {
57
- "correct": 339,
58
  "n": 560,
59
- "pct": 60.5
60
  },
61
  "detail": {
62
- "correct": 237,
63
- "n": 560,
64
- "pct": 42.3
65
- },
66
- "anomaly": {
67
- "correct": 0,
68
- "n": 560,
69
- "pct": 0.0
70
- },
71
- "period_change": {
72
- "correct": 0,
73
  "n": 560,
74
- "pct": 0.0
75
  },
76
  "select_plus_detail": {
77
- "correct": 576,
78
  "n": 1120,
79
- "pct": 51.4
80
- },
81
- "all_answerable": {
82
- "correct": 576,
83
- "n": 2240,
84
- "pct": 25.7
85
  },
86
  "unanswerable_no_runnable_sql": {
87
- "count": 486,
88
  "n": 560,
89
- "pct": 86.8
90
  }
91
  },
 
 
 
 
92
  "per_set": {
93
  "chinook": {
94
  "select": {
95
- "correct": 55,
96
  "n": 80,
97
- "pct": 68.8
98
  },
99
  "detail": {
100
- "correct": 46,
101
- "n": 80,
102
- "pct": 57.5
103
- },
104
- "anomaly": {
105
- "correct": 0,
106
- "n": 80,
107
- "pct": 0.0
108
- },
109
- "period_change": {
110
- "correct": 0,
111
  "n": 80,
112
- "pct": 0.0
113
  },
114
  "select_plus_detail": {
115
- "correct": 101,
116
  "n": 160,
117
- "pct": 63.1
118
- },
119
- "all_answerable": {
120
- "correct": 101,
121
- "n": 320,
122
- "pct": 31.6
123
  },
124
  "unanswerable_no_runnable_sql": {
125
- "count": 66,
126
  "n": 80,
127
- "pct": 82.5
128
  }
129
  },
130
  "hr": {
@@ -134,228 +121,134 @@
134
  "pct": 56.2
135
  },
136
  "detail": {
137
- "correct": 44,
138
- "n": 80,
139
- "pct": 55.0
140
- },
141
- "anomaly": {
142
- "correct": 0,
143
- "n": 80,
144
- "pct": 0.0
145
- },
146
- "period_change": {
147
- "correct": 0,
148
  "n": 80,
149
- "pct": 0.0
150
  },
151
  "select_plus_detail": {
152
- "correct": 89,
153
  "n": 160,
154
- "pct": 55.6
155
- },
156
- "all_answerable": {
157
- "correct": 89,
158
- "n": 320,
159
- "pct": 27.8
160
  },
161
  "unanswerable_no_runnable_sql": {
162
- "count": 71,
163
  "n": 80,
164
- "pct": 88.8
165
  }
166
  },
167
  "adventureworks": {
168
  "select": {
169
- "correct": 44,
170
  "n": 80,
171
- "pct": 55.0
172
  },
173
  "detail": {
174
- "correct": 28,
175
- "n": 80,
176
- "pct": 35.0
177
- },
178
- "anomaly": {
179
- "correct": 0,
180
- "n": 80,
181
- "pct": 0.0
182
- },
183
- "period_change": {
184
- "correct": 0,
185
  "n": 80,
186
- "pct": 0.0
187
  },
188
  "select_plus_detail": {
189
- "correct": 72,
190
  "n": 160,
191
- "pct": 45.0
192
- },
193
- "all_answerable": {
194
- "correct": 72,
195
- "n": 320,
196
- "pct": 22.5
197
  },
198
  "unanswerable_no_runnable_sql": {
199
- "count": 66,
200
  "n": 80,
201
- "pct": 82.5
202
  }
203
  },
204
  "northwind": {
205
  "select": {
206
- "correct": 52,
207
  "n": 80,
208
- "pct": 65.0
209
  },
210
  "detail": {
211
- "correct": 45,
212
- "n": 80,
213
- "pct": 56.2
214
- },
215
- "anomaly": {
216
- "correct": 0,
217
- "n": 80,
218
- "pct": 0.0
219
- },
220
- "period_change": {
221
- "correct": 0,
222
  "n": 80,
223
- "pct": 0.0
224
  },
225
  "select_plus_detail": {
226
- "correct": 97,
227
  "n": 160,
228
- "pct": 60.6
229
- },
230
- "all_answerable": {
231
- "correct": 97,
232
- "n": 320,
233
- "pct": 30.3
234
  },
235
  "unanswerable_no_runnable_sql": {
236
- "count": 69,
237
  "n": 80,
238
- "pct": 86.2
239
  }
240
  },
241
  "nycflights": {
242
  "select": {
243
- "correct": 50,
244
  "n": 80,
245
- "pct": 62.5
246
  },
247
  "detail": {
248
- "correct": 27,
249
- "n": 80,
250
- "pct": 33.8
251
- },
252
- "anomaly": {
253
- "correct": 0,
254
- "n": 80,
255
- "pct": 0.0
256
- },
257
- "period_change": {
258
- "correct": 0,
259
  "n": 80,
260
- "pct": 0.0
261
  },
262
  "select_plus_detail": {
263
- "correct": 77,
264
  "n": 160,
265
- "pct": 48.1
266
- },
267
- "all_answerable": {
268
- "correct": 77,
269
- "n": 320,
270
- "pct": 24.1
271
  },
272
  "unanswerable_no_runnable_sql": {
273
- "count": 74,
274
  "n": 80,
275
- "pct": 92.5
276
  }
277
  },
278
  "pagila": {
279
  "select": {
280
- "correct": 50,
281
  "n": 80,
282
- "pct": 62.5
283
  },
284
  "detail": {
285
- "correct": 27,
286
- "n": 80,
287
- "pct": 33.8
288
- },
289
- "anomaly": {
290
- "correct": 0,
291
- "n": 80,
292
- "pct": 0.0
293
- },
294
- "period_change": {
295
- "correct": 0,
296
  "n": 80,
297
- "pct": 0.0
298
  },
299
  "select_plus_detail": {
300
- "correct": 77,
301
  "n": 160,
302
- "pct": 48.1
303
- },
304
- "all_answerable": {
305
- "correct": 77,
306
- "n": 320,
307
- "pct": 24.1
308
  },
309
  "unanswerable_no_runnable_sql": {
310
- "count": 67,
311
  "n": 80,
312
- "pct": 83.8
313
  }
314
  },
315
  "synthea": {
316
  "select": {
317
- "correct": 43,
318
  "n": 80,
319
- "pct": 53.8
320
  },
321
  "detail": {
322
- "correct": 20,
323
- "n": 80,
324
- "pct": 25.0
325
- },
326
- "anomaly": {
327
- "correct": 0,
328
- "n": 80,
329
- "pct": 0.0
330
- },
331
- "period_change": {
332
- "correct": 0,
333
  "n": 80,
334
- "pct": 0.0
335
  },
336
  "select_plus_detail": {
337
- "correct": 63,
338
  "n": 160,
339
- "pct": 39.4
340
- },
341
- "all_answerable": {
342
- "correct": 63,
343
- "n": 320,
344
- "pct": 19.7
345
  },
346
  "unanswerable_no_runnable_sql": {
347
- "count": 73,
348
  "n": 80,
349
- "pct": 91.2
350
  }
351
  }
352
- },
353
- "explicit_unsupported_marker": {
354
- "on_unanswerable": "387/560",
355
- "on_answerable": "402/2240"
356
  }
357
  },
358
- "Qwen3.5-9B (zero-shot)": {
359
  "sets_scored": [
360
  "chinook",
361
  "hr",
@@ -367,81 +260,51 @@
367
  ],
368
  "overall": {
369
  "select": {
370
- "correct": 387,
371
  "n": 560,
372
- "pct": 69.1
373
  },
374
  "detail": {
375
- "correct": 262,
376
- "n": 560,
377
- "pct": 46.8
378
- },
379
- "anomaly": {
380
- "correct": 0,
381
- "n": 560,
382
- "pct": 0.0
383
- },
384
- "period_change": {
385
- "correct": 0,
386
  "n": 560,
387
- "pct": 0.0
388
  },
389
  "select_plus_detail": {
390
- "correct": 649,
391
  "n": 1120,
392
- "pct": 57.9
393
- },
394
- "all_answerable": {
395
- "correct": 649,
396
- "n": 2240,
397
- "pct": 29.0
398
  },
399
  "unanswerable_no_runnable_sql": {
400
- "count": 407,
401
  "n": 560,
402
- "pct": 72.7
403
  }
404
  },
405
  "explicit_unsupported_marker": {
406
- "on_unanswerable": "287/560",
407
- "on_answerable": "237/2240"
408
  },
409
  "per_set": {
410
  "chinook": {
411
  "select": {
412
- "correct": 63,
413
  "n": 80,
414
- "pct": 78.8
415
  },
416
  "detail": {
417
- "correct": 50,
418
- "n": 80,
419
- "pct": 62.5
420
- },
421
- "anomaly": {
422
- "correct": 0,
423
- "n": 80,
424
- "pct": 0.0
425
- },
426
- "period_change": {
427
- "correct": 0,
428
  "n": 80,
429
- "pct": 0.0
430
  },
431
  "select_plus_detail": {
432
- "correct": 113,
433
  "n": 160,
434
- "pct": 70.6
435
- },
436
- "all_answerable": {
437
- "correct": 113,
438
- "n": 320,
439
- "pct": 35.3
440
  },
441
  "unanswerable_no_runnable_sql": {
442
- "count": 50,
443
  "n": 80,
444
- "pct": 62.5
445
  }
446
  },
447
  "hr": {
@@ -451,219 +314,129 @@
451
  "pct": 56.2
452
  },
453
  "detail": {
454
- "correct": 35,
455
- "n": 80,
456
- "pct": 43.8
457
- },
458
- "anomaly": {
459
- "correct": 0,
460
- "n": 80,
461
- "pct": 0.0
462
- },
463
- "period_change": {
464
- "correct": 0,
465
  "n": 80,
466
- "pct": 0.0
467
  },
468
  "select_plus_detail": {
469
- "correct": 80,
470
  "n": 160,
471
- "pct": 50.0
472
- },
473
- "all_answerable": {
474
- "correct": 80,
475
- "n": 320,
476
- "pct": 25.0
477
  },
478
  "unanswerable_no_runnable_sql": {
479
- "count": 52,
480
  "n": 80,
481
- "pct": 65.0
482
  }
483
  },
484
  "adventureworks": {
485
  "select": {
486
- "correct": 52,
487
  "n": 80,
488
- "pct": 65.0
489
  },
490
  "detail": {
491
- "correct": 35,
492
- "n": 80,
493
- "pct": 43.8
494
- },
495
- "anomaly": {
496
- "correct": 0,
497
- "n": 80,
498
- "pct": 0.0
499
- },
500
- "period_change": {
501
- "correct": 0,
502
  "n": 80,
503
- "pct": 0.0
504
  },
505
  "select_plus_detail": {
506
- "correct": 87,
507
  "n": 160,
508
- "pct": 54.4
509
- },
510
- "all_answerable": {
511
- "correct": 87,
512
- "n": 320,
513
- "pct": 27.2
514
  },
515
  "unanswerable_no_runnable_sql": {
516
- "count": 56,
517
  "n": 80,
518
- "pct": 70.0
519
  }
520
  },
521
  "northwind": {
522
  "select": {
523
- "correct": 59,
524
  "n": 80,
525
- "pct": 73.8
526
  },
527
  "detail": {
528
- "correct": 46,
529
- "n": 80,
530
- "pct": 57.5
531
- },
532
- "anomaly": {
533
- "correct": 0,
534
- "n": 80,
535
- "pct": 0.0
536
- },
537
- "period_change": {
538
- "correct": 0,
539
  "n": 80,
540
- "pct": 0.0
541
  },
542
  "select_plus_detail": {
543
- "correct": 105,
544
  "n": 160,
545
- "pct": 65.6
546
- },
547
- "all_answerable": {
548
- "correct": 105,
549
- "n": 320,
550
- "pct": 32.8
551
  },
552
  "unanswerable_no_runnable_sql": {
553
- "count": 61,
554
  "n": 80,
555
- "pct": 76.2
556
  }
557
  },
558
  "nycflights": {
559
  "select": {
560
- "correct": 53,
561
  "n": 80,
562
- "pct": 66.2
563
  },
564
  "detail": {
565
- "correct": 33,
566
- "n": 80,
567
- "pct": 41.2
568
- },
569
- "anomaly": {
570
- "correct": 0,
571
- "n": 80,
572
- "pct": 0.0
573
- },
574
- "period_change": {
575
- "correct": 0,
576
  "n": 80,
577
- "pct": 0.0
578
  },
579
  "select_plus_detail": {
580
- "correct": 86,
581
  "n": 160,
582
- "pct": 53.8
583
- },
584
- "all_answerable": {
585
- "correct": 86,
586
- "n": 320,
587
- "pct": 26.9
588
  },
589
  "unanswerable_no_runnable_sql": {
590
- "count": 65,
591
  "n": 80,
592
- "pct": 81.2
593
  }
594
  },
595
  "pagila": {
596
  "select": {
597
- "correct": 55,
598
  "n": 80,
599
- "pct": 68.8
600
  },
601
  "detail": {
602
- "correct": 36,
603
- "n": 80,
604
- "pct": 45.0
605
- },
606
- "anomaly": {
607
- "correct": 0,
608
- "n": 80,
609
- "pct": 0.0
610
- },
611
- "period_change": {
612
- "correct": 0,
613
  "n": 80,
614
- "pct": 0.0
615
  },
616
  "select_plus_detail": {
617
- "correct": 91,
618
  "n": 160,
619
- "pct": 56.9
620
- },
621
- "all_answerable": {
622
- "correct": 91,
623
- "n": 320,
624
- "pct": 28.4
625
  },
626
  "unanswerable_no_runnable_sql": {
627
- "count": 54,
628
  "n": 80,
629
- "pct": 67.5
630
  }
631
  },
632
  "synthea": {
633
  "select": {
634
- "correct": 60,
635
  "n": 80,
636
- "pct": 75.0
637
  },
638
  "detail": {
639
- "correct": 27,
640
- "n": 80,
641
- "pct": 33.8
642
- },
643
- "anomaly": {
644
- "correct": 0,
645
- "n": 80,
646
- "pct": 0.0
647
- },
648
- "period_change": {
649
- "correct": 0,
650
  "n": 80,
651
- "pct": 0.0
652
  },
653
  "select_plus_detail": {
654
- "correct": 87,
655
  "n": 160,
656
- "pct": 54.4
657
- },
658
- "all_answerable": {
659
- "correct": 87,
660
- "n": 320,
661
- "pct": 27.2
662
  },
663
  "unanswerable_no_runnable_sql": {
664
- "count": 69,
665
  "n": 80,
666
- "pct": 86.2
667
  }
668
  }
669
  }
@@ -689,26 +462,11 @@
689
  "n": 560,
690
  "pct": 33.4
691
  },
692
- "anomaly": {
693
- "correct": 0,
694
- "n": 560,
695
- "pct": 0.0
696
- },
697
- "period_change": {
698
- "correct": 0,
699
- "n": 560,
700
- "pct": 0.0
701
- },
702
  "select_plus_detail": {
703
  "correct": 469,
704
  "n": 1120,
705
  "pct": 41.9
706
  },
707
- "all_answerable": {
708
- "correct": 469,
709
- "n": 2240,
710
- "pct": 20.9
711
- },
712
  "unanswerable_no_runnable_sql": {
713
  "count": 323,
714
  "n": 560,
@@ -727,26 +485,11 @@
727
  "n": 80,
728
  "pct": 56.2
729
  },
730
- "anomaly": {
731
- "correct": 0,
732
- "n": 80,
733
- "pct": 0.0
734
- },
735
- "period_change": {
736
- "correct": 0,
737
- "n": 80,
738
- "pct": 0.0
739
- },
740
  "select_plus_detail": {
741
  "correct": 98,
742
  "n": 160,
743
  "pct": 61.2
744
  },
745
- "all_answerable": {
746
- "correct": 98,
747
- "n": 320,
748
- "pct": 30.6
749
- },
750
  "unanswerable_no_runnable_sql": {
751
  "count": 35,
752
  "n": 80,
@@ -764,26 +507,11 @@
764
  "n": 80,
765
  "pct": 37.5
766
  },
767
- "anomaly": {
768
- "correct": 0,
769
- "n": 80,
770
- "pct": 0.0
771
- },
772
- "period_change": {
773
- "correct": 0,
774
- "n": 80,
775
- "pct": 0.0
776
- },
777
  "select_plus_detail": {
778
  "correct": 63,
779
  "n": 160,
780
  "pct": 39.4
781
  },
782
- "all_answerable": {
783
- "correct": 63,
784
- "n": 320,
785
- "pct": 19.7
786
- },
787
  "unanswerable_no_runnable_sql": {
788
  "count": 42,
789
  "n": 80,
@@ -801,26 +529,11 @@
801
  "n": 80,
802
  "pct": 33.8
803
  },
804
- "anomaly": {
805
- "correct": 0,
806
- "n": 80,
807
- "pct": 0.0
808
- },
809
- "period_change": {
810
- "correct": 0,
811
- "n": 80,
812
- "pct": 0.0
813
- },
814
  "select_plus_detail": {
815
  "correct": 61,
816
  "n": 160,
817
  "pct": 38.1
818
  },
819
- "all_answerable": {
820
- "correct": 61,
821
- "n": 320,
822
- "pct": 19.1
823
- },
824
  "unanswerable_no_runnable_sql": {
825
  "count": 58,
826
  "n": 80,
@@ -838,26 +551,11 @@
838
  "n": 80,
839
  "pct": 41.2
840
  },
841
- "anomaly": {
842
- "correct": 0,
843
- "n": 80,
844
- "pct": 0.0
845
- },
846
- "period_change": {
847
- "correct": 0,
848
- "n": 80,
849
- "pct": 0.0
850
- },
851
  "select_plus_detail": {
852
  "correct": 81,
853
  "n": 160,
854
  "pct": 50.6
855
  },
856
- "all_answerable": {
857
- "correct": 81,
858
- "n": 320,
859
- "pct": 25.3
860
- },
861
  "unanswerable_no_runnable_sql": {
862
  "count": 49,
863
  "n": 80,
@@ -875,26 +573,11 @@
875
  "n": 80,
876
  "pct": 31.2
877
  },
878
- "anomaly": {
879
- "correct": 0,
880
- "n": 80,
881
- "pct": 0.0
882
- },
883
- "period_change": {
884
- "correct": 0,
885
- "n": 80,
886
- "pct": 0.0
887
- },
888
  "select_plus_detail": {
889
  "correct": 68,
890
  "n": 160,
891
  "pct": 42.5
892
  },
893
- "all_answerable": {
894
- "correct": 68,
895
- "n": 320,
896
- "pct": 21.2
897
- },
898
  "unanswerable_no_runnable_sql": {
899
  "count": 49,
900
  "n": 80,
@@ -912,26 +595,11 @@
912
  "n": 80,
913
  "pct": 26.2
914
  },
915
- "anomaly": {
916
- "correct": 0,
917
- "n": 80,
918
- "pct": 0.0
919
- },
920
- "period_change": {
921
- "correct": 0,
922
- "n": 80,
923
- "pct": 0.0
924
- },
925
  "select_plus_detail": {
926
  "correct": 61,
927
  "n": 160,
928
  "pct": 38.1
929
  },
930
- "all_answerable": {
931
- "correct": 61,
932
- "n": 320,
933
- "pct": 19.1
934
- },
935
  "unanswerable_no_runnable_sql": {
936
  "count": 43,
937
  "n": 80,
@@ -949,26 +617,11 @@
949
  "n": 80,
950
  "pct": 7.5
951
  },
952
- "anomaly": {
953
- "correct": 0,
954
- "n": 80,
955
- "pct": 0.0
956
- },
957
- "period_change": {
958
- "correct": 0,
959
- "n": 80,
960
- "pct": 0.0
961
- },
962
  "select_plus_detail": {
963
  "correct": 37,
964
  "n": 160,
965
  "pct": 23.1
966
  },
967
- "all_answerable": {
968
- "correct": 37,
969
- "n": 320,
970
- "pct": 11.6
971
- },
972
  "unanswerable_no_runnable_sql": {
973
  "count": 47,
974
  "n": 80,
@@ -998,26 +651,11 @@
998
  "n": 560,
999
  "pct": 23.4
1000
  },
1001
- "anomaly": {
1002
- "correct": 0,
1003
- "n": 560,
1004
- "pct": 0.0
1005
- },
1006
- "period_change": {
1007
- "correct": 0,
1008
- "n": 560,
1009
- "pct": 0.0
1010
- },
1011
  "select_plus_detail": {
1012
  "correct": 341,
1013
  "n": 1120,
1014
  "pct": 30.4
1015
  },
1016
- "all_answerable": {
1017
- "correct": 341,
1018
- "n": 2240,
1019
- "pct": 15.2
1020
- },
1021
  "unanswerable_no_runnable_sql": {
1022
  "count": 333,
1023
  "n": 560,
@@ -1036,26 +674,11 @@
1036
  "n": 80,
1037
  "pct": 42.5
1038
  },
1039
- "anomaly": {
1040
- "correct": 0,
1041
- "n": 80,
1042
- "pct": 0.0
1043
- },
1044
- "period_change": {
1045
- "correct": 0,
1046
- "n": 80,
1047
- "pct": 0.0
1048
- },
1049
  "select_plus_detail": {
1050
  "correct": 74,
1051
  "n": 160,
1052
  "pct": 46.2
1053
  },
1054
- "all_answerable": {
1055
- "correct": 74,
1056
- "n": 320,
1057
- "pct": 23.1
1058
- },
1059
  "unanswerable_no_runnable_sql": {
1060
  "count": 31,
1061
  "n": 80,
@@ -1073,26 +696,11 @@
1073
  "n": 80,
1074
  "pct": 33.8
1075
  },
1076
- "anomaly": {
1077
- "correct": 0,
1078
- "n": 80,
1079
- "pct": 0.0
1080
- },
1081
- "period_change": {
1082
- "correct": 0,
1083
- "n": 80,
1084
- "pct": 0.0
1085
- },
1086
  "select_plus_detail": {
1087
  "correct": 63,
1088
  "n": 160,
1089
  "pct": 39.4
1090
  },
1091
- "all_answerable": {
1092
- "correct": 63,
1093
- "n": 320,
1094
- "pct": 19.7
1095
- },
1096
  "unanswerable_no_runnable_sql": {
1097
  "count": 43,
1098
  "n": 80,
@@ -1110,26 +718,11 @@
1110
  "n": 80,
1111
  "pct": 12.5
1112
  },
1113
- "anomaly": {
1114
- "correct": 0,
1115
- "n": 80,
1116
- "pct": 0.0
1117
- },
1118
- "period_change": {
1119
- "correct": 0,
1120
- "n": 80,
1121
- "pct": 0.0
1122
- },
1123
  "select_plus_detail": {
1124
  "correct": 23,
1125
  "n": 160,
1126
  "pct": 14.4
1127
  },
1128
- "all_answerable": {
1129
- "correct": 23,
1130
- "n": 320,
1131
- "pct": 7.2
1132
- },
1133
  "unanswerable_no_runnable_sql": {
1134
  "count": 72,
1135
  "n": 80,
@@ -1147,26 +740,11 @@
1147
  "n": 80,
1148
  "pct": 30.0
1149
  },
1150
- "anomaly": {
1151
- "correct": 0,
1152
- "n": 80,
1153
- "pct": 0.0
1154
- },
1155
- "period_change": {
1156
- "correct": 0,
1157
- "n": 80,
1158
- "pct": 0.0
1159
- },
1160
  "select_plus_detail": {
1161
  "correct": 58,
1162
  "n": 160,
1163
  "pct": 36.2
1164
  },
1165
- "all_answerable": {
1166
- "correct": 58,
1167
- "n": 320,
1168
- "pct": 18.1
1169
- },
1170
  "unanswerable_no_runnable_sql": {
1171
  "count": 50,
1172
  "n": 80,
@@ -1184,26 +762,11 @@
1184
  "n": 80,
1185
  "pct": 20.0
1186
  },
1187
- "anomaly": {
1188
- "correct": 0,
1189
- "n": 80,
1190
- "pct": 0.0
1191
- },
1192
- "period_change": {
1193
- "correct": 0,
1194
- "n": 80,
1195
- "pct": 0.0
1196
- },
1197
  "select_plus_detail": {
1198
  "correct": 52,
1199
  "n": 160,
1200
  "pct": 32.5
1201
  },
1202
- "all_answerable": {
1203
- "correct": 52,
1204
- "n": 320,
1205
- "pct": 16.2
1206
- },
1207
  "unanswerable_no_runnable_sql": {
1208
  "count": 46,
1209
  "n": 80,
@@ -1221,26 +784,11 @@
1221
  "n": 80,
1222
  "pct": 21.2
1223
  },
1224
- "anomaly": {
1225
- "correct": 0,
1226
- "n": 80,
1227
- "pct": 0.0
1228
- },
1229
- "period_change": {
1230
- "correct": 0,
1231
- "n": 80,
1232
- "pct": 0.0
1233
- },
1234
  "select_plus_detail": {
1235
  "correct": 48,
1236
  "n": 160,
1237
  "pct": 30.0
1238
  },
1239
- "all_answerable": {
1240
- "correct": 48,
1241
- "n": 320,
1242
- "pct": 15.0
1243
- },
1244
  "unanswerable_no_runnable_sql": {
1245
  "count": 39,
1246
  "n": 80,
@@ -1258,26 +806,11 @@
1258
  "n": 80,
1259
  "pct": 3.8
1260
  },
1261
- "anomaly": {
1262
- "correct": 0,
1263
- "n": 80,
1264
- "pct": 0.0
1265
- },
1266
- "period_change": {
1267
- "correct": 0,
1268
- "n": 80,
1269
- "pct": 0.0
1270
- },
1271
  "select_plus_detail": {
1272
  "correct": 23,
1273
  "n": 160,
1274
  "pct": 14.4
1275
  },
1276
- "all_answerable": {
1277
- "correct": 23,
1278
- "n": 320,
1279
- "pct": 7.2
1280
- },
1281
  "unanswerable_no_runnable_sql": {
1282
  "count": 52,
1283
  "n": 80,
@@ -1305,26 +838,11 @@
1305
  "n": 400,
1306
  "pct": 13.0
1307
  },
1308
- "anomaly": {
1309
- "correct": 0,
1310
- "n": 400,
1311
- "pct": 0.0
1312
- },
1313
- "period_change": {
1314
- "correct": 0,
1315
- "n": 400,
1316
- "pct": 0.0
1317
- },
1318
  "select_plus_detail": {
1319
  "correct": 160,
1320
  "n": 800,
1321
  "pct": 20.0
1322
  },
1323
- "all_answerable": {
1324
- "correct": 160,
1325
- "n": 1600,
1326
- "pct": 10.0
1327
- },
1328
  "unanswerable_no_runnable_sql": {
1329
  "count": 279,
1330
  "n": 400,
@@ -1343,26 +861,11 @@
1343
  "n": 80,
1344
  "pct": 17.5
1345
  },
1346
- "anomaly": {
1347
- "correct": 0,
1348
- "n": 80,
1349
- "pct": 0.0
1350
- },
1351
- "period_change": {
1352
- "correct": 0,
1353
- "n": 80,
1354
- "pct": 0.0
1355
- },
1356
  "select_plus_detail": {
1357
  "correct": 34,
1358
  "n": 160,
1359
  "pct": 21.2
1360
  },
1361
- "all_answerable": {
1362
- "correct": 34,
1363
- "n": 320,
1364
- "pct": 10.6
1365
- },
1366
  "unanswerable_no_runnable_sql": {
1367
  "count": 55,
1368
  "n": 80,
@@ -1380,26 +883,11 @@
1380
  "n": 80,
1381
  "pct": 10.0
1382
  },
1383
- "anomaly": {
1384
- "correct": 0,
1385
- "n": 80,
1386
- "pct": 0.0
1387
- },
1388
- "period_change": {
1389
- "correct": 0,
1390
- "n": 80,
1391
- "pct": 0.0
1392
- },
1393
  "select_plus_detail": {
1394
  "correct": 31,
1395
  "n": 160,
1396
  "pct": 19.4
1397
  },
1398
- "all_answerable": {
1399
- "correct": 31,
1400
- "n": 320,
1401
- "pct": 9.7
1402
- },
1403
  "unanswerable_no_runnable_sql": {
1404
  "count": 42,
1405
  "n": 80,
@@ -1417,26 +905,11 @@
1417
  "n": 80,
1418
  "pct": 13.8
1419
  },
1420
- "anomaly": {
1421
- "correct": 0,
1422
- "n": 80,
1423
- "pct": 0.0
1424
- },
1425
- "period_change": {
1426
- "correct": 0,
1427
- "n": 80,
1428
- "pct": 0.0
1429
- },
1430
  "select_plus_detail": {
1431
  "correct": 33,
1432
  "n": 160,
1433
  "pct": 20.6
1434
  },
1435
- "all_answerable": {
1436
- "correct": 33,
1437
- "n": 320,
1438
- "pct": 10.3
1439
- },
1440
  "unanswerable_no_runnable_sql": {
1441
  "count": 61,
1442
  "n": 80,
@@ -1454,26 +927,11 @@
1454
  "n": 80,
1455
  "pct": 16.2
1456
  },
1457
- "anomaly": {
1458
- "correct": 0,
1459
- "n": 80,
1460
- "pct": 0.0
1461
- },
1462
- "period_change": {
1463
- "correct": 0,
1464
- "n": 80,
1465
- "pct": 0.0
1466
- },
1467
  "select_plus_detail": {
1468
  "correct": 40,
1469
  "n": 160,
1470
  "pct": 25.0
1471
  },
1472
- "all_answerable": {
1473
- "correct": 40,
1474
- "n": 320,
1475
- "pct": 12.5
1476
- },
1477
  "unanswerable_no_runnable_sql": {
1478
  "count": 64,
1479
  "n": 80,
@@ -1491,26 +949,11 @@
1491
  "n": 80,
1492
  "pct": 7.5
1493
  },
1494
- "anomaly": {
1495
- "correct": 0,
1496
- "n": 80,
1497
- "pct": 0.0
1498
- },
1499
- "period_change": {
1500
- "correct": 0,
1501
- "n": 80,
1502
- "pct": 0.0
1503
- },
1504
  "select_plus_detail": {
1505
  "correct": 22,
1506
  "n": 160,
1507
  "pct": 13.8
1508
  },
1509
- "all_answerable": {
1510
- "correct": 22,
1511
- "n": 320,
1512
- "pct": 6.9
1513
- },
1514
  "unanswerable_no_runnable_sql": {
1515
  "count": 57,
1516
  "n": 80,
@@ -1519,145 +962,5 @@
1519
  }
1520
  }
1521
  }
1522
- },
1523
- "prompt_context_ablation": {
1524
- "baseline": {
1525
- "anomaly": {
1526
- "correct": 0,
1527
- "n": 560,
1528
- "pct": 0.0,
1529
- "unsupported": 56,
1530
- "did_not_run": 185
1531
- },
1532
- "period_change": {
1533
- "correct": 0,
1534
- "n": 560,
1535
- "pct": 0.0,
1536
- "unsupported": 305,
1537
- "did_not_run": 368
1538
- },
1539
- "per_database_correct_of_80": {
1540
- "chinook": {
1541
- "anomaly": 0,
1542
- "period_change": 0
1543
- },
1544
- "hr": {
1545
- "anomaly": 0,
1546
- "period_change": 0
1547
- },
1548
- "adventureworks": {
1549
- "anomaly": 0,
1550
- "period_change": 0
1551
- },
1552
- "northwind": {
1553
- "anomaly": 0,
1554
- "period_change": 0
1555
- },
1556
- "nycflights": {
1557
- "anomaly": 0,
1558
- "period_change": 0
1559
- },
1560
- "pagila": {
1561
- "anomaly": 0,
1562
- "period_change": 0
1563
- },
1564
- "synthea": {
1565
- "anomaly": 0,
1566
- "period_change": 0
1567
- }
1568
- }
1569
- },
1570
- "date_and_spec": {
1571
- "anomaly": {
1572
- "correct": 2,
1573
- "n": 560,
1574
- "pct": 0.4,
1575
- "unsupported": 17,
1576
- "did_not_run": 486
1577
- },
1578
- "period_change": {
1579
- "correct": 66,
1580
- "n": 560,
1581
- "pct": 11.8,
1582
- "unsupported": 24,
1583
- "did_not_run": 203
1584
- },
1585
- "per_database_correct_of_80": {
1586
- "chinook": {
1587
- "anomaly": 1,
1588
- "period_change": 11
1589
- },
1590
- "hr": {
1591
- "anomaly": 1,
1592
- "period_change": 14
1593
- },
1594
- "adventureworks": {
1595
- "anomaly": 0,
1596
- "period_change": 8
1597
- },
1598
- "northwind": {
1599
- "anomaly": 0,
1600
- "period_change": 6
1601
- },
1602
- "nycflights": {
1603
- "anomaly": 0,
1604
- "period_change": 6
1605
- },
1606
- "pagila": {
1607
- "anomaly": 0,
1608
- "period_change": 14
1609
- },
1610
- "synthea": {
1611
- "anomaly": 0,
1612
- "period_change": 7
1613
- }
1614
- }
1615
- },
1616
- "date_spec_templates": {
1617
- "anomaly": {
1618
- "correct": 104,
1619
- "n": 560,
1620
- "pct": 18.6,
1621
- "unsupported": 21,
1622
- "did_not_run": 227
1623
- },
1624
- "period_change": {
1625
- "correct": 57,
1626
- "n": 560,
1627
- "pct": 10.2,
1628
- "unsupported": 9,
1629
- "did_not_run": 204
1630
- },
1631
- "per_database_correct_of_80": {
1632
- "chinook": {
1633
- "anomaly": 26,
1634
- "period_change": 18
1635
- },
1636
- "hr": {
1637
- "anomaly": 16,
1638
- "period_change": 8
1639
- },
1640
- "adventureworks": {
1641
- "anomaly": 7,
1642
- "period_change": 4
1643
- },
1644
- "northwind": {
1645
- "anomaly": 8,
1646
- "period_change": 3
1647
- },
1648
- "nycflights": {
1649
- "anomaly": 25,
1650
- "period_change": 7
1651
- },
1652
- "pagila": {
1653
- "anomaly": 13,
1654
- "period_change": 9
1655
- },
1656
- "synthea": {
1657
- "anomaly": 9,
1658
- "period_change": 8
1659
- }
1660
- }
1661
- }
1662
  }
1663
  }
 
1
  {
2
  "meta": {
3
+ "benchmark": "App text-to-SQL held-out sample for open models under 10B parameters, PostgreSQL execution accuracy",
 
4
  "sets": [
5
  "chinook",
6
  "hr",
 
10
  "pagila",
11
  "synthea"
12
  ],
13
+ "per_set": "240 prompts: 80 each of select, detail and unanswerable",
14
+ "sample_rule": "Per database, first 80 indices per category ordered by sha256(\"database|index\"); independent of model outputs.",
15
+ "sample_files_sha256": "b87f4fbcc2c63425c75030d50e8b92d7cfdc9a6516bc50afc6ce0c3c18cb6354",
16
+ "sample_files_note": "The hashed sample files hold 400 prompts per database; this release scores the 240 select, detail and unanswerable prompts among them.",
 
 
 
 
 
 
 
17
  "scoring": "Generated SQL is executed on PostgreSQL and its result table is compared with the gold query result; row order counts only for gold queries marked as ordered. An unanswerable prompt counts as handled when the output does not run as SQL.",
18
  "prompts": "Nemotron, Qwen3.5-9B and Qwen3.5-0.8B share one prompt text. SLM-SQL used its own published template with the same schema and question and no unanswerable instruction.",
19
+ "inference": {
20
+ "Qwen3.5-9B (zero-shot, thinking off)": {
21
+ "endpoint": "self-hosted OpenAI-compatible server",
22
+ "model": "Qwen/Qwen3.5-9B",
23
+ "chat_template_kwargs": {
24
+ "enable_thinking": false
25
+ },
26
+ "temperature": 0,
27
+ "max_tokens": 1024,
28
+ "samples": 1,
29
+ "failed_requests_after_retries": 0,
30
+ "serving_precision": "not recorded",
31
+ "date": "2026-10-29"
32
+ },
33
+ "Nemotron 3.5 Bolt E4B (zero-shot)": {
34
+ "endpoint": "NVIDIA API catalog integrate.api.nvidia.com/v1/chat/completions",
35
+ "model": "nvidia/nemotron-3.5-bolt-e4b",
36
+ "temperature": 0,
37
+ "max_tokens": 1024,
38
+ "samples": 1,
39
+ "reasoning_tokens_returned": 0,
40
+ "failed_requests": 0,
41
+ "date": "2026-10-27"
42
+ }
43
+ },
44
+ "paired_tests_qwen35_9b_vs_nemotron": {
45
  "select_plus_detail": {
46
  "qwen35_9b_only_correct": 119,
47
  "nemotron_only_correct": 46,
 
52
  "nemotron_only": 97,
53
  "exact_mcnemar_p": 2.8e-14
54
  }
55
+ }
 
56
  },
57
  "results": {
58
+ "Qwen3.5-9B (zero-shot, thinking off)": {
59
  "sets_scored": [
60
  "chinook",
61
  "hr",
 
67
  ],
68
  "overall": {
69
  "select": {
70
+ "correct": 387,
71
  "n": 560,
72
+ "pct": 69.1
73
  },
74
  "detail": {
75
+ "correct": 262,
 
 
 
 
 
 
 
 
 
 
76
  "n": 560,
77
+ "pct": 46.8
78
  },
79
  "select_plus_detail": {
80
+ "correct": 649,
81
  "n": 1120,
82
+ "pct": 57.9
 
 
 
 
 
83
  },
84
  "unanswerable_no_runnable_sql": {
85
+ "count": 407,
86
  "n": 560,
87
+ "pct": 72.7
88
  }
89
  },
90
+ "explicit_unsupported_marker": {
91
+ "on_unanswerable": "287/560",
92
+ "on_answerable": "5/1120"
93
+ },
94
  "per_set": {
95
  "chinook": {
96
  "select": {
97
+ "correct": 63,
98
  "n": 80,
99
+ "pct": 78.8
100
  },
101
  "detail": {
102
+ "correct": 50,
 
 
 
 
 
 
 
 
 
 
103
  "n": 80,
104
+ "pct": 62.5
105
  },
106
  "select_plus_detail": {
107
+ "correct": 113,
108
  "n": 160,
109
+ "pct": 70.6
 
 
 
 
 
110
  },
111
  "unanswerable_no_runnable_sql": {
112
+ "count": 50,
113
  "n": 80,
114
+ "pct": 62.5
115
  }
116
  },
117
  "hr": {
 
121
  "pct": 56.2
122
  },
123
  "detail": {
124
+ "correct": 35,
 
 
 
 
 
 
 
 
 
 
125
  "n": 80,
126
+ "pct": 43.8
127
  },
128
  "select_plus_detail": {
129
+ "correct": 80,
130
  "n": 160,
131
+ "pct": 50.0
 
 
 
 
 
132
  },
133
  "unanswerable_no_runnable_sql": {
134
+ "count": 52,
135
  "n": 80,
136
+ "pct": 65.0
137
  }
138
  },
139
  "adventureworks": {
140
  "select": {
141
+ "correct": 52,
142
  "n": 80,
143
+ "pct": 65.0
144
  },
145
  "detail": {
146
+ "correct": 35,
 
 
 
 
 
 
 
 
 
 
147
  "n": 80,
148
+ "pct": 43.8
149
  },
150
  "select_plus_detail": {
151
+ "correct": 87,
152
  "n": 160,
153
+ "pct": 54.4
 
 
 
 
 
154
  },
155
  "unanswerable_no_runnable_sql": {
156
+ "count": 56,
157
  "n": 80,
158
+ "pct": 70.0
159
  }
160
  },
161
  "northwind": {
162
  "select": {
163
+ "correct": 59,
164
  "n": 80,
165
+ "pct": 73.8
166
  },
167
  "detail": {
168
+ "correct": 46,
 
 
 
 
 
 
 
 
 
 
169
  "n": 80,
170
+ "pct": 57.5
171
  },
172
  "select_plus_detail": {
173
+ "correct": 105,
174
  "n": 160,
175
+ "pct": 65.6
 
 
 
 
 
176
  },
177
  "unanswerable_no_runnable_sql": {
178
+ "count": 61,
179
  "n": 80,
180
+ "pct": 76.2
181
  }
182
  },
183
  "nycflights": {
184
  "select": {
185
+ "correct": 53,
186
  "n": 80,
187
+ "pct": 66.2
188
  },
189
  "detail": {
190
+ "correct": 33,
 
 
 
 
 
 
 
 
 
 
191
  "n": 80,
192
+ "pct": 41.2
193
  },
194
  "select_plus_detail": {
195
+ "correct": 86,
196
  "n": 160,
197
+ "pct": 53.8
 
 
 
 
 
198
  },
199
  "unanswerable_no_runnable_sql": {
200
+ "count": 65,
201
  "n": 80,
202
+ "pct": 81.2
203
  }
204
  },
205
  "pagila": {
206
  "select": {
207
+ "correct": 55,
208
  "n": 80,
209
+ "pct": 68.8
210
  },
211
  "detail": {
212
+ "correct": 36,
 
 
 
 
 
 
 
 
 
 
213
  "n": 80,
214
+ "pct": 45.0
215
  },
216
  "select_plus_detail": {
217
+ "correct": 91,
218
  "n": 160,
219
+ "pct": 56.9
 
 
 
 
 
220
  },
221
  "unanswerable_no_runnable_sql": {
222
+ "count": 54,
223
  "n": 80,
224
+ "pct": 67.5
225
  }
226
  },
227
  "synthea": {
228
  "select": {
229
+ "correct": 60,
230
  "n": 80,
231
+ "pct": 75.0
232
  },
233
  "detail": {
234
+ "correct": 27,
 
 
 
 
 
 
 
 
 
 
235
  "n": 80,
236
+ "pct": 33.8
237
  },
238
  "select_plus_detail": {
239
+ "correct": 87,
240
  "n": 160,
241
+ "pct": 54.4
 
 
 
 
 
242
  },
243
  "unanswerable_no_runnable_sql": {
244
+ "count": 69,
245
  "n": 80,
246
+ "pct": 86.2
247
  }
248
  }
 
 
 
 
249
  }
250
  },
251
+ "Nemotron 3.5 Bolt E4B (zero-shot)": {
252
  "sets_scored": [
253
  "chinook",
254
  "hr",
 
260
  ],
261
  "overall": {
262
  "select": {
263
+ "correct": 339,
264
  "n": 560,
265
+ "pct": 60.5
266
  },
267
  "detail": {
268
+ "correct": 237,
 
 
 
 
 
 
 
 
 
 
269
  "n": 560,
270
+ "pct": 42.3
271
  },
272
  "select_plus_detail": {
273
+ "correct": 576,
274
  "n": 1120,
275
+ "pct": 51.4
 
 
 
 
 
276
  },
277
  "unanswerable_no_runnable_sql": {
278
+ "count": 486,
279
  "n": 560,
280
+ "pct": 86.8
281
  }
282
  },
283
  "explicit_unsupported_marker": {
284
+ "on_unanswerable": "387/560",
285
+ "on_answerable": "41/1120"
286
  },
287
  "per_set": {
288
  "chinook": {
289
  "select": {
290
+ "correct": 55,
291
  "n": 80,
292
+ "pct": 68.8
293
  },
294
  "detail": {
295
+ "correct": 46,
 
 
 
 
 
 
 
 
 
 
296
  "n": 80,
297
+ "pct": 57.5
298
  },
299
  "select_plus_detail": {
300
+ "correct": 101,
301
  "n": 160,
302
+ "pct": 63.1
 
 
 
 
 
303
  },
304
  "unanswerable_no_runnable_sql": {
305
+ "count": 66,
306
  "n": 80,
307
+ "pct": 82.5
308
  }
309
  },
310
  "hr": {
 
314
  "pct": 56.2
315
  },
316
  "detail": {
317
+ "correct": 44,
 
 
 
 
 
 
 
 
 
 
318
  "n": 80,
319
+ "pct": 55.0
320
  },
321
  "select_plus_detail": {
322
+ "correct": 89,
323
  "n": 160,
324
+ "pct": 55.6
 
 
 
 
 
325
  },
326
  "unanswerable_no_runnable_sql": {
327
+ "count": 71,
328
  "n": 80,
329
+ "pct": 88.8
330
  }
331
  },
332
  "adventureworks": {
333
  "select": {
334
+ "correct": 44,
335
  "n": 80,
336
+ "pct": 55.0
337
  },
338
  "detail": {
339
+ "correct": 28,
 
 
 
 
 
 
 
 
 
 
340
  "n": 80,
341
+ "pct": 35.0
342
  },
343
  "select_plus_detail": {
344
+ "correct": 72,
345
  "n": 160,
346
+ "pct": 45.0
 
 
 
 
 
347
  },
348
  "unanswerable_no_runnable_sql": {
349
+ "count": 66,
350
  "n": 80,
351
+ "pct": 82.5
352
  }
353
  },
354
  "northwind": {
355
  "select": {
356
+ "correct": 52,
357
  "n": 80,
358
+ "pct": 65.0
359
  },
360
  "detail": {
361
+ "correct": 45,
 
 
 
 
 
 
 
 
 
 
362
  "n": 80,
363
+ "pct": 56.2
364
  },
365
  "select_plus_detail": {
366
+ "correct": 97,
367
  "n": 160,
368
+ "pct": 60.6
 
 
 
 
 
369
  },
370
  "unanswerable_no_runnable_sql": {
371
+ "count": 69,
372
  "n": 80,
373
+ "pct": 86.2
374
  }
375
  },
376
  "nycflights": {
377
  "select": {
378
+ "correct": 50,
379
  "n": 80,
380
+ "pct": 62.5
381
  },
382
  "detail": {
383
+ "correct": 27,
 
 
 
 
 
 
 
 
 
 
384
  "n": 80,
385
+ "pct": 33.8
386
  },
387
  "select_plus_detail": {
388
+ "correct": 77,
389
  "n": 160,
390
+ "pct": 48.1
 
 
 
 
 
391
  },
392
  "unanswerable_no_runnable_sql": {
393
+ "count": 74,
394
  "n": 80,
395
+ "pct": 92.5
396
  }
397
  },
398
  "pagila": {
399
  "select": {
400
+ "correct": 50,
401
  "n": 80,
402
+ "pct": 62.5
403
  },
404
  "detail": {
405
+ "correct": 27,
 
 
 
 
 
 
 
 
 
 
406
  "n": 80,
407
+ "pct": 33.8
408
  },
409
  "select_plus_detail": {
410
+ "correct": 77,
411
  "n": 160,
412
+ "pct": 48.1
 
 
 
 
 
413
  },
414
  "unanswerable_no_runnable_sql": {
415
+ "count": 67,
416
  "n": 80,
417
+ "pct": 83.8
418
  }
419
  },
420
  "synthea": {
421
  "select": {
422
+ "correct": 43,
423
  "n": 80,
424
+ "pct": 53.8
425
  },
426
  "detail": {
427
+ "correct": 20,
 
 
 
 
 
 
 
 
 
 
428
  "n": 80,
429
+ "pct": 25.0
430
  },
431
  "select_plus_detail": {
432
+ "correct": 63,
433
  "n": 160,
434
+ "pct": 39.4
 
 
 
 
 
435
  },
436
  "unanswerable_no_runnable_sql": {
437
+ "count": 73,
438
  "n": 80,
439
+ "pct": 91.2
440
  }
441
  }
442
  }
 
462
  "n": 560,
463
  "pct": 33.4
464
  },
 
 
 
 
 
 
 
 
 
 
465
  "select_plus_detail": {
466
  "correct": 469,
467
  "n": 1120,
468
  "pct": 41.9
469
  },
 
 
 
 
 
470
  "unanswerable_no_runnable_sql": {
471
  "count": 323,
472
  "n": 560,
 
485
  "n": 80,
486
  "pct": 56.2
487
  },
 
 
 
 
 
 
 
 
 
 
488
  "select_plus_detail": {
489
  "correct": 98,
490
  "n": 160,
491
  "pct": 61.2
492
  },
 
 
 
 
 
493
  "unanswerable_no_runnable_sql": {
494
  "count": 35,
495
  "n": 80,
 
507
  "n": 80,
508
  "pct": 37.5
509
  },
 
 
 
 
 
 
 
 
 
 
510
  "select_plus_detail": {
511
  "correct": 63,
512
  "n": 160,
513
  "pct": 39.4
514
  },
 
 
 
 
 
515
  "unanswerable_no_runnable_sql": {
516
  "count": 42,
517
  "n": 80,
 
529
  "n": 80,
530
  "pct": 33.8
531
  },
 
 
 
 
 
 
 
 
 
 
532
  "select_plus_detail": {
533
  "correct": 61,
534
  "n": 160,
535
  "pct": 38.1
536
  },
 
 
 
 
 
537
  "unanswerable_no_runnable_sql": {
538
  "count": 58,
539
  "n": 80,
 
551
  "n": 80,
552
  "pct": 41.2
553
  },
 
 
 
 
 
 
 
 
 
 
554
  "select_plus_detail": {
555
  "correct": 81,
556
  "n": 160,
557
  "pct": 50.6
558
  },
 
 
 
 
 
559
  "unanswerable_no_runnable_sql": {
560
  "count": 49,
561
  "n": 80,
 
573
  "n": 80,
574
  "pct": 31.2
575
  },
 
 
 
 
 
 
 
 
 
 
576
  "select_plus_detail": {
577
  "correct": 68,
578
  "n": 160,
579
  "pct": 42.5
580
  },
 
 
 
 
 
581
  "unanswerable_no_runnable_sql": {
582
  "count": 49,
583
  "n": 80,
 
595
  "n": 80,
596
  "pct": 26.2
597
  },
 
 
 
 
 
 
 
 
 
 
598
  "select_plus_detail": {
599
  "correct": 61,
600
  "n": 160,
601
  "pct": 38.1
602
  },
 
 
 
 
 
603
  "unanswerable_no_runnable_sql": {
604
  "count": 43,
605
  "n": 80,
 
617
  "n": 80,
618
  "pct": 7.5
619
  },
 
 
 
 
 
 
 
 
 
 
620
  "select_plus_detail": {
621
  "correct": 37,
622
  "n": 160,
623
  "pct": 23.1
624
  },
 
 
 
 
 
625
  "unanswerable_no_runnable_sql": {
626
  "count": 47,
627
  "n": 80,
 
651
  "n": 560,
652
  "pct": 23.4
653
  },
 
 
 
 
 
 
 
 
 
 
654
  "select_plus_detail": {
655
  "correct": 341,
656
  "n": 1120,
657
  "pct": 30.4
658
  },
 
 
 
 
 
659
  "unanswerable_no_runnable_sql": {
660
  "count": 333,
661
  "n": 560,
 
674
  "n": 80,
675
  "pct": 42.5
676
  },
 
 
 
 
 
 
 
 
 
 
677
  "select_plus_detail": {
678
  "correct": 74,
679
  "n": 160,
680
  "pct": 46.2
681
  },
 
 
 
 
 
682
  "unanswerable_no_runnable_sql": {
683
  "count": 31,
684
  "n": 80,
 
696
  "n": 80,
697
  "pct": 33.8
698
  },
 
 
 
 
 
 
 
 
 
 
699
  "select_plus_detail": {
700
  "correct": 63,
701
  "n": 160,
702
  "pct": 39.4
703
  },
 
 
 
 
 
704
  "unanswerable_no_runnable_sql": {
705
  "count": 43,
706
  "n": 80,
 
718
  "n": 80,
719
  "pct": 12.5
720
  },
 
 
 
 
 
 
 
 
 
 
721
  "select_plus_detail": {
722
  "correct": 23,
723
  "n": 160,
724
  "pct": 14.4
725
  },
 
 
 
 
 
726
  "unanswerable_no_runnable_sql": {
727
  "count": 72,
728
  "n": 80,
 
740
  "n": 80,
741
  "pct": 30.0
742
  },
 
 
 
 
 
 
 
 
 
 
743
  "select_plus_detail": {
744
  "correct": 58,
745
  "n": 160,
746
  "pct": 36.2
747
  },
 
 
 
 
 
748
  "unanswerable_no_runnable_sql": {
749
  "count": 50,
750
  "n": 80,
 
762
  "n": 80,
763
  "pct": 20.0
764
  },
 
 
 
 
 
 
 
 
 
 
765
  "select_plus_detail": {
766
  "correct": 52,
767
  "n": 160,
768
  "pct": 32.5
769
  },
 
 
 
 
 
770
  "unanswerable_no_runnable_sql": {
771
  "count": 46,
772
  "n": 80,
 
784
  "n": 80,
785
  "pct": 21.2
786
  },
 
 
 
 
 
 
 
 
 
 
787
  "select_plus_detail": {
788
  "correct": 48,
789
  "n": 160,
790
  "pct": 30.0
791
  },
 
 
 
 
 
792
  "unanswerable_no_runnable_sql": {
793
  "count": 39,
794
  "n": 80,
 
806
  "n": 80,
807
  "pct": 3.8
808
  },
 
 
 
 
 
 
 
 
 
 
809
  "select_plus_detail": {
810
  "correct": 23,
811
  "n": 160,
812
  "pct": 14.4
813
  },
 
 
 
 
 
814
  "unanswerable_no_runnable_sql": {
815
  "count": 52,
816
  "n": 80,
 
838
  "n": 400,
839
  "pct": 13.0
840
  },
 
 
 
 
 
 
 
 
 
 
841
  "select_plus_detail": {
842
  "correct": 160,
843
  "n": 800,
844
  "pct": 20.0
845
  },
 
 
 
 
 
846
  "unanswerable_no_runnable_sql": {
847
  "count": 279,
848
  "n": 400,
 
861
  "n": 80,
862
  "pct": 17.5
863
  },
 
 
 
 
 
 
 
 
 
 
864
  "select_plus_detail": {
865
  "correct": 34,
866
  "n": 160,
867
  "pct": 21.2
868
  },
 
 
 
 
 
869
  "unanswerable_no_runnable_sql": {
870
  "count": 55,
871
  "n": 80,
 
883
  "n": 80,
884
  "pct": 10.0
885
  },
 
 
 
 
 
 
 
 
 
 
886
  "select_plus_detail": {
887
  "correct": 31,
888
  "n": 160,
889
  "pct": 19.4
890
  },
 
 
 
 
 
891
  "unanswerable_no_runnable_sql": {
892
  "count": 42,
893
  "n": 80,
 
905
  "n": 80,
906
  "pct": 13.8
907
  },
 
 
 
 
 
 
 
 
 
 
908
  "select_plus_detail": {
909
  "correct": 33,
910
  "n": 160,
911
  "pct": 20.6
912
  },
 
 
 
 
 
913
  "unanswerable_no_runnable_sql": {
914
  "count": 61,
915
  "n": 80,
 
927
  "n": 80,
928
  "pct": 16.2
929
  },
 
 
 
 
 
 
 
 
 
 
930
  "select_plus_detail": {
931
  "correct": 40,
932
  "n": 160,
933
  "pct": 25.0
934
  },
 
 
 
 
 
935
  "unanswerable_no_runnable_sql": {
936
  "count": 64,
937
  "n": 80,
 
949
  "n": 80,
950
  "pct": 7.5
951
  },
 
 
 
 
 
 
 
 
 
 
952
  "select_plus_detail": {
953
  "correct": 22,
954
  "n": 160,
955
  "pct": 13.8
956
  },
 
 
 
 
 
957
  "unanswerable_no_runnable_sql": {
958
  "count": 57,
959
  "n": 80,
 
962
  }
963
  }
964
  }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
965
  }
966
  }