Datasets:
task_id stringlengths 5 16 | split stringclasses 1
value | domain stringclasses 8
values | difficulty stringclasses 2
values | prompt stringlengths 196 3.66k | gold stringlengths 1 74 | grader unknown |
|---|---|---|---|---|---|---|
se-log-5xx | train | software | easy | Here is an excerpt from a web server access log.
113.36.130.50 - - [12/Oct/2026:10:52:27 +0000] "GET /api/pay HTTP/1.1" 200 5502
202.166.122.164 - - [12/Oct/2026:10:14:35 +0000] "GET /api/orders HTTP/1.1" 500 7049
202.166.122.164 - - [12/Oct/2026:10:44:16 +0000] "GET /health HTTP/1.1" 200 3670
138.187.98.121 - - [12/O... | 11 | {
"type": "int"
} |
se-log-uniq-ip | train | software | medium | Here is an excerpt from a web server access log.
14.94.226.230 - - [12/Oct/2026:10:37:57 +0000] "GET /api/orders HTTP/1.1" 200 8474
129.113.74.126 - - [12/Oct/2026:10:33:23 +0000] "GET /api/orders HTTP/1.1" 500 2856
183.93.196.75 - - [12/Oct/2026:10:37:18 +0000] "GET /static/app.js HTTP/1.1" 400 4951
18.126.47.213 - -... | 6 | {
"type": "int"
} |
se-pyexpr | train | software | easy | Evaluate this Python 3 expression exactly as the interpreter would:
int('101101', 2) + max(4, 9, 2) * min(7, 3)
What integer does it produce? Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or extra text, inside the answer field. | 72 | {
"type": "int"
} |
se-json-patch | train | software | medium | Start from this JSON object:
{"active": true, "count": 14, "name": "alpha", "score": 9.5, "tags": ["x", "y"]}
Apply these operations in order:
1. set key 'count' to 21
2. remove key 'active'
3. append 'z' to the list at key 'tags'
4. set key 'owner' to 'ops'
Answer with the resulting JSON object as a sin... | {"count":21,"name":"alpha","owner":"ops","score":9.5,"tags":["x","y","z"]} | {
"type": "json_fields"
} |
se-ipv4-count | train | software | medium | Below is a firewall draft. A 'bare IPv4 address' means exactly four dot-separated decimal octets, each 0-255, standing alone as a token.
route 8.8.8.8 via gw1
route ::1 via gw2
note version 1.2.3.4.5 released
allow 256.1.1.1 any
permit 172.16.0.9 log
forward 10.0.300.14 to pool_b
deny 192.168.4 any
mirror 203.0.113.7 ... | 6 | {
"type": "int"
} |
se-log-errors | train | software | easy | An application log shows:
2026-10-12 09:17:47 [INFO] worker-2 processed batch 495
2026-10-12 15:27:55 [DEBUG] worker-3 processed batch 872
2026-10-12 12:53:36 [WARN] worker-4 processed batch 768
2026-10-12 22:14:42 [WARN] worker-1 processed batch 404
2026-10-12 03:55:32 [INFO] worker-1 processed batch 782
2026-10-12 1... | 2 | {
"type": "int"
} |
in-btu-kw | train | industrial | easy | A heater is rated at 42000 BTU per hour. Convert this to kilowatts (1 BTU/h = 0.000293071 kW). Answer in kW rounded to two decimal places. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or extra text, inside the answer field. | 12.31 | {
"type": "num",
"rel_tol": 0.005
} |
in-sensor-median | train | industrial | medium | A temperature sensor logged these readings (time, deg C):
8:00 27.9
8:30 26.4
9:00 23.3
9:30 20.6
10:00 26.8
10:30 30.6
11:00 23.6
11:30 29.1
12:00 18.5
12:30 24.9
13:00 20.6
13:30 24.4
14:00 25.9
14:30 29.1
15:00 24.8
15:30 29.3
16:00 22.2
16:30 19.7
What is the median of the readings from 10:00 th... | 24.6 | {
"type": "num",
"rel_tol": 0.002
} |
in-battery-soc | train | industrial | easy | A battery has a rated capacity of 100 Ah and starts fully charged. It discharges at a constant 7.5 A for 6 hours (ignore efficiency losses). What is its state of charge afterward, as a percentage? Round to one decimal place; answer with just the number. Reply with exactly one JSON object and nothing else: {"answer": ".... | 55.0 | {
"type": "num",
"rel_tol": 0.005
} |
in-parts-ooc | train | industrial | medium | Twelve machined parts must measure 25.00 mm +/- 0.08 mm. Measured diameters (mm):
P-01 24.969
P-02 24.876
P-03 25.025
P-04 24.890
P-05 24.995
P-06 24.950
P-07 25.059
P-08 24.897
P-09 25.140
P-10 24.872
P-11 24.870
P-12 25.052
How many parts are out of tolerance? Answer is one integer. Reply with exactly o... | 6 | {
"type": "int"
} |
sc-rate-const | train | science | easy | A first-order reaction has initial concentration [A]0 = 0.8 mol/L. After t = 14 minutes the concentration is [A] = 0.26 mol/L. Using ln([A]0/[A]) = k t, find the rate constant k in min^-1. Round to four decimal places. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for... | 0.0803 | {
"type": "num",
"rel_tol": 0.01
} |
sc-moles | train | science | easy | How many moles are in 37.0 g of NaCl? Use a molar mass of 58.44 g/mol. Round to three decimal places. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or extra text, inside the answer field. | 0.633 | {
"type": "num",
"rel_tol": 0.005
} |
sc-ph | train | science | medium | 20 mL of 0.05 M HCl is diluted with water to a final volume of 500 mL. HCl is a strong acid. What is the pH of the diluted solution? Round to two decimal places. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or extra text, inside the answer field. | 2.70 | {
"type": "num",
"rel_tol": 0.01
} |
sc-half-life | train | science | medium | A cobalt-60 source has activity proportional to a mass of 480 mg and a half-life of 5.27 years. How much remains after 15.81 years? Round to the nearest whole milligram. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or extra text, inside the answer field... | 60 | {
"type": "num",
"rel_tol": 0.01
} |
of-invoice | train | office | easy | An invoice lists these line items:
Consulting, 12 h: $180.00 x 12
License fee: $450.00 x 1
Travel: $612.40 x 1
Support retainer: $300.00 x 1
A correction memo says: the Travel charge is withdrawn and replaced by a $480.00 flat expense, and a 10% courtesy credit applies to the corrected total. What is the fina... | 3051.00 | {
"type": "num",
"rel_tol": 0.001
} |
of-meeting | train | office | medium | Three colleagues are busy during these intervals (work day 09:00-17:00):
Ana: 09:00-09:45, 10:30-11:15, 13:00-14:30, 16:00-17:00
Bo: 09:00-10:00, 11:00-12:00, 13:30-15:00
Cy: 09:30-10:30, 11:30-12:30, 14:00-16:30
They need a 30-minute meeting, aligned to quarter-hour starts. What is the earliest start time that... | 12:30 | {
"type": "string"
} |
of-long-words | train | office | easy | Here is an internal memo:
"Please route the quarterly vendor summary to procurement before Friday. The revised totals exclude the duplicate freight charge flagged in June. Confirm receipt with accounting and archive the signed copy. Do not circulate drafts outside the finance group."
How many words in the memo ar... | 23 | {
"type": "int"
} |
of-expense | train | office | medium | Expense policy caps each line item at: meals $60.00, lodging $200.00, transport $50.00.
Submitted items:
meals: $46.20
meals: $78.40
lodging: $210.00
lodging: $185.00
transport: $64.00
transport: $55.00
What is the total reimbursable amount after applying the per-item caps? Answer in dollars to two decim... | 591.20 | {
"type": "num",
"rel_tol": 0.001
} |
fi-loan | train | finance | easy | A loan of $24,000 at 7.2% annual interest is repaid in 36 equal monthly payments. Using M = P r / (1 - (1+r)^-n) with r the monthly rate, what is the monthly payment? Round to two decimals; answer with just the number. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for... | 743.25 | {
"type": "num",
"rel_tol": 0.005
} |
fi-npv | train | finance | medium | Cash flows for a project are: year 0: $-5,000, year 1: $+1,600, year 2: $+1,900, year 3: $+2,100, year 4: $+1,700 (year 0 is paid today). At a discount rate of 9%, what is the net present value in dollars? Round to two decimals; answer with just the number (negative if the project loses money). Reply with exactly one J... | 892.99 | {
"type": "num",
"rel_tol": 0.01
} |
fi-cagr | train | finance | easy | An investment grows from $18,500 to $31,200 over 4 years. What is the compound annual growth rate as a percentage? Round to one decimal place; answer with just the number. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or extra text, inside the answer fie... | 14.0 | {
"type": "num",
"rel_tol": 0.01
} |
fi-tax | train | finance | medium | Progressive tax brackets: first $12,000 at 0%; next $20,000 at 10%; next $38,000 at 22%; remainder at 32%. What is the total tax on an income of $84,500? Round to two decimals; answer with just the number. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or... | 15000.00 | {
"type": "num",
"rel_tol": 0.002
} |
fi-simple-int | train | finance | easy | $7,600 is placed in a simple-interest account at 4.8% per year. How much interest accrues after 30 months? Round to two decimals; answer with just the number. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or extra text, inside the answer field. | 912.00 | {
"type": "num",
"rel_tol": 0.005
} |
ma-edges | train | math | easy | An undirected graph is given as an adjacency list:
0: [1, 4]
1: [0]
2: [4, 5]
3: [5, 6]
4: [0, 2, 5, 7]
5: [2, 3, 4, 6]
6: [3, 5]
7: [4]
How many edges does it have? Answer is one integer. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units o... | 9 | {
"type": "int"
} |
ma-triangles | train | math | medium | An undirected graph is given as an adjacency list:
0: [1, 2, 3]
1: [0, 2, 3]
2: [0, 1, 3]
3: [0, 1, 2, 4, 5]
4: [3, 5]
5: [3, 4, 6]
6: [5]
How many triangles (three mutually connected vertices) does it contain? Answer is one integer. Reply with exactly one JSON object and nothing else: {"answer": "..."}... | 5 | {
"type": "int"
} |
ma-modpow | train | math | easy | Compute 7^233 mod 19. Answer is one integer. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or extra text, inside the answer field. | 11 | {
"type": "int"
} |
ma-recurrence | train | math | medium | A sequence obeys a_n = 3 a_(n-1) + (-2) with a_0 = 2. Find a_8. Answer is one integer. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or extra text, inside the answer field. | 6562 | {
"type": "int"
} |
ma-set-int | train | math | medium | Set A = {12, 21, 29, 32, 35, 38, 50, 51, 58}
Set B = {28, 29, 34, 38, 40, 47, 55, 63, 67}
List the elements of A intersect B as comma-separated integers in ascending order (or 'none' if empty). Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or extra text... | 29,38 | {
"type": "set"
} |
cy-cidr | train | cybersecurity | easy | Which of these addresses fall inside the network 10.20.0.0/22?
10.20.1.44
10.20.3.200
10.20.4.9
10.19.255.1
10.20.0.1
10.20.5.12
10.21.0.7
10.20.2.77
Answer with the count as one integer. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or... | 4 | {
"type": "int"
} |
cy-acl | train | cybersecurity | medium | A firewall evaluates rules top-down and stops at the first match:
1 deny tcp 10.0.0.0/8 any port 23
2 allow tcp 10.4.0.0/16 any port 22
3 deny tcp any any port 22
4 allow tcp 10.0.0.0/8 any any
5 deny any any any
Packet: tcp 10.4.1.20 -> 172.16.5.5 port 22
Is it allowed or denied? Answer 'allow' or 'deny'.... | allow | {
"type": "choice",
"options": [
"allow",
"deny"
]
} |
cy-b64 | train | cybersecurity | easy | A config file contains this base64-encoded value:
api_token = "cm90YXRlLWtleXMtcXVhcnRlcmx5"
Decode it and answer with the plaintext string. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or extra text, inside the answer field. | rotate-keys-quarterly | {
"type": "string"
} |
cy-bruteforce | train | cybersecurity | medium | An authentication log shows:
Oct 12 06:17:52 sshd[7447]: Failed password for svc-monitor from 198.51.100.94 port 25812 ssh2
Oct 12 23:49:39 sshd[6197]: Accepted password for www-data from 203.0.113.215 port 56287 ssh2
Oct 12 13:20:30 sshd[8168]: Failed password for svc-monitor from 198.51.100.231 port 33714 ssh2
Oct 1... | svc-monitor | {
"type": "string"
} |
cy-subnet | train | cybersecurity | easy | A subnet is configured as 192.168.44.0/26. How many usable host addresses does it have (excluding network and broadcast addresses)? Answer is one integer. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or extra text, inside the answer field. | 62 | {
"type": "int"
} |
me-srt-shift | train | media | easy | An SRT subtitle file has these cues:
1
00:00:04,200 --> 00:00:06,100
(cue text)
2
00:00:07,050 --> 00:00:09,400
(cue text)
3
00:00:11,300 --> 00:00:13,750
(cue text)
Every cue must be shifted later by exactly 2.35 seconds. What is the new start timestamp of cue 2, in SRT format HH:MM:SS,mmm? Reply with exactly one JS... | 00:00:09,400 | {
"type": "string"
} |
me-ppm | train | media | medium | Here is a plain-text PPM (P3) image:
P3
4 3
255
79 177 163
159 94 179
242 194 35
248 80 25
167 162 42
244 35 208
46 123 153
225 26 190
12 242 26
39 217 145
185 223 6
75 24 68
What is the mean value of the red channel across all pixels? Round to one decimal place; answer with just the number. Reply with exactly one JS... | 143.4 | {
"type": "num",
"rel_tol": 0.005
} |
me-readtime | train | media | easy | A script draft is 1140 words long. At a narration pace of 190 words per minute, how long is the read? Answer in M:SS format, rounded to the nearest second. Reply with exactly one JSON object and nothing else: {"answer": "..."}. Give only the value asked for, no units or extra text, inside the answer field. | 6:00 | {
"type": "string"
} |
Quant Desk — OpenEnv Arena dataset
Data for the Quant Desk OpenEnv environment: 36 deterministic, self-contained single-answer tasks spanning all eight Arena domains (software engineering, industrial & physical systems, natural science, office & white-collar work, finance & economics, math & formal reasoning, cybersecurity, media & content production).
What the environment teaches
Analyst-style precision: read a small self-contained brief (a log excerpt, a
sensor table, an invoice plus correction memo, a firewall ruleset, an SRT file),
compute the exact requested value, and reply with exactly one JSON object
{"answer": "..."}. The action schema is a single answer string, which keeps
episodes to one required turn and makes strict-reply format discipline part of
the learned behavior. Up to three attempts are allowed per episode, with
feedback between attempts.
How the reward works
Each step({"answer": ...}) is graded deterministically per task:
int— first integer in the answer must equal the gold integernum— numeric match withinrel_tol(full), 10x tol (0.5), 100x tol (0.2)choice— normalized match against the gold optionset— token-set F1 against the gold setjson_fields— fraction of gold top-level fields present and equalstring— whitespace/case-normalized exact match
Step reward is that attempt's score. Episodes end on a perfect score or after
3 attempts; the terminal reward is the best attempt score, always in [0, 1].
Files
tasks.jsonl— the 36 task instances actually served by the image:task_id,split(alltrain),domain,difficulty,prompt,gold,grader(type + parameters)gen_tasks.py— the generator that producedtasks.jsonlenv/— full environment source (server, graders, models, openenv.yaml)env-README.md— environment documentation
Provenance
- Image:
ghcr.io/vaskrokodile/quant-desk:v1(linux/amd64) - Source: https://github.com/Vaskrokodile/quant-desk
- Built on OpenEnv pinned at
86a180ede21e044f7929b9a7783ad83aa67d83a3
- Downloads last month
- 54