sabin1234/Real_Scarp_Romanized_FINANCE_FAQ_Compile_Dataset

Romanized Nepali Finance FAQ — Instruction-Following Dataset (Latin script)

A Romanized (Latin-script) Nepali question–answer dataset for the finance domain of Nepal: banking, central-bank policy, digital payments, stock market (NEPSE), dividends, mutual funds, pensions, provident fund, insurance, microfinance and cooperatives. Every record is a single-turn conversation (one question, one answer) in ShareHub/ShareGPT-style format, together with full provenance and generation metadata.

This dataset is the row-by-row romanized counterpart of the Devanagari file finance_nepali_faq.jsonl: the 6,635 rows have the same ids in the same order and identical metadata; only the text of the question and answer is transliterated from Devanagari into lower-case Latin letters (e.g. एडीबीएल अनलाइन खाताको लागि को ले आवेदन दिन सक्छ? → edibiela analaina khatako lagi ko le aawedana din sakchha?).

• Rows: 6,635 (single train split)
• Language: Nepali (ne / npi), written in Latin script (romanized) — see the script-metadata warning
• Format: JSON Lines (.jsonl), UTF-8, ~10.7 MB
• Task type: instruction-following (FAQ-style question answering, single turn)
• Origin type: real (FAQ pages / market-data records rewritten into Nepali, then romanized)
• Licence: Apache-2.0 for 6,335 rows; 300 rows (Lemonn) have licence unknown

---

1. Dataset at a glance

Property	Value
Total conversations	6,635
Turns per conversation	2 (1 human + 1 gpt) in 100% of rows
Text script (actual)	Latin, lower-case (36 rows keep upper-case API/AI)
Script declared in metadata	Deva / देवनागरी (does not match the text, see warning below)
Distinct sources (source)	43
Distinct source_name values	44
Distinct categories (generation_category)	56
Distinct sub-domains (generation_sub_domain)	122
Distinct behaviors	2
Distinct question-type labels	3
Unique questions	6,610
Unique (question, answer) pairs	6,631
Mean question length	75 characters
Mean answer length	208 characters (29 words)
Alphabetic vocabulary (romanized)	9,884 distinct words
Questions that end with ?	99.9%

Script-metadata warning

The row-level fields script (Deva) and the metadata keys content_script (देवनागरी), content_language (नेपाली) and behavior_definition (written in Devanagari Nepali) were copied unchanged from the Devanagari release. The text in conversations, however, is Latin script. Do not rely on script to detect the writing system — use the code in §3.6 or override it to Latn.

The corpus has three sub-corpora with very different character:

Sub-corpus	Rows	Share	What it is
ShareHub (market data QA)	2,448	36.9%	Templated questions about listed securities (stocks, debentures, mutual funds): price, market cap, financial ratios, listed shares, etc.
NEPSE Dividend History	2,000	30.1%	Templated questions about cash/bonus dividend percentages per company per fiscal year (Bikram Sambat).
Institutional FAQ (41 sources)	2,187	33.0%	Natural, human-style FAQ pairs from banks, regulators, wallets, insurers, pension funds and investing-education sites.

Heads-up: ~67% of rows are templated structured-data QA; ~33% are free-form FAQ. Filter to the FAQ part for conversational diversity (see §3.7).

2. Data format and schema

One JSON object per line. Example record:

{
  "id": "adbl_online_account_faq_000001",
  "conversations": [
    {
      "from": "human",
      "value": "edibiela analaina khatako lagi ko le aawedana din sakchha?"
    },
    {
      "from": "gpt",
      "value": "kunai pani nepali wyakti ra enaaaraena wyaktiharu jasle krishi wikas bainkasanga wyaktigat bachat/chalu khata kholna chahanchhan, uniharule aawedana din sakchhan."
    }
  ],
  "source": "Agricultural Development Bank Limited",
  "source_name": "adbl_online_account_faq",
  "source_repo": "Agricultural Development Bank Limited",
  "source_config": "default",
  "source_split": "train",
  "source_revision": "v1",
  "source_row_id": "adbl_online_account_faq_000001",
  "language": "ne",
  "language_code": "npi",
  "script": "Deva",
  "license": "Apache-2.0",
  "license_tier": "permissive",
  "task_type": "instruction-following",
  "generation_type": "real",
  "condition": "real",
  "url": "",
  "metadata_json": "{\"generation_domain\": \"बैंकिङ\", \"generation_category\": \"अनलाइन खाता सम्बन्धी प्रायः सोधिने प्रश्न\", \"generation_sub_domain\": \"एडीबीएल अनलाइन खाता\", \"behavior\": \"जानकारीमूलक उत्तर दिने\", \"behavior_definition\": \"मूल प्रश्नको अर्थ र मूल उत्तरको तथ्य कायम राख्दै नेपालीमा रूपान्तरण गरिएको उत्तर दिने।\", \"question_type\": \"तथ्यमा आधारित प्रश्न\", \"content_language\": \"नेपाली\", \"content_script\": \"देवनागरी\", \"english_content_allowed\": false}",
  "g": "Commercial / development banks (retail banking)"
}

2.1 Top-level fields

Field	Type	Description	Observed values
id	string	Unique row id (<source_name>_<6-digit index>); identical to the id of the same row in the Devanagari file	6,635 unique
conversations	list[object]	Always [{"from":"human","value":Q},{"from":"gpt","value":A}] — romanized text	2 turns, always human → gpt
source	string	Human-readable publisher / organisation (mixed English and Devanagari Nepali — not romanized)	43 values
source_name	string	Machine-friendly dataset slug (e.g. nepal_rastra_bank_faq)	44 values (the Agricultural Development Bank source has two slugs)
source_repo	string	Upstream repository / organisation label	43 values
source_config	string	Upstream config	default
source_split	string	Upstream split	train
source_revision	string	Upstream revision	v1
source_row_id	string	Row id in the upstream source (equal to id)	—
language	string	ISO 639-1 language	ne
language_code	string	ISO 639-3 language	npi
script	string	Declared script — inaccurate for this release (text is Latin)	Deva in all rows
license	string	Row-level licence	Apache-2.0 (6,335) / unknown (300)
license_tier	string	Coarse licence class	permissive (6,335) / unknown (300)
task_type	string	Task label	instruction-following
generation_type	string	How the data was produced	real
condition	string	Experimental condition tag	real
url	string	Source URL (empty in this release)	empty string in all rows
metadata_json	string (JSON)	A JSON-encoded string with generation metadata — parse with json.loads; values are in Devanagari Nepali (copied from the Devanagari release)	see 2.2

2.2 metadata_json keys

metadata_json is a string, not a nested object, and it is byte-identical to the Devanagari file's. Two key-sets exist:

Key	Present in	Description
generation_domain	all rows	Coarse domain (8 values, Devanagari: वित्त, वित्तीय सेवा, बैंकिङ, सहकारी, डिजिटल भुक्तानी, बीमा, लघुवित्त, बैंकिङ तथा पूँजी बजार)
generation_category	all rows	Topic category (56 values) — usually “<topic> सम्बन्धी प्रायः सोधिने प्रश्न” (“FAQ about ”)
generation_sub_domain	all rows	Finer topic / sector (122 values; ShareHub sectors are English, others Devanagari)
behavior	all rows	Assistant behavior label (2 values, §7)
behavior_definition	all rows	Natural-language definition of the behavior (4 wording variants, Devanagari)
question_type	all rows	Question-type label (3 values, §8)
content_language	all rows	नेपाली (Nepali)
content_script	all rows	देवनागरी (Devanagari) — not true for the text of this release
english_content_allowed	all rows	false
security_type	ShareHub rows only (2,448)	stock / debenture / mutual_fund
original_symbol	ShareHub rows only (2,448)	NEPSE ticker of the security (415 distinct symbols), Latin upper-case

⚠️ security_type and original_symbol exist only for ShareHub rows, so metadata key sets differ between rows (4,187 rows have 9 keys, 2,448 rows have 11). This matters when flattening with some libraries — see the Hugging Face demo.

---

3. How to load the dataset (code demos)

Assume the file is saved as finance_romanized_faq.jsonl in your working directory. The snippets were executed against this file.

3.1 Plain Python (no dependencies)

import json

rows = []
with open("finance_romanized_faq.jsonl", encoding="utf-8") as f:
    for line in f:
        rows.append(json.loads(line))

print(len(rows))                       # 6635
r = rows[0]
question = r["conversations"][0]["value"]
answer   = r["conversations"][1]["value"]
meta     = json.loads(r["metadata_json"])   # metadata_json is a JSON *string*

print(question)   # edibiela analaina khatako lagi ko le aawedana din sakchha?
print(answer)
print(meta["behavior"], "|", meta["question_type"], "|", meta["generation_category"])

3.2 pandas (recommended for exploration)

import json
import pandas as pd

df = pd.read_json("finance_romanized_faq.jsonl", lines=True)

# expand metadata_json into columns (missing keys become NaN)
meta = pd.json_normalize(df["metadata_json"].map(json.loads))
df = pd.concat([df.drop(columns=["metadata_json"]), meta], axis=1)

df["question"] = df["conversations"].str[0].str["value"]
df["answer"]   = df["conversations"].str[1].str["value"]

print(df.shape)                                  # (6635, 31)
print(df["behavior"].value_counts())
print(df["source_name"].value_counts().head(10))

3.3 Hugging Face datasets

import json
from datasets import load_dataset

ds = load_dataset("json", data_files="finance_romanized_faq.jsonl", split="train")
print(ds)          # 6,635 rows

# IMPORTANT: metadata keys differ between rows (security_type / original_symbol
# exist only for ShareHub rows). Always return the SAME keys from .map(),
# and use "" instead of None, otherwise Arrow raises a null-column / KeyError.
def flatten(r):
    m = json.loads(r["metadata_json"])
    return {
        "question":        r["conversations"][0]["value"],
        "answer":          r["conversations"][1]["value"],
        "behavior":        m["behavior"],
        "question_type":   m["question_type"],
        "domain":          m["generation_domain"],
        "category":        m["generation_category"],
        "sub_domain":      m["generation_sub_domain"],
        "security_type":   m.get("security_type", ""),
        "original_symbol": m.get("original_symbol", ""),
    }

ds = ds.map(flatten)
splits = ds.train_test_split(test_size=0.1, seed=42)   # the file ships only a train split
print(splits)

3.4 Join with the Devanagari original (parallel corpus)

Both files share id and row order, so they form a Devanagari ↔ Latin parallel corpus (useful for transliteration models or for mixed-script training).

import pandas as pd

rom = pd.read_json("finance_romanized_faq.jsonl", lines=True)
dev = pd.read_json("finance_nepali_faq.jsonl",   lines=True)   # the Devanagari release

for d in (rom, dev):
    d["question"] = d["conversations"].str[0].str["value"]
    d["answer"]   = d["conversations"].str[1].str["value"]

pair = rom[["id", "question", "answer"]].merge(
    dev[["id", "question", "answer"]], on="id", suffixes=("_rom", "_dev"), validate="one_to_one"
)
print(pair.shape)                                   # (6635, 5)
print(pair.loc[0, "question_rom"])                  # edibiela analaina khatako lagi ko le aawedana din sakchha?
print(pair.loc[0, "question_dev"])                  # एडीबीएल अनलाइन खाताको लागि को ले आवेदन दिन सक्छ?

3.5 Light-weight text normalisation

The text is clean ASCII for the most part, but a few artifacts exist (§11.3): zero-width joiners (U+200D), two stray Devanagari characters, typographic quotes/dashes. This helper removes them:

import re, unicodedata

def normalize(text: str) -> str:
    text = text.replace("\u200d", "").replace("\u200c", "")      # zero-width joiner/non-joiner
    text = re.sub(r"[\u0900-\u097F]", "", text)                  # stray Devanagari marks (2 rows)
    text = (text.replace("“", '"').replace("”", '"')
                .replace("‘", "'").replace("’", "'")
                .replace("–", "-").replace("—", "-"))
    text = unicodedata.normalize("NFKC", text)
    return re.sub(r"\s+", " ", text).strip()

df["question_clean"] = df["question"].map(normalize)
df["answer_clean"]   = df["answer"].map(normalize)

# a simple romanized word tokenizer (lower-case words and numbers)
tok = lambda s: re.findall(r"[a-z]+|\d+(?:[.,/]\d+)*", s.lower())
print(tok(df.loc[0, "answer"])[:8])   # ['kunai', 'pani', 'nepali', 'wyakti', 'ra', 'enaaaraena', ...]

3.6 Detect script and fix the script field

import re
from collections import Counter

LATIN = re.compile(r"[A-Za-z]"); DEVA = re.compile(r"[\u0900-\u097F]")
print(Counter("Latn" if LATIN.search(a) and not DEVA.search(a) else "other" for a in df["answer"]))
# Counter({'Latn': 6633, 'other': 2})   <- the 2 rows contain stray Devanagari marks

df["script"] = "Latn"          # override the (incorrect) declared 'Deva'

3.7 Filtering by sub-corpus, source or label

import json

for r in rows:                       # parse once, keep the dict next to the row
    r["meta"] = json.loads(r["metadata_json"])

# a) only the free-form FAQ part (drops templated ShareHub + dividend rows)
faq = [r for r in rows if r["source_name"] not in ("sharehub_faq", "nepse_dividend_faq")]
print(len(faq))                                                   # 2187

# b) only ShareHub debentures
debentures = [r for r in rows if r["meta"].get("security_type") == "debenture"]
print(len(debentures))                                            # 440

# c) only procedural questions ("how do I ...")
procedural = [r for r in rows if r["meta"]["question_type"] == "प्रक्रियामूलक प्रश्न"]
print(len(procedural))                                            # 77

# d) only rows with a permissive licence (drops the 300 'unknown' Lemonn rows)
permissive = [r for r in rows if r["license_tier"] == "permissive"]
print(len(permissive))                                            # 6335

# e) a single source
nrb = [r for r in rows if r["source_name"] == "nepal_rastra_bank_faq"]
print(len(nrb))                                                   # 388

3.8 Chat / messages format, with a Romanized system prompt (for SFT)

ROLE = {"human": "user", "gpt": "assistant"}

def to_messages(r):
    return {"messages": [{"role": ROLE[t["from"]], "content": t["value"]}
                         for t in r["conversations"]]}

SYSTEM = "tapaain nepalko witta kshetrasambandhi prashnako uttara dine sahayak hunuhunchha."
chat = [{"messages": [{"role": "system", "content": SYSTEM}] + to_messages(r)["messages"]}
        for r in rows]
print(chat[0])

3.9 Source-stratified train / validation split

Every source has at least 3 rows, so stratify= works directly:

from sklearn.model_selection import train_test_split

train_df, val_df = train_test_split(
    df, test_size=0.1, random_state=42, stratify=df["source_name"]
)
print(len(train_df), len(val_df))                                 # 5971 664

3.10 Quick sanity check

from collections import Counter

assert all(len(r["conversations"]) == 2 for r in rows)
assert all([t["from"] for t in r["conversations"]] == ["human", "gpt"] for r in rows)
assert len({r["id"] for r in rows}) == len(rows)
print(Counter(r["license"] for r in rows))   # Counter({'Apache-2.0': 6335, 'unknown': 300})

---

4. Composition: the three sub-corpora

4.1 ShareHub — security-level market-data QA (2,448 rows)

Each of the 415 securities (tickers) is asked the same fixed set of questions; the answer is a romanized Nepali sentence that embeds numbers (prices, market cap, ratios, dates) and keeps the English company name/ticker in lower-case Latin. Answers are a point-in-time snapshot (e.g. septembar 17 samma = “as of September 17”).

Question intent	Romanized template	Rows	Share of ShareHub	By security type
Listed shares	<security> (<ticker>) ka kati sheyarharu nepsema suchikrit chhan?	415	17.0%	stock 283, debenture 75, mutual_fund 57
Market cap & valuation	<security> (<ticker>) ko kul bajar punjikaran (market kyap) ra mulyankana ke ho?	402	16.4%	stock 282, debenture 74, mutual_fund 46
Last traded price	<security> (<ticker>) ko pachhillo karobar mulya kati chha?	402	16.4%	stock 282, debenture 74, mutual_fund 46
Historical price trend	<security> (<ticker>) ko aitihasik mulya prawritti ra pradarshana kasto chha?	402	16.4%	stock 282, debenture 74, mutual_fund 46
Key financial indicators	<security> (<ticker>) ka mukhya wittiya suchakharu ra pradarshana mapadandaharu ke-ke hun?	401	16.4%	stock 283, debenture 75, mutual_fund 43
Open range (price band)	<security> (<ticker>) ko khula dayara kati ho?	238	9.7%	stock 133, debenture 68, mutual_fund 37
Promoter unlock date	<security> (<ticker>) ko prawarddhak analak miti kahile ho?	157	6.4%	stock 157
Maturity period (mutual funds)	<security> (<ticker>) ko paripakwata awadhi kati ho?	31	1.3%	mutual_fund 31

By security type

security_type	Rows	Share	Meaning
stock	1,702	69.5%	Listed company shares
debenture	440	18.0%	Non-convertible debentures / bonds (Rinpatra)
mutual_fund	306	12.5%	Mutual funds (closed-end & open-end SIP)

By sector / sub-domain

generation_sub_domain	Rows	Share
Hydro Power	743	30.4%
Non-Convertible Debenture	440	18.0%
Mutual Fund	295	12.1%
Microfinance	270	11.0%
Manufacturing And Processing	114	4.7%
Commercial Banks	95	3.9%
Development Banks	82	3.3%
Life Insurance	81	3.3%
Finance	75	3.1%
Non Life Insurance	69	2.8%
Others	62	2.5%
Hotels And Tourism	52	2.1%
Investment	47	1.9%
Tradings	12	0.5%
Open Ended Mutual Fund (SIP)	11	0.4%

Examples (romanized, verbatim)

Type	Question	Answer
debenture	agricultural bond (8 barsha) (adblb86) ka kati sheyarharu nepsema suchikrit chhan?	jamma 60,00,000 (60 lakh) sheyarharu agricultural bond (8 barsha) (adblb86) ka nepsema suchikrit chhan, jasamadhye prawarddhak ra sarwajanik sheyarko wiwaran upalabdha chhaina.
debenture	agricultural bond (8 barsha) (adblb86) ko pachhillo karobar mulya kati chha?	julai 27 samma, pachhillo karobar mulya ru. 1010 (0, 0%). aaja, sheyar khuleko mulya ru. 1010, uchcha ru. 1010, ra nyuna ru. 1010 thiyo. aghillo banda mulya ru. 1000 thiyo, karobar rakam ru. 50500 ra karobar matra 50 sheyarharu 1 karobaraharuma.
debenture	agricultural bond (8 barsha) (adblb86) ko aitihasik mulya prawritti ra pradarshana kasto chha?	gat barshama, agricultural bond (8 barsha) (adblb86) le 52-haptako uchcha ru. 1010 ra nyuna ru. 1000 dekheko chha. sarwakalina uchcha ru. 1010 julai 27, 2026 thiyo, jabaki sarwakalina nyuna ru. 0 janawari 25, 2024 thiyo. 120-dinko ausat mulya ru. 16.75, ra 180-dinko ausat ru. 11.17 chha. 1-barshako pratiphal hal 1% chha.
stock	arun kabeli power ltd. (akpl) ko prawarddhak analak miti kahile ho?	arun kabeli power ltd. (akpl) ko prawarddhak analak miti april 17, 2020 ho. yo miti kampanika prawarddhak (yogya prawarddhak, sthaniya janata, karmachari aadi) sheyarharuko lak-ina awadhiko antyalai janaunchha, jasle prawarddhakharulai stak eksachenjama swatantra rupma karobar garna anumati dinchha.
mutual_fund	citizens super 30 mutual fund (c30mf) ko paripakwata awadhi kati ho?	citizens super 30 mutual fund (c30mf) ko paripakwata awadhi 10 barsha ho. yo agasta 08, 2023 ma suchikrit bhaeko thiyo, ra me 14, 2033 ma paripakwa hunechha.
debenture	4% agricultural bond (adblb) ko khula dayara kati ho?	nepal stak eksachenja (nepse) le nirdharan gareko dishanirdeshaanusar, 4% agricultural bond (adblb) ko khula dayara ru. 1000 dekhi ru. 1000.

4.2 NEPSE Dividend History — company × fiscal-year dividend QA (2,000 rows)

Questions ask for either the total (cash + bonus) dividend % (kul labhansha) or the bonus-share dividend % (bonas sheyar labhansha) of a company in a given Bikram Sambat fiscal year (aarthik barsha 2077/2078). 211 distinct companies, 18 distinct fiscal years (2062/63 – 2082/83). Company tickers are written as letter-name transliterations (e.g. ADBL → edibiela, ACLBSL → esielabiesaela), which makes them long and unusual tokens (see §11.4). Answers are one short sentence ending in the percentage, e.g. … ko kul labhansha 8% thiyo.

Asked quantity	Surface form (3 syntactic variants)	Romanized template	Rows
Bonus-share dividend %	announcement form	<ticker> le aarthik barsha <FY> ka lagi … kati pratishat … ghoshana gar‍yo?	221
Bonus-share dividend %	ticker-first	<ticker> ko aarthik barsha <FY> ko … pratishat kati ho?	231
Bonus-share dividend %	year-first	aarthik barsha <FY> ma <ticker> ko … kati thiyo?	231
Total dividend %	announcement form	<ticker> le aarthik barsha <FY> ka lagi … kati pratishat … ghoshana gar‍yo?	448
Total dividend %	ticker-first	<ticker> ko aarthik barsha <FY> ko … pratishat kati ho?	401
Total dividend %	year-first	aarthik barsha <FY> ma <ticker> ko … kati thiyo?	468

Rows per fiscal year

Fiscal year (B.S.)	Rows	Fiscal year (B.S.)	Rows	Fiscal year (B.S.)	Rows
2062/2063	2	2071/2072	99	2077/2078	178
2066/2067	1	2072/2073	116	2078/2079	180
2067/2068	67	2073/2074	129	2079/2080	134
2068/2069	75	2074/2075	131	2080/2081	172
2069/2070	74	2075/2076	149	2081/2082	206
2070/2071	92	2076/2077	160	2082/2083	35

Examples (romanized, verbatim)

Quantity	Question	Answer
Total dividend %	aarthik barsha 2077/2078 ma esielabiesaela ko kul labhansha kati thiyo?	esielabiesaela ko aarthik barsha 2077/2078 ko kul labhansha 21% thiyo.
Total dividend %	esielabiesaela le aarthik barsha 2076/2077 ka lagi kul kati pratishat labhansha ghoshana gar‍yo?	esielabiesaela ko aarthik barsha 2076/2077 ko kul labhansha 8.85% thiyo.
Total dividend %	edibiela ko aarthik barsha 2081/2082 ko kul labhansha pratishat kati ho?	edibiela ko aarthik barsha 2081/2082 ko kul labhansha 13% thiyo.
Bonus-share dividend %	esielabiesaela ko aarthik barsha 2077/2078 ko bonas seyar labhansha pratishat kati ho?	esielabiesaela ko aarthik barsha 2077/2078 ko bonas seyar labhansha 21% thiyo.
Bonus-share dividend %	edibiela le aarthik barsha 2078/2079 ka lagi kati pratishat bonas seyar ghoshana gar‍yo?	edibiela ko aarthik barsha 2078/2079 ko bonas seyar labhansha 2% thiyo.
Bonus-share dividend %	aarthik barsha 2077/2078 ma edibiela ko bonas seyar labhansha kati thiyo?	edibiela ko aarthik barsha 2077/2078 ko bonas seyar labhansha 20% thiyo.

4.3 Institutional FAQ — free-form question answering (2,187 rows)

Natural FAQ pairs from 41 organisations, romanized. Questions are written the way customers ask them (esaemaesa sewa kasari sakriya garne? = “How to activate SMS service?”); answers are short explanatory paragraphs, procedures, limits, fees, eligibility rules or contact routes. See §6 for topics and §9 for question patterns.

---

5. Source coverage (all 43 sources)

Sorted by rows. Avg Q / Avg A = mean characters of the romanized text. Group is the topic group used in §6. Source labels are copied from the Devanagari release (not romanized).

#	Source	source_name	Rows	Share	Avg Q	Avg A	Group	Licence
1	ShareHub	sharehub_faq	2,448	36.9%	87	261	Market data (ShareHub)	Apache-2.0
2	NEPSE Dividend History	nepse_dividend_faq	2,000	30.1%	80	70	Dividends (NEPSE)	Apache-2.0
3	Nepal Rastra Bank	nepal_rastra_bank_faq	388	5.8%	79	345	Central bank & monetary policy	Apache-2.0
4	Lemonn	lemonn_all_faq	300	4.5%	37	338	Capital-market & investing education	unknown
5	Muktinath Bikas Bank	muktinath_bank_faq	106	1.6%	51	228	Commercial / development banks (retail banking)	Apache-2.0
6	Khalti	khalti_official_faq	105	1.6%	55	158	Digital payments, wallets & clearing	Apache-2.0
7	Nepal Bank Limited	nepal_bank_limited_faq	105	1.6%	64	213	Commercial / development banks (retail banking)	Apache-2.0
8	MOCO Nepal	moco_nepal_faq	96	1.4%	50	277	Digital payments, wallets & clearing	Apache-2.0
9	Fonepay	fonepay_faq	94	1.4%	53	239	Digital payments, wallets & clearing	Apache-2.0
10	Department of Financial Services, Government of India	dfsindia_faq	90	1.4%	61	241	Pension & provident funds	Apache-2.0
11	Nepal Clearing House Limited	nchl_faq	89	1.3%	58	305	Digital payments, wallets & clearing	Apache-2.0
12	Siddhartha Bank	siddhartha_bank_faq	81	1.2%	66	239	Commercial / development banks (retail banking)	Apache-2.0
13	Everest Bank	everest_bank_faq	69	1.0%	58	184	Commercial / development banks (retail banking)	Apache-2.0
14	Prime Commercial Bank	prime_commercial_bank_faq	59	0.9%	53	170	Commercial / development banks (retail banking)	Apache-2.0
15	National Institute of Securities Markets	nism_mutual_fund_faq	50	0.8%	47	192	Capital-market & investing education	Apache-2.0
16	Kumari Bank	kumari_bank_faq	49	0.7%	68	234	Commercial / development banks (retail banking)	Apache-2.0
17	Shine Resunga Development Bank	shine_resunga_faq	44	0.7%	58	223	Commercial / development banks (retail banking)	Apache-2.0
18	कर्मचारी सञ्चय कोष	employees_provident_fund_faq	43	0.6%	61	233	Pension & provident funds	Apache-2.0
19	Agricultural Development Bank Limited	adbl_kisan_faq/adbl_online_account_faq	34	0.5%	57	209	Commercial / development banks (retail banking)	Apache-2.0
20	नासा सेक्युरिटिज कम्पनी लिमिटेड	naasa_securities_faq	31	0.5%	58	251	Capital-market & investing education	Apache-2.0
21	Tata Moneyfy	tata_moneyfy_faq	30	0.5%	50	252	Capital-market & investing education	Apache-2.0
22	COPOMIS	copomis_faq	29	0.4%	91	258	Microfinance, cooperatives, lending & development	Apache-2.0
23	Machhapuchchhre Bank Limited	machhapuchchhre_bank_faq	28	0.4%	61	138	Commercial / development banks (retail banking)	Apache-2.0
24	नेपाल बीमा प्राधिकरण	nia_faq	21	0.3%	64	170	Insurance	Apache-2.0
25	NMB Bank	nmb_bank_faq	20	0.3%	58	220	Commercial / development banks (retail banking)	Apache-2.0
26	WealthSerene	wealthserene_faq	20	0.3%	55	619	Capital-market & investing education	Apache-2.0
27	World Bank	world_bank_faq	20	0.3%	62	452	Microfinance, cooperatives, lending & development	Apache-2.0
28	Foneloan	foneloan_faq	19	0.3%	55	348	Microfinance, cooperatives, lending & development	Apache-2.0
29	NLG Insurance Company Ltd.	nlg_insurance_faq	19	0.3%	45	288	Insurance	Apache-2.0
30	ईसेवा	esewa_faq	18	0.3%	53	249	Digital payments, wallets & clearing	Apache-2.0
31	Citizen Life Insurance Company Limited	citizenlife_faq	16	0.2%	56	308	Insurance	Apache-2.0
32	Farmers Development Microfinance Pvt. Ltd.	fdmpl_faq	15	0.2%	40	157	Microfinance, cooperatives, lending & development	Apache-2.0
33	Nepal Stock Exchange Limited	nepse_faq	14	0.2%	83	225	Capital-market & investing education	Apache-2.0
34	eSewa	esewa_taxpayer_incentive_prize_program_faq	12	0.2%	44	259	Digital payments, wallets & clearing	Apache-2.0
35	Global IME Bank	gbime_faq	11	0.2%	42	338	Commercial / development banks (retail banking)	Apache-2.0
36	Nabil Invest	nabil_invest_mutual_fund_faq	11	0.2%	62	375	Capital-market & investing education	Apache-2.0
37	Reliance Finance Limited	reliancefinance_faq	11	0.2%	77	312	Microfinance, cooperatives, lending & development	Apache-2.0
38	Life Insurance Corporation (Nepal) Ltd.	licnepal_faq	10	0.2%	41	267	Insurance	Apache-2.0
39	Nepal Investment Mega Bank	nimb_smart_faq	9	0.1%	63	247	Commercial / development banks (retail banking)	Apache-2.0
40	Sanima GIC Insurance Ltd.	sanima_gic_faq	7	0.1%	31	380	Insurance	Apache-2.0
41	Deposit & Credit Guarantee Fund	dcgf_faq	6	0.1%	40	193	Microfinance, cooperatives, lending & development	Apache-2.0
42	Invest Nepal	investnepal_faq	5	0.1%	53	659	Capital-market & investing education	Apache-2.0
43	CDS and Clearing Limited	cdsc_faq	3	0.0%	21	270	Capital-market & investing education	Apache-2.0

Some labels are transliterations in the upstream data (e.g. “फोनपे” in generation_category for the Fonepay rows) or follow upstream attribution (the National Pension System FAQ is tagged “Department of Financial Services, Government of India”; its content was localised into Nepali before romanization). Rows are kept exactly as provided.

---

6. Topic and domain diversity

6.1 Topic groups of the Institutional FAQ part

Topic group	Rows	Share of FAQ part	# sources	Sources
Central bank & monetary policy	388	17.7%	1	Nepal Rastra Bank
Commercial / development banks (retail banking)	615	28.1%	12	Agricultural Development Bank Limited, Everest Bank, Global IME Bank, Kumari Bank, Machhapuchchhre Bank Limited, Muktinath Bikas Bank …
Digital payments, wallets & clearing	414	18.9%	6	Fonepay, Khalti, MOCO Nepal, Nepal Clearing House Limited, eSewa, ईसेवा
Capital-market & investing education	464	21.2%	9	CDS and Clearing Limited, Invest Nepal, Lemonn, Nabil Invest, National Institute of Securities Markets, Nepal Stock Exchange Limited …
Pension & provident funds	133	6.1%	2	Department of Financial Services, Government of India, कर्मचारी सञ्चय कोष
Insurance	73	3.3%	5	Citizen Life Insurance Company Limited, Life Insurance Corporation (Nepal) Ltd., NLG Insurance Company Ltd., Sanima GIC Insurance Ltd., नेपाल बीमा प्राधिकरण
Microfinance, cooperatives, lending & development	100	4.6%	6	COPOMIS, Deposit & Credit Guarantee Fund, Farmers Development Microfinance Pvt. Ltd., Foneloan, Reliance Finance Limited, World Bank

6.2 generation_domain (all rows)

generation_domain	English	Rows	Share
वित्त	Finance	3,957	59.6%
वित्तीय सेवा	Financial services	2,527	38.1%
बैंकिङ	Banking	70	1.1%
सहकारी	Cooperative	29	0.4%
डिजिटल भुक्तानी	Digital payments	18	0.3%
बीमा	Insurance	16	0.2%
लघुवित्त	Microfinance	15	0.2%
बैंकिङ तथा पूँजी बजार	Banking & capital market	3	0.0%

6.3 generation_category (all 56 categories)

Metadata labels stay in Devanagari in this release.

#	generation_category	Rows	Share
1	शेयर बजार सम्बन्धी प्रायः सोधिने प्रश्न	2,448	36.9%
2	नेप्से लाभांश इतिहास	2,000	30.1%
3	नेपाल राष्ट्र बैंक सम्बन्धी प्रायः सोधिने प्रश्न	388	5.8%
4	सेयर बजारका आधारभूत विषय	300	4.5%
5	बैंकिङ सम्बन्धी प्रायः सोधिने प्रश्न	134	2.0%
6	खाल्टी सम्बन्धी प्रायः सोधिने प्रश्न	105	1.6%
7	नेपाल बैंक लिमिटेड सम्बन्धी प्रायः सोधिने प्रश्न	105	1.6%
8	भुक्तानी सेवा सम्बन्धी प्रायः सोधिने प्रश्न	96	1.4%
9	फोनपे सम्बन्धी प्रायः सोधिने प्रश्न	94	1.4%
10	एनसीएचएल सम्बन्धी प्रायः सोधिने प्रश्न	89	1.3%
11	सिद्धार्थ बैंक सम्बन्धी प्रायः सोधिने प्रश्न	81	1.2%
12	प्राइम कमर्सियल बैंक सम्बन्धी प्रायः सोधिने प्रश्न	59	0.9%
13	म्युचुअल फन्ड सम्बन्धी प्रायः सोधिने प्रश्न	50	0.8%
14	राष्ट्रिय पेन्सन प्रणाली सम्बन्धी प्रायः सोधिने प्रश्न	47	0.7%
15	शाइन रेसुङ्गा डेभलपमेन्ट बैंक सम्बन्धी प्रायः सोधिने प्रश्न	44	0.7%
16	अटल पेन्सन योजना सम्बन्धी प्रायः सोधिने प्रश्न	43	0.6%
17	खाता तथा सामान्य बैंकिङ सेवा सम्बन्धी प्रायः सोधिने प्रश्न	35	0.5%
18	धितोपत्र दलाली सम्बन्धी प्रायः सोधिने प्रश्न	31	0.5%
19	टाटा मनिफाई सम्बन्धी प्रायः सोधिने प्रश्न	30	0.5%
20	कोपोमिस सम्बन्धी प्रायः सोधिने प्रश्न	29	0.4%
21	जीवन बीमा सम्बन्धी प्रायः सोधिने प्रश्न	26	0.4%
22	खाता व्यवस्थापन तथा इ-सेवा सम्बन्धी प्रायः सोधिने प्रश्न	21	0.3%
23	डेबिट कार्ड तथा सामान्य सेवा सम्बन्धी प्रायः सोधिने प्रश्न	21	0.3%
24	नेपाल बीमा प्राधिकरण सम्बन्धी प्रायः सोधिने प्रश्न	21	0.3%
25	एनएमबी बैंक सम्बन्धी प्रायः सोधिने प्रश्न	20	0.3%
26	वेल्थसेरीन व्यक्तिगत वित्त सम्बन्धी प्रायः सोधिने प्रश्न	20	0.3%
27	विश्व बैंक सम्बन्धी प्रायः सोधिने प्रश्न	20	0.3%
28	अनलाइन खाता सम्बन्धी प्रायः सोधिने प्रश्न	19	0.3%
29	फोनलोन सम्बन्धी प्रायः सोधिने प्रश्न	19	0.3%
30	एनएलजी इन्स्योरेन्स सम्बन्धी प्रायः सोधिने प्रश्न	19	0.3%
31	ईसेवा सम्बन्धी प्रायः सोधिने प्रश्न	18	0.3%
32	किसान एप सम्बन्धी प्रायः सोधिने प्रश्न	15	0.2%
33	फार्मर डेभलपमेन्ट माइक्रोफाइनान्स सम्बन्धी प्रायः सोधिने प्रश्न	15	0.2%
34	मोबाइल बैंकिङ सम्बन्धी प्रायः सोधिने प्रश्न	14	0.2%
35	नेप्से सम्बन्धी प्रायः सोधिने प्रश्न	14	0.2%
36	सञ्चयकर्ता स्वास्थोपचार योजना सम्बन्धी प्रायः सोधिने प्रश्न	13	0.2%
37	करदाता प्रोत्साहन उपहार कार्यक्रम सम्बन्धी प्रायः सोधिने प्रश्न	12	0.2%
38	नाबिल म्युचुअल फण्ड सम्बन्धी प्रायः सोधिने प्रश्न	11	0.2%
39	रिलायन्स फाइनान्स सम्बन्धी प्रायः सोधिने प्रश्न	11	0.2%
40	इन्टरनेट बैंकिङ सम्बन्धी प्रायः सोधिने प्रश्न	10	0.2%
41	रेमिटेन्स सम्बन्धी प्रायः सोधिने प्रश्न	10	0.2%
42	कार्ड व्यवस्थापन तथा शुल्क सम्बन्धी प्रायः सोधिने प्रश्न	9	0.1%
43	नेपाल इन्भेस्टमेन्ट मेगा बैंक स्मार्ट सम्बन्धी प्रायः सोधिने प्रश्न	9	0.1%
44	डिम्याट शेयर तथा धितोपत्र स्थानान्तरण सम्बन्धी प्रायः सोधिने प्रश्न	8	0.1%
45	सामान्य प्रायः सोधिने प्रश्न	8	0.1%
46	सनिमा जीआईसी इन्स्योरेन्स सम्बन्धी प्रायः सोधिने प्रश्न	7	0.1%
47	निक्षेप तथा कर्जा सुरक्षण सम्बन्धी प्रायः सोधिने प्रश्न	6	0.1%
48	स्विफ्ट तथा अन्तर्राष्ट्रिय स्थानान्तरण सम्बन्धी प्रायः सोधिने प्रश्न	6	0.1%
49	कर्मचारी सञ्चय कोष सेवा सम्बन्धी प्रायः सोधिने प्रश्न	5	0.1%
50	लगानी तथा प्रत्यक्ष विदेशी लगानी सम्बन्धी प्रायः सोधिने प्रश्न	5	0.1%
51	डिपोजिटरी सम्बन्धी प्रायः सोधिने प्रश्न	3	0.0%
52	सापटी तथा कर्जा सम्बन्धी प्रायः सोधिने प्रश्न	3	0.0%
53	मिस्ड कल अलर्ट सम्बन्धी प्रायः सोधिने प्रश्न	3	0.0%
54	एकसाथ पाँच सम्बन्धी प्रायः सोधिने प्रश्न	3	0.0%
55	सम्पर्क तथा प्रतिनिधि कार्यालय सम्बन्धी प्रायः सोधिने प्रश्न	2	0.0%
56	प्रसूति स्याहार तथा परिवार सम्बन्धी प्रायः सोधिने प्रश्न	1	0.0%

6.4 generation_sub_domain (Institutional FAQ part only, top 40)

The FAQ part alone has 106 distinct sub-domains; the 40 largest:

#	generation_sub_domain	Rows
1	केन्द्रीय बैंक र मौद्रिक नीति	388
2	बैंकिङ सेवा	309
3	म्युचुअल फन्ड, आईपीओ तथा व्यक्तिगत लगानी	300
4	डिजिटल भुक्तानी	211
5	भुक्तानी प्रणाली	89
6	म्युचुअल फन्ड लगानी	50
7	राष्ट्रिय पेन्सन प्रणाली	47
8	अटल पेन्सन योजना	43
9	खाता तथा सामान्य सेवा	35
10	नासा सेक्युरिटिज	31
11	ऋण र म्युचुअल फन्ड	30
12	सहकारी व्यवस्थापन सूचना प्रणाली	29
13	माछापुच्छ्रे बैंक सेवा	28
14	निर्जीवन बीमा	26
15	मोबाइल बैंकिङ	23
16	खाता व्यवस्थापन तथा इ-सेवा	21
17	डेबिट कार्ड तथा सामान्य सेवा	21
18	महिला उद्यम विकास कर्जा	21
19	बीमा नियमन	21
20	बजेटिङ र आपतकालीन कोष	20
21	एडीबीएल अनलाइन खाता	19
22	डिजिटल ऋण सेवा	19
23	डिजिटल वालेट	18
24	सिटीजन लाइफ इन्स्योरेन्स	16
25	एडीबीएल किसान एप	15
26	किसान लघुवित्त	15
27	शेयर बजार र ट्रेड व्यवस्थापन प्रणाली	14
28	सञ्चयकर्ता स्वास्थोपचार योजना	13
29	ग्लोबल आइएमई बैंक	11
30	म्युचुअल फण्ड	11
31	वित्तीय सेवा	11
32	इन्टरनेट बैंकिङ	10
33	रेमिटेन्स	10
34	जीवन बीमा	10
35	ऋण तथ्याङ्क र पारदर्शिता	10
36	ऋण तथ्याङ्क र धोखाधडी जागरूकता	10
37	कार्ड व्यवस्थापन तथा शुल्क	9
38	डिम्याट शेयर तथा धितोपत्र	8
39	भुक्तानी रकम फिर्ता	7
40	निक्षेप तथा कर्जा सुरक्षण कोष	6

6.5 Topics covered, in plain English

Theme	Typical questions answered
Central-bank & regulation	Who regulates microfinance? How are currency notes exchanged? Commemorative coins, clearing sessions, monetary-policy terms, supervision procedures
Retail banking	Account opening, passwords/PIN reset, debit & credit cards, mobile/internet banking, bank guarantees, loan limits, interest-tax, branch services
Digital payments	QR payments (NepalPay/Fonepay), wallet KYC, refunds, transaction limits, merchant onboarding, clearing systems (IPS / connectIPS), remittance
Stock market	DEMAT accounts, IPO investing, SIP and mutual funds, brokerage, share transfer, lock-in/unlock, options and market vocabulary
Pension & provident fund	NPS account access, contribution rules, exit at 60, Atal Pension, EPF loans, medical-treatment reimbursement, claim limits
Insurance	Life/non-life policies, premiums, claims, multiple policies, regulator (Nepal Insurance Authority)
Personal finance & lending	Emergency fund, sinking fund, budgeting, digital loans (collateral-free), women-entrepreneur loans, home/auto/share loans, credit information & fraud awareness

---

7. Behavior diversity

Every record carries a behavior label and a behavior_definition (both in Devanagari, copied from the Devanagari release). This release has only two behaviors; the label is tightly coupled to the sub-corpus.

behavior	Meaning	Rows	Share	behavior_definition (verbatim)	Main sources
जानकारीमूलक उत्तर दिने	Gives an informational answer	3,696	55.7%	मूल प्रश्नको अर्थ र मूल उत्तरको तथ्य कायम राख्दै नेपालीमा रूपान्तरण गरिएको उत्तर दिने।	sharehub_faq (2448), nepal_rastra_bank_faq (388), khalti_official_faq (105) …
तथ्यात्मक लाभांश जानकारी दिने	Gives factual dividend information	2,000	30.1%	निर्दिष्ट कम्पनी र आर्थिक वर्षको लाभांश तथ्यांक प्रयोग गरी प्रश्नको प्रत्यक्ष उत्तर दिने।	nepse_dividend_faq (2000)
जानकारीमूलक उत्तर दिने	Gives an informational answer	639	9.6%	मूल प्रश्नको अर्थ र मूल उत्तरको तथ्य कायम राख्दै शुद्ध नेपालीमा रूपान्तरण गरिएको उत्तर दिने।	muktinath_bank_faq (106), moco_nepal_faq (96), dfsindia_faq (90) …
जानकारीमूलक उत्तर दिने	Gives an informational answer	300	4.5%	मूल प्रश्नको अर्थ र मूल उत्तरका तथ्य कायम राख्दै शुद्ध नेपालीमा रूपान्तरण गरिएको उत्तर दिने।	lemonn_all_faq (300)

English meaning of the behavior definitions

Definition (summary)	Meaning
“मूल प्रश्नको अर्थ र मूल उत्तरको तथ्य कायम राख्दै नेपालीमा रूपान्तरण गरिएको उत्तर दिने।”	Answer is a Nepali rendering that preserves the meaning of the original question and the facts of the original answer
“… शुद्ध नेपालीमा रूपान्तरण …” / “… उत्तरका तथ्य कायम राख्दै शुद्ध नेपालीमा …”	Same, with emphasis on pure Nepali — used for 939 rows (e.g. all Lemonn rows, Muktinath, MOCO, DFS-India, Everest)
“निर्दिष्ट कम्पनी र आर्थिक वर्षको लाभांश तथ्यांक प्रयोग गरी प्रश्नको प्रत्यक्ष उत्तर दिने।”	Answer the question directly using the dividend figures for the named company and fiscal year

The definitions describe the Nepali (Devanagari) generation step. The romanization was applied afterwards, so the label says nothing about romanization.

How diverse is the behavior actually? Honestly, limited. The diversity in this dataset comes mainly from topics, sources, question intents and answer formats (§6, §9, §10), not from different assistant behaviors.

Behavior aspect	Present?	Notes
Direct factual answer	Yes (dominant)	Answers state the fact or instruction directly rather than hedging
Explanatory / informational answer	Yes (FAQ part and ShareHub)	1–5 sentence explanations
Step-by-step procedure	Yes (FAQ part)	Imperative -nuhos steps; see §10
Numeric/structured data answer	Yes (ShareHub, dividends)	Prices, ratios, percentages, dates
Yes/No answer with justification	Yes (FAQ part)	ho, … / hoina. … openings
Refusal / out-of-scope handling	No	—
Clarifying questions back to the user	No	—
Multi-turn dialogue	No	Always exactly 1 human + 1 gpt turn
Chain-of-thought / reasoning	No	Answers are concise final answers
System prompts	No	Add your own (see §3.8)
Mixed-script / code-switching behavior	No	Text is consistently Latin-script; no Devanagari ↔ Latin switching within a row

---

8. Question-type diversity (metadata labels)

The question_type metadata has three values. The labels are coarse and unevenly applied: 95% of rows carry the same label.

question_type	English	Rows	Share	Top sources
तथ्यमा आधारित प्रश्न	Fact-based question	6,335	95.5%	sharehub_faq (2448), nepse_dividend_faq (2000), nepal_rastra_bank_faq (388)
जानकारीमूलक प्रश्न	Informational (explanatory) question	223	3.4%	lemonn_all_faq (223)
प्रक्रियामूलक प्रश्न	Procedural (“how-to”) question	77	1.2%	lemonn_all_faq (77)

जानकारीमूलक प्रश्न and प्रक्रियामूलक प्रश्न appear only in the Lemonn source (223 + 77 = 300 rows). Use the text-based patterns in §9 for finer-grained analysis.

Label	Example question (romanized)	Example answer (romanized)
जानकारीमूलक प्रश्न	ke myuchuala phandama kar lagchha?	ho, bharatma myuchuala phandabata prapta pratiphalma kar lagchha. karko wyawastha myuchuala phandako prakar, jastai ikwiti, rin wa haibrid phanda, ra lagani kati samaya rakhieko thiyo bhanne awadhima nirbhar hunchha. witta aina 2024 le aarthik barsha 2024–25 …
प्रक्रियामूलक प्रश्न	aaipiole kasari kam garcha?	aaipio arthat prarambhik sarwajanik nishkasana niji kampanilai sarwajanik laganikartaka lagi seyar upalabdha garaune ra tyasapachhi enaesai wa biesaijasta manyataprapta stak eksachenjama suchikrit garne niyamakiya prakriya ho. yas prakriyama sebidwara …
तथ्यमा आधारित प्रश्न	nepal rashtra bainkako duradrishti ke raheko chha?	nepal rashtra bainkako duradrishti “aadhunik, gatishil, wishwasaniya ra prabhawakari kendriya bainka” bhanne raheko chha.

---

9. Question-pattern diversity

Because the romanized file is row-aligned with the Devanagari file, patterns are detected on the Devanagari question text (where Nepali spelling is stable and keyword rules are reliable) and reported with the romanized questions. A question can match several patterns, so shares do not sum to 100%.

9.1 Templated questions (ShareHub + dividends = 4,448 rows, 67.0%)

These have a fixed surface form with only the entity (security / company / year) changing — 8 templates for ShareHub and 3 syntactic forms × 2 quantities for dividends (§4.1, §4.2). Only 4,446 unique questions come from 4,448 templated rows because the entities vary, not the phrasing.

9.2 Free-form FAQ questions (2,187 rows, 33.0%) — intent patterns

Pattern	English gloss	Romanized cue	Rows	Share of FAQ part	Example question (romanized)
Definition — “X के हो?”	What is X?	ke ho / ke hun	556	25.4%	aaiesaaaiena ke ho?
Procedure — “कसरी …?”	How do I / how to …?	kasari	473	21.6%	yogadana kasari lagani hunchha?
Ability / eligibility — “… सक्छ / सकिन्छ?”	Can I / can one …? Who can …?	sakchha, sakinchha	373	17.1%	ke hami bhansar pragyapanapatrama samawesh rakamabhanda badhi titi marphat rakam sthanantaran garna sakchhau?
Closed-form question — ends “छ/हुन्छ/हो/छन्?” with no other question word	Is it…? Does it…? (also “what are the …?” forms)	… chha?, … huncha?, … ho?	461	21.1%	sahakari sansthale kopomisma sadasyako wiwaran prawishta garne kramma sadasyasambandhi wiwaran apug bhaema ke …
Quantity / limit / fee — “कति …?”	How much / how many / how long?	kati	202	9.2%	kati yogadana garnuparchha?
Reason / purpose — “किन …?”	Why …?	kina	58	2.7%	naya niyukti bhaepachhi sanchaya koshma dui thari rakam pathaunuparne kina?
Selection — “कुन / कुनै …?”	Which …? (also “any …?”)	kun, kunai	148	6.8%	ke rojagaradatale karmachariko kosh wyawasthapanama kunai bhumika khelchha?
Place — “कहाँ …?”	Where …?	kaha…	55	2.5%	parikshan ra laibha kij kahabata prapta garne?
Time — “कहिले …?”	When …?	kahile	27	1.2%	byaj kahile khatama jamma garinchha?
Consequence — “… भएमा के हुन्छ?”	What happens if …?	ke hunchha	44	2.0%	atal pensana yojanama sthanantaranapachhi rashtriya pensana pranali laita / swawalambana yojanaantargat jamma …
Conditional — “यदि … भने …”	If … then …	yadi	24	1.1%	yadi karobar suru garepachhi pratikriya esaemaesa prapta bhaena bhane ke hunchha?
Agent — “कसले … ?”	Who …?	kasle, ko ho/hun	36	1.6%	dhito rakhieka dhitopatrama karporet labh kasle paunchha?
Comparison / difference — “फरक / बीच / तुलना”	How is X different from Y?	pharak, bicha, tulana	37	1.7%	khalti pementa getawe (kepiji) nepalka anya pementa getawebhanda kasari pharak chha?
First-person / personal — “म / मेरो / मैले / मलाई”	I / my … (customer-voice)	ma, mero, maile, malai	325	14.9%	maile aaphno pasawarda birse bhane ke garne?
Multi-question — more than one “?”	Two questions in one turn	two ?	52	2.4%	mero masik yogadana kasari lagani hunchha? ke swaichchhik yogadana garna sakinchha?

2,092 of 2,187 FAQ questions (95.7%) match at least one pattern; the rest are mostly noun-phrase / statement-style questions.

Cross-check: can the patterns be detected directly from the romanized text? Yes, for most of them — a simple regex on the romanized question reproduces the Devanagari-derived label almost exactly:

Romanized cue	Rows (Devanagari rule)	Rows (romanized regex)	Rows matched by both
kasari	473	474	473
kati	202	199	199
kina	58	58	58
kun*	148	148	148
kahile	27	24	24
ke hunchha	44	44	44
yadi	24	24	24
kasle	26	26	26
sakchha/sakinchha	373	373	373

9.3 Question wording features (all rows)

Feature	Rows	Share
Question contains digits (all ASCII in this release)	2,609	39.3%
Question contains a parenthesised ticker, e.g. (aclbsl)	2,505	37.8%
Free-form questions containing digits	28	1.3%
Ends with ?	6,627	99.9%
Space before ? (… garne ?) — typical of source typing style	44	0.7%
Question contains an upper-case letter	28	0.4%

9.4 Example questions per pattern (verbatim romanized, with Devanagari original of the question)

Definition — “X के हो?” — What is X?

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
kisana kredit karda ke ho?	किसान क्रेडिट कार्ड के हो?	kisana kredit karda bainkadwara kisanalai krishima aadharit utpadana tatha sewaharuko kharid/bikriko kramma ilektronik bhuktanika lagi ra bankma rakhieko aaphno sambandhit khatama nagad jhikna/jamma …	Agricultural Development Bank Limited
wittiya yojana bhaneko ke ho?	वित्तीय योजना भनेको के हो?	wittiya yojana bhaneko halko aarthik awastha mulyankana gari chhoto tatha dirghakalina lakshya pura garna samagra rananiti banaune prakriya ho. yasma bajet, bachat, lagani, kar, bima, awakash yojana …	Lemonn
ekamushta punarakarjaka lagi aawedana dinka lagi aawashyak kagajat ke ke hun?	एकमुष्ट पुनरकर्जाका लागि आवेदन दिनका लागि आवश्यक कागजात के के हुन्?	ekamushta punarakarjako lagi aawedana aahriwana garda punarakarja sambandhi karyawidhi, 2077 ma tokie bamojimko dhanchama aawedana tatha grahak riniko wiwaran matra aawashyak hunchha. anya kagajat …	Nepal Rastra Bank

Procedure — “कसरी …?” — How do I / how to …?

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
sahakari sansthale aaphno sansthako koshasambandhi wiwaran kasari rakhnupardachha?	सहकारी संस्थाले आफ्नो संस्थाको कोषसम्बन्धी विवरण कसरी राख्नुपर्दछ?	wibhagle prakashana gareko sahakari aina, 2074 ra niyamawali, 2075 ma ullekh bhaeka koshharu matra prawishta garnuparne hunchha. pranalima sahakari aina, 2074 ra niyamawali, 2075 ma ullekh bhaeka …	COPOMIS
masik aamdani kasari lagani garne?	मासिक आम्दानी कसरी लगानी गर्ने?	talab wa aamdani khatama jamma hunebittikai pahile nai taya garieko rakam laganima chhutyaunu ramro hunchha. yaslai pahile aaphulai bhuktani garne tarika bhanna sakinchha, jasle laganilai swachalit …	Lemonn
chhawiharu kasari kyapchar garinchhan?	छविहरू कसरी क्याप्चर गरिन्छन्?	chekka widyutiya chhawi ra emaaaisiaara deta prastut garne bainkale chek skyanar prayog garera kyapchar gardachha. prastut garne bainkale skyana garnuaghi sabai chekko pachhadi (dayantir) prastut …	Nepal Clearing House Limited

Ability / eligibility — “… सक्छ / सकिन्छ?” — Can I / can one …? Who can …?

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
naya wyawasayale pani aawedana din sakchha wa puranale matra?	नयाँ व्यवसायले पनि आवेदन दिन सक्छ वा पुरानाले मात्र?	samanyataya sanchalanama raheka purana wyawasayalai prathamikata diinchha, tara bharapardo ra sambhawya wyawasayik yojana pesh gare naya wyawasaya ra startaapa pani aawedana din sakchhan.	Muktinath Bikas Bank
nepalma raheka wideshi nagarikle wideshi mudrama khata kholna sakchhan/sakdainan?	नेपालमा रहेका विदेशी नागरिकले विदेशी मुद्रामा खाता खोल्न सक्छन्/सक्दैनन्?	nepalma raheka wideshi nagarikle nepalsthit bainka tatha wittiya sansthama pariwartya wideshi mudrama khata kholna sakchhan. yasto khata bhisa awadhisamma matra sanchalana garna sakinchha.	Nepal Rastra Bank
eutai rinile ekabhanda badhi bainka tatha wittiya sansthabata punarakarja prapta garna sakchhan/sakdainana?	एउटै ऋणीले एकभन्दा बढी बैंक तथा वित्तीय संस्थाबाट पुनरकर्जा प्राप्त गर्न सक्छन/सक्दैनन?	sakdainana. punarakarja sambandhi karyawidhi, 2077 ma kunai pani rinile eka patakma ekabhanda badhi ijajatapatraprapta sansthamarphat punarakarja suwidha upayog garna paune chhainan bhanne spashta …	Nepal Rastra Bank

Closed-form question — ends “छ/हुन्छ/हो/छन्?” with no other question word — Is it…? Does it…? (also “what are the …?” forms)

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
jiwana bimale ke kabhar garcha?	जीवन बीमाले के कभर गर्छ?	mulatah, jiwana bimale bimit wyaktiko mrityupachhi bancheka aashritharulai mrityu labh (aarthik suraksha) pradana garchha. polisile aamdaniko noksani, antyeshti kharcha ra polisiholdarko mrityuko …	Citizen Life Insurance Company Limited
kebiela le pradana garne byajadarharu ke–kasta chhan?	केबीएल ले प्रदान गर्ने ब्याजदरहरू के–कस्ता छन्?	kebieladwara pradana garine adyawadhik byajadarharu sambandhi janakarika lagi kripaya tala diieko linka anusaran garnuhos:	Kumari Bank
chalu punji karjako sima sampurna karjako lagi lagu hunchha?	चालू पुँजी कर्जाको सीमा सम्पूर्ण कर्जाको लागि लागु हुन्छ?	chalupunji karjako sima chalu punji bahek anya karjaka lagi lagu hundaina. wyawasayaka lagi aawashyak parne anya karja widyamana niti niyam anusar bainka tatha wittiya sansthale sohi prayojanaka lagi …	Nepal Rastra Bank

Quantity / limit / fee — “कति …?” — How much / how many / how long?

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
khata kholnako lagi nyunatam maujdat kati chahinchha?	खाता खोल्नको लागि न्यूनतम मौज्दात कति चाहिन्छ?	khata shunya maujdatsanga kholna sakinchha, tara karobarka lagi tapaile nyunatam maujdat kayam rakhnuparchha. nyunatam maujdat tapaile analaina kholna chhaneko khatako prakarma nirbhar gardachha. …	Agricultural Development Bank Limited
asaphal karobar phirta huna kati samaya lagchha?	असफल कारोबार फिर्ता हुन कति समय लाग्छ?	samanyataya paisa turuntai phirta garinchha; yadyapi, kehi paristhitiharuma, 7 dinasammako lagna saknechha.	MOCO Nepal
trabhal/treking, hotal, airlines, kargo tatha anya (manichenjar tatha wipreshan kampanibahek) sansthale wideshi mudra karobar garne …	ट्राभल/ट्रेकिङ, होटल, एयरलाइन्स, कार्गो तथा अन्य (मनिचेञ्जर तथा विप्रेषण कम्पनीबाहेक) संस्थाले विदेशी मुद्रा कारोबार …	trabhal/treking, hotal, airlines, kargo tatha anya (manichenjar tatha wipreshan kampanibahek) sansthale wideshi mudra karobar garne ijajatapatra prapta garna tatha nawikaran garna pratyek karobar …	Nepal Rastra Bank

Reason / purpose — “किन …?” — Why …?

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
sahakari sansthale aaphno masik wiwaranharu prawishta garisakepachhi pani dyasabordama kina dekhaundaina?	सहकारी संस्थाले आफ्नो मासिक विवरणहरू प्रविष्ट गरिसकेपछि पनि ड्यासबोर्डमा किन देखाउँदैन?	ukta karya kopomis sambandhi saphtaweyaraantargatko karya ho ra ukta samasyako samadhanaka lagi saphtaweyarko karya kramabaddha rupma bhairaheko chha ra so karya bhaisakepachhi janakari webasaita …	COPOMIS
namankana kina aawashyak chha?	नामाङ्कन किन आवश्यक छ?	bimit wyaktiko asamayik mrityu bhaema namankanale namankit wyaktika pakshama dabi chhito samadhana garna sahaj banaunchha. prastav pharamma namankit wyaktiko naam spashta ullekh garnuparchha. kunai …	Life Insurance Corporation (Nepal) Ltd.
siddhartha phonape kredit karda (esaephasisi) byanar dekhchhu tara aawedana dinda karda asaina hundaina. kina?	सिद्धार्थ फोनपे क्रेडिट कार्ड (एसएफसीसी) ब्यानर देख्छु तर आवेदन दिँदा कार्ड असाइन हुँदैन। किन?	esaephasisi aawedana dinda karda sakriya garnuko satta antimma ma ruchi rakhchhu batana dekhiyo bhane tapai hal tatkal aawedana din ayogya hunuhunchha. ma ruchi rakhchhu klik garera bainkalai aaphno …	Siddhartha Bank

Selection — “कुन / कुनै …?” — Which …? (also “any …?”)

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
sahakari sansthale aaphno sansthako rakam prawishta garda kuna ankama rakhnupardachha?	सहकारी संस्थाले आफ्नो संस्थाको रकम प्रविष्ट गर्दा कुन अंकमा राख्नुपर्दछ?	kopomis saphtaweyarma sahakari sansthale aaphno sansthako kunai pani rakam prawishta garda pura ankama rakhnupardachha. jastai eka lakh punji prawishta garda eka lakh nai prawishta garna aawashyak …	COPOMIS
nikshep wa jhikaiko sankhya sambandhi kunai sima chha ki chhaina?	निक्षेप वा झिकाइको संख्या सम्बन्धी कुनै सीमा छ कि छैन?	grahak harule nikshep tatha rakam jhikai kunai pani sankhya patak bina kunai simabina garna sakdachhan. tathapi, debit kardamarphat garine jhikaiko sankhya dainik adhikatam 10 patakasamma matra simit …	Kumari Bank
bainking wibhagma widyutiya chek samashodhana pranali (isisi)/antarabainka bhuktani pranali (aaipiesa) karobarko samaya kuna ho?	बैंकिङ विभागमा विद्युतीय चेक समाशोधन प्रणाली (ईसीसी)/अन्तरबैंक भुक्तानी प्रणाली (आईपीएस) कारोबारको समय कुन हो?	isisi/aaipiesa ko karobar samaya mahina din isisi aaipiesa magh 16 dekhi kattik 15 aaitabar – bihibarasamma bihana 10:00 dekhi aparahna 3:00 bajesamma bihana 10:00 bajedekhi aparahna 5:00 bajesamma …	Nepal Rastra Bank

Place — “कहाँ …?” — Where …?

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
aawedana pharam kaha upalabdha hunchha?	आवेदन फाराम कहाँ उपलब्ध हुन्छ ?	yas yojanako aawedana pharam thamelsthit koshko niwrittabharan wyawasthapana tatha samajik suraksha wibhag, koshka sabai shakha karyalayaharu ra koshko webasaitabata lina sakinchha.	कर्मचारी सञ्चय कोष
kanektaaaipiesa mobile epa kahabata daunalod garne?	कनेक्टआईपीएस मोबाइल एप कहाँबाट डाउनलोड गर्ने?	kanektaaaipiesa mobile epa ple stor wa epa storbata daunalod garna sakinchha.	Nepal Clearing House Limited
nepal bima pradhikaranko karyalaya kaha raheko chha?	नेपाल बीमा प्राधिकरणको कार्यालय कहाँ रहेको छ?	nepal bima pradhikaranko kendriya karyalaya lalitapur mahanagarapalika–10 sthit kupandol, kandewatama raheko chha.	नेपाल बीमा प्राधिकरण

Time — “कहिले …?” — When …?

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
nepalma nikshep tatha karja surakshan kosh kahile sthapana bhaeko ho?	नेपालमा निक्षेप तथा कर्जा सुरक्षण कोष कहिले स्थापना भएको हो?	nepalma nikshep tatha karja surakshan kosh wi.san. 2031 asoj 4 ma sthapana bhaeko ho.	Deposit & Credit Guarantee Fund
aaipioko phirta rakam kahile aaunchha?	आईपीओको फिर्ता रकम कहिले आउँछ?	bharatko eesabie pranalima aaipio aawedana garda rakam prayah khatabata turunta katindaina, baru chhutyaera rokka rakhinchha. seyar naparema bandaphantapachhi samanyataya ekadekhi dui …	Lemonn
wyapari khatama rakam kahile sthanantaran hunchha?	व्यापारी खातामा रकम कहिले स्थानान्तरण हुन्छ?	dainik karobar rakam ru. 50 bhanda mathi bhae pranalile karobar dinko rati 11 bajepachhi swatah wyapariko khatama rakam sthanantaran garchha. ru. 50 bhanda kam bhae wyaparile phonape epabata myanuala …	Shine Resunga Development Bank

Consequence — “… भएमा के हुन्छ?” — What happens if …?

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
khata kholepachhi nagarik narahe ke hunchha?	खाता खोलेपछि नागरिक नरहे के हुन्छ?	khata banda gari yogadanako wastawik byaj phirta garinchha. sarakari saha-yogadana phirta hundaina.	Department of Financial Services, Government of India
yadi maile rakam galat wyaktilai pathae bhane ke hunchha?	यदि मैले रकम गलत व्यक्तिलाई पठाएँ भने के हुन्छ?	praptakartako mobile number taipa garda atyanta sawadhani apanauna hami sujhav dinchhaun. yadi asawadhaniwash tapaile anjanama galat wyaktilai rakam pathaunu bhaema, kripaya turuntai +977-01-5970499 …	Kumari Bank
suraksha prashna birsiema ke hunchha?	सुरक्षा प्रश्न बिर्सिएमा के हुन्छ?	khata riset ra pasawarda riset garna najikko shakhama januhos.	Nepal Bank Limited

Conditional — “यदि … भने …” — If … then …

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
yadi ma mero lagaina pasawarda/karobar pina birse bhane ma ke garnuparchha?	यदि म मेरो लगइन पासवर्ड/कारोबार पिन बिर्सें भने म के गर्नुपर्छ?	yadi tapaile aaphno lagaina pasawarda ra karobar pina birsanubhayo bhane kripaya sewa punah suru garna najikaiko shakhama januhos.	Agricultural Development Bank Limited
yadi maile mero mobile number pariwartana gare bhane ke hunchha?	यदि मैले मेरो मोबाइल नम्बर परिवर्तन गरेँ भने के हुन्छ?	yadi tapaile aaphno mobile number pariwartana garnubhaeko chha bhane, kripaya najikko kumari bainka limitedko shakhama samparka garnuhos.	Kumari Bank
yadi adhikatam ra nyunatam byajadar briddhi garda byajadar bichko antar pancha pratishat kayam bhaema darharu samayojana garna aawashyak …	यदि अधिकतम र न्यूनतम ब्याजदर बृद्धि गर्दा ब्याजदर बीचको अन्तर पाँच प्रतिशत कायम भएमा दरहरु समायोजन गर्न आवश्यक पर्दछ?	nirdharan hune adhikatam ra nyunatam byajadar 5 pratishatko antar bhitra kayam bhaema samayojana garna paidaina. prashna nan. 1 ma diieko udaharanma yasto antar 3.3 pratishat matra dekhaieko chha …	Nepal Rastra Bank

Agent — “कसले … ?” — Who …?

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
krishi wyapariharu ko hun?	कृषि व्यापारीहरू को हुन्?	krishi wyapariharu bainkadwara niyukta wyapariharu hun jasle kisana epa ra kisana kredit karda marphat kisanaharulai krishi utpadana tatha sewaharu kharid ra bikri garchhan.	Agricultural Development Bank Limited
enasiechaela-aaipiesa pranali kasle niyamana garcha?	एनसीएचएल-आईपीएस प्रणाली कसले नियमन गर्छ?	enasiechaela-aaipiesa pranali nepal rashtra bainkadwara swikrit ra niyamana garinchha.	Nepal Clearing House Limited
laghuwitta wittiya sansthako niyamana tatha supariwekshan kasle garcha?	लघुवित्त वित्तीय संस्थाको नियमन तथा सुपरिवेक्षण कसले गर्छ?	laghuwitta wittiya sansthaharuko niyamana tatha supariwekshan nepal rashtra bainkale nai gardachha.	Nepal Rastra Bank

Comparison / difference — “फरक / बीच / तुलना” — How is X different from Y?

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
mera dherai bainka khata chhan tara tinako mobile bainking number pharak chha. ke ma sabai linka garna sakchhu?	मेरा धेरै बैंक खाता छन् तर तिनको मोबाइल बैंकिङ नम्बर फरक छ। के म सबै लिङ्क गर्न सक्छु?	akaunta linkaka lagi eutai darta bhaeko mobile number aniwarya chha. phonape epa ra tapaika sabai bainkako mobile bainking epama darta bhaeko mobile number eutai hunuparchha. tapai aaphno bainka …	Fonepay
pranalile bankharu ra enasiechaelabich chek chhawiharu surakshit rupma sthanantaran bhaeko kasari sunishchit garcha?	प्रणालीले बैंकहरू र एनसीएचएलबीच चेक छविहरू सुरक्षित रूपमा स्थानान्तरण भएको कसरी सुनिश्चित गर्छ?	bankharu ra enasiechaelabich chek chhawiharu uchcha surakshit ra atyadhunik prawidhi purwadhar marphat surakshit rupma sthanantaran garinchha jasle sthanantaran hune chhawiharuko pramanikaran ra …	Nepal Clearing House Limited
bankma lagani garnu ra myuchuala phandama lagani garnuma ke pharak chha?	बैंकमा लगानी गर्नु र म्युचुअल फन्डमा लगानी गर्नुमा के फरक छ?	bainka nikshepma lagani garnelai nikshepakarta bhaninchha; myuchuala phandama lagani garnelai yunit holdar bhaninchha. nikshepakartale mag anusar byajasahit nikshep phirta dabi garna sakchhan. …	National Institute of Securities Markets

First-person / personal — “म / मेरो / मैले / मलाई” — I / my … (customer-voice)

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
maile ipiepha / pipiepha / sewaniwritti kosh / bhawishya nidhima yogadana gareko chhu. ke ma rashtriya pensana pranalima samel huna yogya …	मैले ईपीएफ / पीपीएफ / सेवानिवृत्ति कोष / भविष्य निधिमा योगदान गरेको छु। के म राष्ट्रिय पेन्सन प्रणालीमा सामेल हुन योग्य …	ho. rashtriya pensana pranali anya kunai pani pensana yojanasangai swechchhale samel huna sakinchha. tara euta wyaktisanga ekabhanda badhi rashtriya pensana pranali khata huna sakdaina.	Department of Financial Services, Government of India
malai prawidhik sahayog chahiyo. kaslai samparka garne?	मलाई प्राविधिक सहयोग चाहियो। कसलाई सम्पर्क गर्ने?	prawidhik sahayogka lagi 9801165565 ma samparka garnuhos wa merchantcare@khalti.com ma imel garnuhos.	Khalti
mobile bainking prayog garera mero paisa thagi bhaema ke garne?	मोबाइल बैंकिङ प्रयोग गरेर मेरो पैसा ठगी भएमा के गर्ने?	mobile bainking sewa ra khata blak garna turunta najikko shakhama januhos. tatkal mobile blakka lagi nepal bainka kal sentar 01-5971222 ma samparka garnuhos. thap rupma sabai sambandhit wiwaranasahit …	Nepal Bank Limited

Multi-question — more than one “?” — Two questions in one turn

Question (romanized)	Question (Devanagari original)	Answer (romanized, truncated)	Source
pensana ke ho? malai yasko aawashyakata kina chha?	पेन्सन के हो? मलाई यसको आवश्यकता किन छ?	pensanale anutpadak barshaharuma masik aamdani dinchha. aawashyakata: umerasangai aamdani kshamata ghatnu, ekal pariwar wriddhi, basainsarai, jiwanayapana kharcha badhnu, aayu badhnu ra …	Department of Financial Services, Government of India
ke khaltile phirta samarthana garcha? yadi ho bhane phirta kasari suru garne?	के खाल्टीले फिर्ता समर्थन गर्छ? यदि हो भने फिर्ता कसरी सुरु गर्ने?	ho, khaltima phirta wikalpa pani chha. yaska lagi wyaparile tranjeksana aaidima gaera phirta batanama klik gari grahak ko khalti aaidima ichchhit rakam phirta pathaunuparchha.	Khalti
tranjeksana pina matra birsie? ke mobile bainking pasawarda pani riset garna aawashyak chha wa pina matra riset garna sakinchha?	ट्रान्जेक्सन पिन मात्र बिर्सिएँ? के मोबाइल बैंकिङ पासवर्ड पनि रिसेट गर्न आवश्यक छ वा पिन मात्र रिसेट गर्न सकिन्छ?	tranjeksana pina matra birsie pani mobile bainking pasawarda ra pina duwai riset garnuparchha.	Nepal Bank Limited

---

10. Answer-style diversity

Answer shapes are classified from the Devanagari answer text (sentence-final । is a reliable boundary there) and shown with romanized examples.

10.1 Answer shape (rule-based classification)

Answer shape	Rows	Share	Avg chars (romanized)	Example (romanized, truncated)
Numeric / market-data sentence(s)	2,448	36.9%	261	5% ncc debenture 2086 (nccd86) ko prati sheyar aamdani 0 ra pi/i anupat 0 chha. prati sheyar buk bhyalyu 0, ra mulya-dekhi-buk mulya 0 chha. sheyarko 1-barshako …
One-sentence numeric fact	2,000	30.1%	70	esaeenaaaiemae ko aarthik barsha 2082/2083 ko kul labhansha 10% thiyo.
Multi-sentence explanation	1,047	15.8%	341	manisharule seyarma lagani garera kampani tatha arthatantrako dirghakalina wriddhibata labh lina khojchhan. seyarko mulya badhnuka sathai kehi kampanile labhansha pani …
Single-sentence fact	671	10.1%	178	nepal bainka limited grahak samarthanama tol phri number 16600137373 ra kal sentar 01-5971222 ma samparka garnuhos wa sahayogka lagi najikko shakhama januhos.
Step-by-step / instructional	244	3.7%	338	tranjeksana pasawarda 6 ankako pasawarda ho juna karobar pura garna prayog hunchha ra karobar simaka aadharma otipisanga sanyojanama prayog garna sakinchha. …
Yes/No-led answer	225	3.4%	168	ho, enaaaraaaile enaaarai wa enaaarao bainka khatama raheko rakam prayog garera bharatiya myuchuala phandama esaaaipi garna sakchhan. prakriya bharatiya basindajastai …

10.2 Answer features (computed on the romanized text)

Feature	All rows	% of all	FAQ part	% of FAQ part
Contains digits (ASCII)	4,996	75.3%	548	25.1%
Contains currency (ru. / rupaiya / neru)	1,986	29.9%	139	6.4%
Contains a percentage (%)	3,730	56.2%	27	1.2%
Contains an imperative instruction (…nuhos)	324	4.9%	324	14.8%
Contains ;-separated list items	43	0.6%	43	2.0%
Contains an arrow →	10	0.2%	10	0.5%
Contains a URL	4	0.1%	4	0.2%
Contains an e-mail / @ sign	25	0.4%	25	1.1%
Contains a newline	0	0.0%	0	0.0%
Contains upper-case letters	35	0.5%	9	0.4%

10.3 Number of sentences per answer (counted on the Devanagari answer: ।, ?, !)

Sentences	All rows	% of all	FAQ part	% of FAQ part
1	3,370	50.8%	740	33.8%
2	936	14.1%	725	33.2%
3	1,256	18.9%	452	20.7%
4	947	14.3%	144	6.6%
5	69	1.0%	69	3.2%
6+	57	0.9%	57	2.6%

10.4 Example answers of different shapes (FAQ part, romanized)

Shape	Question	Answer (truncated)	Source
Yes/No-led answer	ke bachat bainka khata bina atal pensana yojana khata kholna sakinchha?	hoina, bachat bainka wa hulak bachat khata aniwarya chha.	Department of Financial Services, Government of India
Yes/No-led answer	ke namankana aniwarya chha?	ho, atal pensana yojana khatama namankana wiwaran aniwarya chha.	Department of Financial Services, Government of India
Step-by-step / instructional	karyalaya kewaisi wiwaran adyawadhik garna ke garnuparchha?	koshakatti garne karyalayale koshko webasaitama gai karyalaya kewaisima klik gari wiwaran bharnupardachha. pesh garepachhi imelma linka aaunchha, tyasma klik garnupardachha. kosh karyalayabata swikrit bhaepachhi sambandhit karyalayako imelma prayogakarta naam tatha pasawarda prapta hunechha.	कर्मचारी सञ्चय कोष
Step-by-step / instructional	analaina sewako prayogakarta naam tatha pasawarda birsie bhane kasari punah set garne?	koshko webasaitama gai lagaina → wyaktigat lagainama januhos. pasawarda birsie bhannema klik garepachhi prayogakarta pahichana (naam ra janma miti) bharnuhos. mobilema aaeko ekapatak prayog hune pasawarda prawishta gari naya pasawarda pariwartana garna sakinchha.	कर्मचारी सञ्चय कोष
Single-sentence fact	kyuaara kod ke ho?	kyuaara kod bhaneko sarbharma bhandaran garieko kisanako janakari samawesh bhaeko kod ho jasbata ilektronik upakaran marphat grahak ko karobar ra wiwaran herna sakinchha.	Agricultural Development Bank Limited
Single-sentence fact	yo epa prayog garne mukhya sarokarawalaharu ko hun?	yo epa prayog garne mukhya sarokarawalaharu kisana, krishi wyapari ra kisana kyashapointa ejenta, laina ejensiharu ra niyamak nikayaharu hun.	Agricultural Development Bank Limited
Multi-sentence explanation	mero naya khata kahile sakriya hunechha?	khata pushti sandesh prapta garepachhi tapai tyas khatama paisa jamma garna saknuhunchha. rakam nikalne suwidhaka lagi, tapaile shakhama aawashyak sabai mul kagajatharu pesh garnuparchha. tapaile aawedanama hastakshar ra aunlachhap garnuparchha. kagajat pramanikaran pachhi shakhale hastakshar karda tayar garnechha ra chek buk …	Agricultural Development Bank Limited
Multi-sentence explanation	mero naya khata kahiledekhi byaj jamma huna thalchha ra punjikrit hunchha?	byaj dine khataharuma tapaiko khatama paisa jamma bhaepachhi byaj jamma huna thalchha. bainkale dainik maujdatma byaj pradana gardachha ra pratyek traimasik antyama byajko rakam tapaiko khatama punjikrit hunechha.	Agricultural Development Bank Limited

---

11. Romanization characteristics

11.1 What the romanization looks like

The Latin text is a phonetic, rule-based transliteration of the Devanagari spelling (not the informal “Romanized Nepali” people type in chats). Observed regularities, with word pairs taken from the aligned files:

Devanagari	Romanized	Aligned occurrences
छ	chha	5,307
छन्	chhan	1,371
हुन्छ	hunchha	1,038
सक्छ	sakchha	257
थियो	thiyo	4,350
आर्थिक	aarthik	4,488
वर्ष	barsha	4,101
लाभांश	labhansha	3,786
प्रतिशत	pratishat	1,381
शेयर	sheyar	1,909
सेयर	seyar	1,574
पूँजीकरण	punjikaran	1,212
ऋण	rin	205
ऋणपत्र	rinapatra	27
विवरण	wiwaran	256
विद्युतीय	widyutiya	65
भुक्तानी	bhuktani	502
बैंक	bainka	576
फन्ड	phanda	180
पेन्सन	pensana	155
अनलाइन	analaina	133
म्युचुअल	myuchuala	223
क्यूआर	kyuaara	144
एसएमएस	esaemaesa	64
एनआरएन	enaaaraena	15
अन्तर्राष्ट्रिय	antarrashtriya	52
नेपाल	nepal	576
गर्‍यो	gar‍yo	668
१०००	1000	244

Takeaways from these pairs:

• Everything is lower-case ASCII except 36 rows that keep upper-case API / AI.
• Devanagari digits are converted to ASCII digits everywhere (१००० → 1000, २०७७/२०७८ → 2077/2078). In the Devanagari file 75% of answers use Devanagari digits; here digits are 100% ASCII.
• Vowel length is expanded rather than dropped (आ → aa: aarthik, analaina), छ → chha, श → sh, फ → ph, भ → bh, ऋ → ri, ंश → nsha.
• A trailing a is added to many consonant-ending words (बैंक → bainka, फन्ड → phanda, पेन्सन → pensana, अनलाइन → analaina) but not to all (नेपाल → nepal, शेयर → sheyar). So the output differs from how people usually type these words (bank, fund, pension, online).
• English loanwords written in Devanagari are transliterated back phonetically, not restored (म्युचुअल फन्ड → myuchuala phanda, क्यूआर → kyuaara).
• English text that was already in Latin letters in the source (company names, tickers, product names such as Gurkhas Finance Ltd. (GUFL)) is simply lower-cased (gurkhas finance ltd. (gufl)).
• Treatment of व is lexical, not uniform: वर्ष → barsha but विवरण → wiwaran, विद्युतीय → widyutiya.
• Spelling variants in the Devanagari source carry over: शेयर (sheyar, 1,909 tokens) and सेयर (seyar, 1,574 tokens) are the same word; likewise फन्ड and फण्ड both become phanda.

11.2 Character inventory

Character class	Count	Share of all characters
Lower-case letters a–z	1,419,199	75.6%
Upper-case letters A–Z	184	0.01%
ASCII digits 0–9	122,031	6.5%
Spaces	248,603	13.2%
ASCII punctuation / symbols	86,466	4.6%
Non-ASCII characters	1,242	0.07%

All non-ASCII characters in the file

Character	Code point	Name	Occurrences	Rows
(ZWJ)	U+200D	ZERO WIDTH JOINER	676	676
–	U+2013	EN DASH	154	89
“	U+201C	LEFT DOUBLE QUOTE	106	56
”	U+201D	RIGHT DOUBLE QUOTE	106	56
‘	U+2018	LEFT SINGLE QUOTE	65	41
’	U+2019	RIGHT SINGLE QUOTE	65	41
—	U+2014	EM DASH	36	20
→	U+2192	RIGHTWARDS ARROW	22	10
÷	U+00F7	DIVISION SIGN	7	6
×	U+00D7	MULTIPLICATION SIGN	2	2
ॉ	U+0949	DEVANAGARI VOWEL SIGN CANDRA O	1	1
…	U+2026	ELLIPSIS	1	1
़	U+093C	DEVANAGARI SIGN NUKTA	1	1

11.3 Artifacts in the romanized text

Artifact	Rows	Details	Suggested handling
Zero-width joiner U+200D	676	Comes from the Devanagari गर्‍यो (“did”) written as gar‍yo; 669 of the rows are NEPSE-dividend questions (… ghoshana gar‍yo?); also pur‍yau	Strip with text.replace("\u200d","") (§3.5)
Stray Devanagari characters	2	2 rows (lemonn_all_faq_000127: praॉspektas; wealthserene_faq_000012: rudha़wadi) — vowel sign / nukta left untransliterated	Remove with a Devanagari-range regex
Upper-case letters	36	API / AI kept in Latin capitals; everything else lower-cased	Lower-case if you need uniform case
Rows with any non-ASCII character	886	ZWJ, typographic quotes/dashes, →, ×, ÷, …, stray Devanagari	Apply normalize() (§3.5)
Typographic quotes / dashes	196	Carried over from the source FAQ text	Normalise to ASCII

11.4 Tickers and names

• ShareHub tickers are the original NEPSE symbols in lower-case Latin (aclbsl, gufl, lbld86) — compact and unambiguous, 415 distinct symbols.
• Dividend tickers are different: they were Nepali-transliterated letter names in the Devanagari file (e.g. एसीएलबीएसएल), and in the romanized file they become letter-name strings:

Romanized ticker in dividend rows	Devanagari form	Characters
esielabiesaela	एसीएलबीएसएल	14
edibiela	एडीबीएल	8
eechapisi	एएचपीसी	9
ekepiela	एकेपीएल	8
eelabiesaela	एएलबीएसएल	12
eelaaaisiela	एएलआईसीएल	12
eenaelabi	एएनएलबी	9
bieaarayuena	बीएआरयूएन	12

The ticker is therefore hard to recover by de-romanizing; if you need the true ticker, map it via the Devanagari file / the ShareHub symbol list.

11.5 Vocabulary statistics

Measure	Romanized (this file)	Devanagari original
Alphabetic tokens (all occurrences)	239,137	264,602
Distinct alphabetic types	9,884	12,144
Type/token ratio	0.041	0.046
Types occurring once	3,958	5,452
Mean characters per row (Q+A)	283	250

Romanization merges some distinct Devanagari spellings (and drops case/punctuation attached to tokens), so there are 19% fewer word types than in the Devanagari text. The romanized text is on average 1.14× as long in characters (per-sub-corpus ratios below).

Subset	Romanized / Devanagari length (characters)
ShareHub	1.10×
NEPSE Dividend	1.18×
Institutional FAQ	1.16×

30 most frequent romanized words

#	Word	Count	Gloss
1	ko	9,055	of / 's
2	ra	6,737	and
3	ru	6,272	Rs. (rupees)
4	chha	5,361	is
5	aarthik	4,496	financial / economic
6	thiyo	4,354	was
7	mulya	4,232	price / value
8	barsha	4,108	year
9	labhansha	3,795	dividend
10	kul	3,496	total
11	kati	3,317	how much
12	ho	2,866	is (copula)
13	limited	2,546	Ltd. (kept in Latin)
14	ka	2,342	of (plural)
15	ke	2,331	what
16	karobar	2,314	transaction / trading
17	lagi	1,970	for
18	sheyar	1,911	share
19	garna	1,724	to do
20	buk	1,621
21	seyar	1,576	share (variant spelling)
22	le	1,416	by (agent marker)
23	pratishat	1,383	percent
24	chhan	1,378	are
25	bonas	1,369	bonus
26	bajar	1,353	market
27	uchcha	1,269	high
28	sheyarharu	1,232	shares
29	barshako	1,231	of the year
30	ma	1,228	in

---

12. Length statistics

Characters of the romanized text; words are whitespace-separated tokens.

Subset	Field	Min	Mean	Median	P95	Max	Mean words
All rows	Question	11	75	76	110	902	10.8
All rows	Answer	6	208	200	411	1,719	28.7
ShareHub	Question	46	87	87.0	115	132	12.6
ShareHub	Answer	129	261	264.0	337	357	38.3
NEPSE Dividend	Question	60	80	78.0	95	101	11.6
NEPSE Dividend	Answer	58	70	70.0	78	82	10.3
Institutional FAQ	Question	11	58	52	110	902	8.0
Institutional FAQ	Answer	6	273	243	593	1,719	34.8

Answer-length distribution (all rows)

Answer length (chars)	All rows	% of all	FAQ part	% of FAQ part
< 50	27	0.4%	27	1.2%
50–99	2,219	33.4%	219	10.0%
100–199	1,060	16.0%	618	28.3%
200–299	1,692	25.5%	530	24.2%
300–499	1,446	21.8%	602	27.5%
500–999	178	2.7%	178	8.1%
1000+	13	0.2%	13	0.6%

Question-length distribution (all rows)

Question length (chars)	All rows	% of all	FAQ part	% of FAQ part
< 30	205	3.1%	205	9.4%
30–49	794	12.0%	785	35.9%
50–69	1,322	19.9%	607	27.8%
70–99	3,542	53.4%	421	19.3%
100–149	744	11.2%	141	6.4%
150+	28	0.4%	28	1.3%

Per-topic-group length (FAQ part)

Topic group	Rows	Mean Q	Mean A	Median A	P95 A	Max A
Central bank & monetary policy	388	79	345	281.5	824	1,719
Commercial / development banks (retail banking)	615	59	213	185	475	808
Digital payments, wallets & clearing	414	53	243	189.0	613	1,209
Capital-market & investing education	464	43	323	317.5	593	928
Pension & provident funds	133	61	238	205	537	758
Insurance	73	51	264	240	534	585
Microfinance, cooperatives, lending & development	100	66	301	256.0	621	832

Tokenizer note: romanized Nepali is split into many more sub-word tokens by typical LLM tokenizers than the same content in English, so budget sequence length accordingly and measure with your own tokenizer.

---

13. Data quality notes and known limitations

Topic	Observation	Suggested handling
Script metadata mismatch	script = Deva and content_script = देवनागरी in every row, but the text is Latin	Override script to Latn (§3.6); do not filter on it
Romanization scheme	Rule-based phonetic transliteration with some lexical irregularities (e.g. bainka, phanda, wiwaran vs barsha); differs from natural chat-style Romanized Nepali	If you target user-typed Romanized Nepali, evaluate on natural data or add spelling-noise augmentation
No spelling normalisation	Source spelling variants stay different after romanization (sheyar/seyar)	Merge variants with a small lexicon if needed
Dividend tickers	Letter-name strings (edibiela) instead of real tickers	Use the Devanagari file or ShareHub symbols to recover tickers
Template dominance	67.0% of rows are templated; ShareHub has 8 fixed question forms, dividends 6 forms	Down-sample if you need linguistic diversity; weight by source_name
Time-sensitive facts	ShareHub prices, market caps, ratios and “as of ” statements are a snapshot (most last-price answers say septembar 17 samma, a few carry earlier dates); dividend values are historical	Do not treat as current market data
Behavior diversity	Only 2 behaviors; no refusals, clarifications, multi-turn, system prompts	Mix with other instruction data if you need these behaviors
Label coverage	question_type ‘जानकारीमूलक’ and ‘प्रक्रियामूलक’ exist only for Lemonn rows	Use text-based patterns (§9)
Duplicates	25 duplicate questions and 4 fully duplicated (Q, A) pairs — same as the Devanagari file (romanization neither creates nor removes duplicates)	Deduplicate on (question, answer) if needed; ids are unique
Licence	300 Lemonn rows have license=unknown	Filter with license_tier == "permissive" for strictly-licensed use
Source labelling	Labels kept as in the Devanagari release (mixed English/Devanagari; NPS FAQ attributed to an Indian ministry; Fonepay category written “फोनपे”)	Use source_name as the stable key
Metadata stays Devanagari	Category / sub-domain / behavior labels are not romanized	Map them via a lookup table if you need Latin labels
Empty url	url is empty for every row	—
Metadata key sets differ	ShareHub rows carry 2 extra keys	Use .get(); see HF demo in §3.3
Translation artifacts	FAQ answers were localised into Nepali from other-language originals before romanization, so occasional literal phrasing or transliterated terms occur	Spot-check before using as gold answers
Placeholder zeros	In the “key financial indicators” answers, 121 of 401 rows (75 debentures, 43 mutual funds, 3 stocks) report EPS and P/E as 0 because the metric does not apply	Filter or treat 0 as “not applicable”
Unit/range oddities	Some debenture “open range” answers show identical lower and upper bounds (ru. 1000 dekhi ru. 1000) because the source band is a single price	Drop the open-range rows of debentures if undesirable

---

14. Recommended usage

Goal	Suggested subset / approach
Instruction-tune a Romanized-Nepali finance assistant	FAQ part (2,187) + a capped sample of ShareHub/dividend rows (e.g. ≤ 500 each) so templated data does not dominate; apply normalize()
Transliteration / script-conversion research	Join with the Devanagari file on id (§3.4) — 6,635 aligned (Devanagari, Latin) question and answer pairs
Cross-script robustness	Train on a mix of both releases, evaluate on each script separately
Numeric grounding / structured-data-to-text	ShareHub + dividends (4,448 rows); digits are already ASCII, which simplifies number handling
Evaluate Nepali finance QA	Hold out stratified by source_name (§3.9); evaluate templated and free-form parts separately
Retrieval / RAG evaluation	FAQ-part questions as queries and answers as ground-truth passages
Strictly-licensed training	license_tier == "permissive" (6,335 rows)

Licensing

Row-level licence is stored in license / license_tier. 6,335 rows are Apache-2.0 (permissive) and 300 rows (source: Lemonn) are unknown. Several sources are commercial/company FAQ pages or public market data; confirm that your use case (especially commercial redistribution) is compatible with the upstream terms before use. The romanized text is a transliteration of the Devanagari release and carries the same licence status per row.

Citation

Romanized Nepali Finance FAQ — instruction-following dataset (6,635 single-turn conversations), v1.