id string | skill string | difficulty string | bank string | trap_id string | trap string | instruction string | instruction_summary string | license string | github_repo string | github_tag string | github_task_url string | compact_shell_version string | benchmark_version string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
loc-member-discount | loc | hard | base-47 | L1 | Config already correct; live code hardcodes the wrong value | The shop project is at `/app/repo`. Member customers should get a 10% discount, but they are charged the full list price.
Identify every file that must be changed to resolve this issue. Write each path relative to `/app/repo` using forward slashes, one path per line, to `/app/answer.txt`. Order does not matter. Do not... | The shop project is at `/app/repo`. Member customers should get a 10% discount, but they are charged the full list price. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/loc-member-discount | 0.1.1 | benchmark-v1.0.1 |
loc-bind-host | loc | uncalibrated | base-47 | L2 | Same as L1, plus an unused file that shares the bad string | The service at `/app/repo` should bind `0.0.0.0` from config. It currently binds `127.0.0.1`.
Identify every file that must be changed to resolve this issue. Write each path relative to `/app/repo` using forward slashes, one path per line, to `/app/answer.txt`. Order does not matter. Do not list test files unless they... | The service at `/app/repo` should bind `0.0.0.0` from config. It currently binds `127.0.0.1`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/loc-bind-host | 0.1.1 | benchmark-v1.0.1 |
loc-vip-two-files | loc | hard | base-47 | L3 | Gold is two files among similar names | The shop at `/app/repo` is supposed to charge VIP customers 80% of list price. VIP customers currently pay the full list price.
Identify every file that must be changed to resolve this issue. Write each path relative to `/app/repo` using forward slashes, one path per line, to `/app/answer.txt`. Order does not matter. ... | The shop at `/app/repo` is supposed to charge VIP customers 80% of list price. VIP customers currently pay the full list price. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/loc-vip-two-files | 0.1.1 | benchmark-v1.0.1 |
loc-similar-filenames | loc | medium | base-47 | L4 | Similar helper names; gold is one nested path | The CLI at `/app/repo` should collapse internal whitespace in user names before saving. It currently does not.
Identify every file that must be changed to resolve this issue. Write each path relative to `/app/repo` using forward slashes, one path per line, to `/app/answer.txt`. Order does not matter. Do not list test ... | The CLI at `/app/repo` should collapse internal whitespace in user names before saving. It currently does not. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/loc-similar-filenames | 0.1.1 | benchmark-v1.0.1 |
loc-traceback-helper | loc | medium | base-47 | L5 | Do not list the traceback wrapper; list the parser | Running `python3 /app/repo/app.py` fails while loading a record file.
Identify every file that must be changed to resolve this issue. Write each path relative to `/app/repo` using forward slashes, one path per line, to `/app/answer.txt`. Order does not matter. Do not list test files unless they themselves contain the ... | Running `python3 /app/repo/app.py` fails while loading a record file. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/loc-traceback-helper | 0.1.1 | benchmark-v1.0.1 |
loc-failing-test-impl | loc | medium | base-47 | L6 | Do not list the failing test; list the implementation | The project at `/app/repo` has a failing unit test for invoice totals.
Identify every file that must be changed to resolve this issue. Write each path relative to `/app/repo` using forward slashes, one path per line, to `/app/answer.txt`. Order does not matter. Do not list test files unless they themselves contain the... | The project at `/app/repo` has a failing unit test for invoice totals. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/loc-failing-test-impl | 0.1.1 | benchmark-v1.0.1 |
loc-reexport | loc | uncalibrated | base-47 | L7 | Do not list a re-export `__init__.py`; list the defining module | The service at `/app/repo` should bind `0.0.0.0` from config. It currently binds `127.0.0.1`.
Identify every file that must be changed to resolve this issue. Write each path relative to `/app/repo` using forward slashes, one path per line, to `/app/answer.txt`. Order does not matter. Do not list test files unless they... | The service at `/app/repo` should bind `0.0.0.0` from config. It currently binds `127.0.0.1`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/loc-reexport | 0.1.1 | benchmark-v1.0.1 |
loc-unused-fix | loc | easy | base-47 | L8 | Unused file already holds the “fix”; it is never imported | The service at `/app/repo` should bind `0.0.0.0` from config. It currently binds `127.0.0.1`.
Identify every file that must be changed to resolve this issue. Write each path relative to `/app/repo` using forward slashes, one path per line, to `/app/answer.txt`. Order does not matter. Do not list test files unless they... | The service at `/app/repo` should bind `0.0.0.0` from config. It currently binds `127.0.0.1`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/loc-unused-fix | 0.1.1 | benchmark-v1.0.1 |
edit-slugify | edit | medium | base-47 | E1 | Text normalize (case, punctuation, hyphens) | In `/app/repo/slug.py`, implement `slugify(text)`.
Rules: lowercase; replace spaces with hyphens; drop characters that are not letters, digits, or hyphens; collapse repeated hyphens; strip leading and trailing hyphens.
Do not change tests. You may run `python3 -m pytest /app/repo/tests -q`. You have 180 seconds. | In `/app/repo/slug.py`, implement `slugify(text)`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-slugify | 0.1.1 | benchmark-v1.0.1 |
edit-covered-length | edit | easy | base-47 | E2 | Half-open interval union length | In `/app/repo/span.py`, implement `covered(segments)`.
Each segment is a half-open interval `[start, end)` as a pair of numbers. Return the total length of the union of all segments. Empty and reversed segments (`end <= start`) contribute 0. Adjacent segments such as `[0, 1)` and `[1, 2)` cover length 2. Overlaps are ... | In `/app/repo/span.py`, implement `covered(segments)`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-covered-length | 0.1.1 | benchmark-v1.0.1 |
edit-deep-merge | edit | easy | base-47 | E3 | Nested merge, list replace, no mutate | In `/app/repo/merge.py`, implement `deep_merge(left, right)`.
Return a new mapping. Nested dictionaries merge recursively. For any key whose value is not a dictionary on either side, the value from `right` replaces the value from `left` (lists are replaced, not concatenated). Do not modify `left` or `right`.
Do not c... | In `/app/repo/merge.py`, implement `deep_merge(left, right)`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-deep-merge | 0.1.1 | benchmark-v1.0.1 |
edit-int-list | edit | medium | base-47 | E4 | Parse comma ints; whitespace and empty items | In `/app/repo/parse.py`, implement `parse_int_list(text)` so it splits on commas, ignores surrounding whitespace, skips empty items, and returns integers.
Do not change tests. You may run `python3 -m pytest /app/repo/tests -q`. You have 180 seconds. | In `/app/repo/parse.py`, implement `parse_int_list(text)` so it splits on commas, ignores surrounding whitespace, skips empty items, and returns integers. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-int-list | 0.1.1 | benchmark-v1.0.1 |
edit-top-k | edit | medium | base-47 | E5 | Stable top-k by score | In `/app/repo/rank.py`, implement `top_k(rows, k)` where each row is `{"name": str, "score": number}`.
Return the `k` rows with the highest scores. Ties keep the original order. If `k` is larger than the list, return all rows in that order.
Do not change tests. You may run `python3 -m pytest /app/repo/tests -q`. You ... | In `/app/repo/rank.py`, implement `top_k(rows, k)` where each row is `{"name": str, "score": number}`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-top-k | 0.1.1 | benchmark-v1.0.1 |
edit-jsonl-keep | edit | medium | base-47 | E6 | Filter JSONL; skip bad lines | In `/app/repo/filter_jsonl.py`, implement `keep_ok(lines)`: `lines` is a list of JSON object strings. Return the objects whose `status` field is exactly `ok`, in the original order. Skip blank lines. Invalid JSON lines must be skipped, not crash.
Do not change tests. You may run `python3 -m pytest /app/repo/tests -q`.... | In `/app/repo/filter_jsonl.py`, implement `keep_ok(lines)`: `lines` is a list of JSON object strings. Return the objects whose `status` field is exactly `ok`, in the original order. Skip blank lines. Invalid JSON lines must be skipped, not crash. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-jsonl-keep | 0.1.1 | benchmark-v1.0.1 |
edit-hhmmss | edit | medium | base-47 | E7 | Duration format with zero padding | In `/app/repo/timefmt.py`, implement `format_hms(seconds)` for a non-negative integer number of seconds.
Return `H:MM:SS` where `MM` and `SS` are two digits. Hours may be larger than 24.
Do not change tests. You may run `python3 -m pytest /app/repo/tests -q`. You have 180 seconds. | In `/app/repo/timefmt.py`, implement `format_hms(seconds)` for a non-negative integer number of seconds. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-hhmmss | 0.1.1 | benchmark-v1.0.1 |
edit-prefix-sums | edit | easy | base-47 | E8 | Inclusive prefix sums; no mutate | In `/app/repo/scan.py`, implement `prefix_sums(nums)`.
Return a new list where index `i` is the sum of `nums[0]` through `nums[i]` inclusive. An empty input yields an empty list. Do not modify `nums`.
Do not change tests. You may run `python3 -m pytest /app/repo/tests -q`. You have 180 seconds. | In `/app/repo/scan.py`, implement `prefix_sums(nums)`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-prefix-sums | 0.1.1 | benchmark-v1.0.1 |
edit-clip | edit | medium | base-47 | E9 | Clamp both bounds | In `/app/repo/clip.py`, implement `clip(value, lo, hi)`.
Return `lo` if `value` is below `lo`, `hi` if it is above `hi`, otherwise `value`.
Do not change tests. You may run `python3 -m pytest /app/repo/tests -q`. You have 180 seconds. | In `/app/repo/clip.py`, implement `clip(value, lo, hi)`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-clip | 0.1.1 | benchmark-v1.0.1 |
edit-timeout-zero | edit | easy | base-47 | E10 | Explicit `0` is not “missing” (`or` default) | In `/app/repo/limits.py`, implement `get_timeout(cfg)`.
If `timeout` is present, return that value (including `0`). If it is omitted, return `30`.
Do not change tests. You may run `python3 -m pytest /app/repo/tests -q`. You have 180 seconds. | In `/app/repo/limits.py`, implement `get_timeout(cfg)`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-timeout-zero | 0.1.1 | benchmark-v1.0.1 |
edit-unique-keep | edit | medium | base-47 | E11 | Unique while keeping first-seen order | In `/app/repo/unique.py`, implement `unique(items)`.
Return a new list of first occurrences, in original order. Do not modify `items`.
Do not change tests. You may run `python3 -m pytest /app/repo/tests -q`. You have 180 seconds. | In `/app/repo/unique.py`, implement `unique(items)`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-unique-keep | 0.1.1 | benchmark-v1.0.1 |
edit-pad-left | edit | medium | base-47 | E12 | No-op when already long enough (pad) | In `/app/repo/text.py`, implement `pad_left(text, width, fill=" ")`.
Return `text` left-padded with `fill` until its length is `width`. If `text` is already at least `width` characters, return it unchanged. `fill` is a single character.
Do not change tests. You may run `python3 -m pytest /app/repo/tests -q`. You have... | In `/app/repo/text.py`, implement `pad_left(text, width, fill=" ")`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-pad-left | 0.1.1 | benchmark-v1.0.1 |
testgen-clip | testgen | medium | base-47 | T1 | Missing lo / hi / identity | `/app/repo/clip.py` implements `clip(value, lo, hi)`, which clamps `value` into the closed interval `[lo, hi]`.
Write unit tests for this function in `/app/repo/tests/test_clip.py`. Do not change `clip.py`. You have 180 seconds. | `/app/repo/clip.py` implements `clip(value, lo, hi)`, which clamps `value` into the closed interval `[lo, hi]`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/testgen-clip | 0.1.1 | benchmark-v1.0.1 |
testgen-unique-order | testgen | medium | base-47 | T2 | `set()` drops order; last-wins | `/app/repo/unique.py` implements `unique_keep_order(items)`: drop later duplicates and keep the first-seen order.
Write unit tests in `/app/repo/tests/test_unique.py`. Do not change `unique.py`. You have 240 seconds. | `/app/repo/unique.py` implements `unique_keep_order(items)`: drop later duplicates and keep the first-seen order. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/testgen-unique-order | 0.1.1 | benchmark-v1.0.1 |
testgen-gregorian | testgen | medium | base-47 | T3 | Leap year is not `year % 4` | `/app/repo/dates.py` implements `valid_date(year, month, day)` for the Gregorian calendar, including century leap-year rules.
Write unit tests in `/app/repo/tests/test_dates.py`. Do not change `dates.py`. You have 300 seconds. | `/app/repo/dates.py` implements `valid_date(year, month, day)` for the Gregorian calendar, including century leap-year rules. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/testgen-gregorian | 0.1.1 | benchmark-v1.0.1 |
testgen-mean | testgen | medium | base-47 | T4 | Sum or first element instead of mean | `/app/repo/stats.py` implements `mean(nums)` as the arithmetic mean of a non-empty list of numbers.
Write unit tests for this function in `/app/repo/tests/test_mean.py`. Do not change `stats.py`. You have 180 seconds. | `/app/repo/stats.py` implements `mean(nums)` as the arithmetic mean of a non-empty list of numbers. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/testgen-mean | 0.1.1 | benchmark-v1.0.1 |
testgen-parse | testgen | medium | base-47 | T5 | No strip / keep empty tokens | `/app/repo/parse.py` implements `parse_int_list(text)`: split on commas, strip whitespace, skip empty items, return integers.
Write unit tests in `/app/repo/tests/test_parse.py`. Do not change `parse.py`. You have 180 seconds. | `/app/repo/parse.py` implements `parse_int_list(text)`: split on commas, strip whitespace, skip empty items, return integers. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/testgen-parse | 0.1.1 | benchmark-v1.0.1 |
testgen-anagram | testgen | hard | base-47 | T6 | Case and spaces | `/app/repo/anagram.py` implements `is_anagram(a, b)`: ignore spaces and letter case, then check whether the remaining characters are the same with the same counts.
Write unit tests in `/app/repo/tests/test_anagram.py`. Do not change `anagram.py`. You have 180 seconds. | `/app/repo/anagram.py` implements `is_anagram(a, b)`: ignore spaces and letter case, then check whether the remaining characters are the same with the same counts. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/testgen-anagram | 0.1.1 | benchmark-v1.0.1 |
testgen-timeout-zero | testgen | easy | base-47 | T7 | `or` default drops explicit `0` | `/app/repo/limits.py` implements `get_timeout(cfg)`: return `cfg["timeout"]` when present, including `0`; return `30` when omitted.
Write unit tests in `/app/repo/tests/test_limits.py`. Do not change `limits.py`. You have 180 seconds. | `/app/repo/limits.py` implements `get_timeout(cfg)`: return `cfg["timeout"]` when present, including `0`; return `30` when omitted. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/testgen-timeout-zero | 0.1.1 | benchmark-v1.0.1 |
testgen-greet-none | testgen | medium | base-47 | T8 | Missing key / `None` vs present name | `/app/repo/greet.py` implements `label(user)`: return `Hello, <name>`, and use `guest` when `name` is missing or `None`.
Write unit tests in `/app/repo/tests/test_greet.py`. Do not change `greet.py`. You have 180 seconds. | `/app/repo/greet.py` implements `label(user)`: return `Hello, <name>`, and use `guest` when `name` is missing or `None`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/testgen-greet-none | 0.1.1 | benchmark-v1.0.1 |
testgen-cents | testgen | easy | base-47 | T9 | `float` cents | `/app/repo/cents.py` implements `to_cents(amount)`: convert a dollar string to an integer number of cents exactly (`'0.29'` is `29`).
Write unit tests in `/app/repo/tests/test_cents.py`. Do not change `cents.py`. You have 180 seconds. | `/app/repo/cents.py` implements `to_cents(amount)`: convert a dollar string to an integer number of cents exactly (`'0.29'` is `29`). | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/testgen-cents | 0.1.1 | benchmark-v1.0.1 |
testgen-window | testgen | medium | base-47 | T10 | Slice `end+1` / `start+1` | `/app/repo/window.py` implements `take(items, start, end)`: the half-open slice `items[start:end]`.
Write unit tests in `/app/repo/tests/test_window.py`. Do not change `window.py`. You have 180 seconds. | `/app/repo/window.py` implements `take(items, start, end)`: the half-open slice `items[start:end]`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/testgen-window | 0.1.1 | benchmark-v1.0.1 |
repro-off-by-one | repro | medium | base-47 | R1 | Slice `end+1` | In `/app/repo`, `window.take(items, start, end)` is supposed to return the half-open slice `items[start:end]`. Calling `take([10, 20, 30, 40], 1, 3)` does not return `[20, 30]`.
Write a minimal Python script to `/app/repro.py` that reproduces this failure: it must exit with a non-zero status on the current repository,... | In `/app/repo`, `window.take(items, start, end)` is supposed to return the half-open slice `items[start:end]`. Calling `take([10, 20, 30, 40], 1, 3)` does not return `[20, 30]`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/repro-off-by-one | 0.1.1 | benchmark-v1.0.1 |
repro-end-exclusive | repro | medium | base-47 | R2 | Inclusive end; interior point does not fail | In `/app/repo`, `window.contains(start, end, point)` should treat `[start, end)` as half-open: `point == end` is outside.
`contains(0, 10, 10)` is currently treated as inside.
Write a minimal Python script to `/app/repro.py` that reproduces this failure: it must exit with a non-zero status on the current repository, ... | In `/app/repo`, `window.contains(start, end, point)` should treat `[start, end)` as half-open: `point == end` is outside. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/repro-end-exclusive | 0.1.1 | benchmark-v1.0.1 |
repro-zero-timeout | repro | medium | base-47 | R3 | `or 30` drops present `0` | In `/app/repo`, `limits.get_timeout(cfg)` reads a timeout in seconds from a dict. If the `timeout` key is omitted, the result must be `30`. An explicit timeout of `0` is valid and must be preserved.
Write a minimal Python script to `/app/repro.py` that reproduces this failure: it must exit with a non-zero status on th... | In `/app/repo`, `limits.get_timeout(cfg)` reads a timeout in seconds from a dict. If the `timeout` key is omitted, the result must be `30`. An explicit timeout of `0` is valid and must be preserved. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/repro-zero-timeout | 0.1.1 | benchmark-v1.0.1 |
repro-keep-zero | repro | medium | base-47 | R4 | `if n` drops `0` | In `/app/repo`, `nums.parse_ints(text)` splits on commas and returns integers. `parse_ints("0,1,2")` is supposed to be `[0, 1, 2]` but currently is not.
Write a minimal Python script to `/app/repro.py` that reproduces this failure: it must exit with a non-zero status on the current repository, and exit 0 after the def... | In `/app/repo`, `nums.parse_ints(text)` splits on commas and returns integers. `parse_ints("0,1,2")` is supposed to be `[0, 1, 2]` but currently is not. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/repro-keep-zero | 0.1.1 | benchmark-v1.0.1 |
repro-none-name | repro | medium | base-47 | R5 | Missing / `None` name | In `/app/repo`, `greet.label(user)` should return `Hello, <name>`, and use `guest` when `name` is missing or `None`. `label({})` currently does not do that.
Write a minimal Python script to `/app/repro.py` that reproduces this failure: it must exit with a non-zero status on the current repository, and exit 0 after the... | In `/app/repo`, `greet.label(user)` should return `Hello, <name>`, and use `guest` when `name` is missing or `None`. `label({})` currently does not do that. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/repro-none-name | 0.1.1 | benchmark-v1.0.1 |
repro-float-cents | repro | medium | base-47 | R6 | `float` cents | In `/app/repo`, `cents.to_cents(amount)` should convert a dollar string to an integer number of cents. `to_cents('0.29')` is supposed to be `29`.
Write a minimal Python script to `/app/repro.py` that reproduces this failure: it must exit with a non-zero status on the current repository, and exit 0 after the defect is ... | In `/app/repo`, `cents.to_cents(amount)` should convert a dollar string to an integer number of cents. `to_cents('0.29')` is supposed to be `29`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/repro-float-cents | 0.1.1 | benchmark-v1.0.1 |
repro-first-index | repro | hard | base-47 | R7 | Last-wins; needs a repeated needle | In `/app/repo`, `search.index_of(items, needle)` should return the first index of `needle`. `index_of(["a", "b", "a"], "a")` is supposed to be `0` but currently is not.
Write a minimal Python script to `/app/repro.py` that reproduces this failure: it must exit with a non-zero status on the current repository, and exit... | In `/app/repo`, `search.index_of(items, needle)` should return the first index of `needle`. `index_of(["a", "b", "a"], "a")` is supposed to be `0` but currently is not. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/repro-first-index | 0.1.1 | benchmark-v1.0.1 |
repro-empty-mean | repro | medium | base-47 | R8 | Empty input must error (returning `0` hides it) | In `/app/repo`, `stats.mean(nums)` returns the arithmetic mean of a non-empty list. `mean([])` currently returns `0` instead of raising `ValueError`.
Write a minimal Python script to `/app/repro.py` that reproduces this failure: it must exit with a non-zero status on the current repository, and exit 0 after the defect... | In `/app/repo`, `stats.mean(nums)` returns the arithmetic mean of a non-empty list. `mean([])` currently returns `0` instead of raising `ValueError`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/repro-empty-mean | 0.1.1 | benchmark-v1.0.1 |
repro-whitespace | repro | hard | base-47 | R9 | Surrounding whitespace; `isdigit()` without strip drops padded tokens | In `/app/repo`, `nums.parse_ints(text)` splits on commas and returns integers. `parse_ints(" 1, 2")` is supposed to be `[1, 2]` but currently is not.
Write a minimal Python script to `/app/repro.py` that reproduces this failure: it must exit with a non-zero status on the current repository, and exit 0 after the defect... | In `/app/repo`, `nums.parse_ints(text)` splits on commas and returns integers. `parse_ints(" 1, 2")` is supposed to be `[1, 2]` but currently is not. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/repro-whitespace | 0.1.1 | benchmark-v1.0.1 |
repro-truthy-flag | repro | medium | base-47 | R10 | Omitted key vs explicit `0` (flag) | In `/app/repo`, `flags.enabled(cfg)` returns whether the feature is on. If `on` is omitted, the result is True. If `on` is present, return its boolean value (`0` is False). `enabled({})` currently errors.
Write a minimal Python script to `/app/repro.py` that reproduces this failure: it must exit with a non-zero status... | In `/app/repo`, `flags.enabled(cfg)` returns whether the feature is on. If `on` is omitted, the result is True. If `on` is present, return its boolean value (`0` is False). `enabled({})` currently errors. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/repro-truthy-flag | 0.1.1 | benchmark-v1.0.1 |
review-clip-incomplete | review | medium | base-47 | V1 | Partial clamp (lo only); example fails | Issue:
`clip(value, lo, hi)` must clamp `value` into `[lo, hi]`.
The candidate change is already applied in `/app/repo`. Decide whether this repository fully addresses the issue.
Write a single character to `/app/answer.txt`: `1` if it fully addresses the issue, `0` otherwise. You may inspect and run code. Do not cha... | Issue: `clip(value, lo, hi)` must clamp `value` into `[lo, hi]`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/review-clip-incomplete | 0.1.1 | benchmark-v1.0.1 |
review-slug-almost | review | easy | base-47 | V2 | Near-miss slug (collapse/strip) | Issue:
`slugify(text)` must lowercase, replace spaces with hyphens, drop other punctuation, collapse repeated hyphens, and strip leading and trailing hyphens. `slugify('Hello, World!!')` must be `hello-world`. `slugify(' ABC ')` must be `abc`.
The candidate change is already applied in `/app/repo`. Decide whether t... | Issue: `slugify(text)` must lowercase, replace spaces with hyphens, drop other punctuation, collapse repeated hyphens, and strip leading and trailing hyphens. `slugify('Hello, World!!')` must be `hello-world`. `slugify(' ABC ')` must be `abc`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/review-slug-almost | 0.1.1 | benchmark-v1.0.1 |
review-mean-wrong | review | medium | base-47 | V3 | Sum instead of mean; example fails | Issue:
`mean(nums)` must return the arithmetic mean of a non-empty list. `mean([1, 2, 3])` must be `2`.
The candidate change is already applied in `/app/repo`. Decide whether this repository fully addresses the issue.
Write a single character to `/app/answer.txt`: `1` if it fully addresses the issue, `0` otherwise. Y... | Issue: `mean(nums)` must return the arithmetic mean of a non-empty list. `mean([1, 2, 3])` must be `2`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/review-mean-wrong | 0.1.1 | benchmark-v1.0.1 |
review-slug-complete | review | easy | base-47 | V4 | Complete slug; label 1 | Issue:
`slugify(text)` must lowercase, replace spaces with hyphens, drop other punctuation, collapse repeated hyphens, and strip leading/trailing hyphens. `slugify('Hello, World!!')` must be `hello-world`.
The candidate change is already applied in `/app/repo`. Decide whether this repository fully addresses the issue... | Issue: `slugify(text)` must lowercase, replace spaces with hyphens, drop other punctuation, collapse repeated hyphens, and strip leading/trailing hyphens. `slugify('Hello, World!!')` must be `hello-world`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/review-slug-complete | 0.1.1 | benchmark-v1.0.1 |
review-configured-timeout | review | medium | base-47 | V5 | `or` default drops `0`; contract is explicit | Issue:
`get_timeout(cfg)` reads a timeout from a dict. If the `timeout` key is omitted, the result must be `30`. `get_timeout({})` must be `30`. Any value that is present in the dict must be returned unchanged.
The candidate change is already applied in `/app/repo`. Decide whether this repository fully addresses the i... | Issue: `get_timeout(cfg)` reads a timeout from a dict. If the `timeout` key is omitted, the result must be `30`. `get_timeout({})` must be `30`. Any value that is present in the dict must be returned unchanged. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/review-configured-timeout | 0.1.1 | benchmark-v1.0.1 |
review-rotate-right | review | easy | base-47 | V6 | Rotates the wrong way | Issue:
`rotate_left(items)` must return a new list rotated one position left. `rotate_left([1, 2, 3])` must be `[2, 3, 1]`.
The candidate change is already applied in `/app/repo`. Decide whether this repository fully addresses the issue.
Write a single character to `/app/answer.txt`: `1` if it fully addresses the iss... | Issue: `rotate_left(items)` must return a new list rotated one position left. `rotate_left([1, 2, 3])` must be `[2, 3, 1]`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/review-rotate-right | 0.1.1 | benchmark-v1.0.1 |
review-prefix-complete | review | medium | base-47 | V7 | Complete prefix sums; label 1 (second domain) | Issue:
`prefix_sums(nums)` must return inclusive prefix sums without modifying `nums`. `prefix_sums([1, 2, 3])` must be `[1, 3, 6]`.
The candidate change is already applied in `/app/repo`. Decide whether this repository fully addresses the issue.
Write a single character to `/app/answer.txt`: `1` if it fully addresse... | Issue: `prefix_sums(nums)` must return inclusive prefix sums without modifying `nums`. `prefix_sums([1, 2, 3])` must be `[1, 3, 6]`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/review-prefix-complete | 0.1.1 | benchmark-v1.0.1 |
loc-vendor-shadow | loc | uncalibrated | hard-release-15 | LR1 | `sys.path` vendor shadow; unused top-level module looks correct | The shop project is at `/app/repo`. Uploads up to 25 MB should be accepted. A 20 MB file is still rejected.
Identify every file that must be changed to resolve this issue. Write each path relative to `/app/repo` using forward slashes, one path per line, to `/app/answer.txt`. Order does not matter. Do not list test fil... | The shop project is at `/app/repo`. Uploads up to 25 MB should be accepted. A 20 MB file is still rejected. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/loc-vendor-shadow | 0.1.1 | benchmark-v1.0.1 |
loc-env-wrapper | loc | uncalibrated | hard-release-15 | LR2 | Shell wrapper exports the live value; Python default / yaml are decoys | The shop project is at `/app/repo`. The service should write logs to `/var/log/app.log`. It still writes them somewhere else.
Identify every file that must be changed to resolve this issue. Write each path relative to `/app/repo` using forward slashes, one path per line, to `/app/answer.txt`. Order does not matter. Do... | The shop project is at `/app/repo`. The service should write logs to `/var/log/app.log`. It still writes them somewhere else. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/loc-env-wrapper | 0.1.1 | benchmark-v1.0.1 |
loc-hook-plugin | loc | hard | hard-release-15 | LR3 | Plugin path in `hooks.json`; unused sibling plugin looks correct | The shop project is at `/app/repo`. Failed jobs should follow the retry policy in `docs/ops.md`. They currently retry fewer times than that policy.
Identify every file that must be changed to resolve this issue. Write each path relative to `/app/repo` using forward slashes, one path per line, to `/app/answer.txt`. Ord... | The shop project is at `/app/repo`. Failed jobs should follow the retry policy in `docs/ops.md`. They currently retry fewer times than that policy. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/loc-hook-plugin | 0.1.1 | benchmark-v1.0.1 |
edit-config-beside | edit | uncalibrated | hard-release-15 | ER1 | Config open is cwd-relative; file next to the module is the real one | In `/app/repo/settings.py`, implement `load()`.
It must read the JSON settings file that ships with this module and return that
object. Callers need at least the `retry` and `mode` fields.
Do not change tests. You may run `python3 -m pytest /app/repo/tests -q`. You have 180 seconds. | In `/app/repo/settings.py`, implement `load()`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-config-beside | 0.1.1 | benchmark-v1.0.1 |
edit-retry-discount | edit | easy | hard-release-15 | ER2 | Retried checkout stacks the member discount | In `/app/repo/checkout.py`, implement `charge(cart)`.
Member checkout is 10% off the list price. Charging the same cart again must
still be 10% off that original list, not a smaller amount.
Do not change tests. You may run `python3 -m pytest /app/repo/tests -q`. You have 180 seconds. | In `/app/repo/checkout.py`, implement `charge(cart)`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-retry-discount | 0.1.1 | benchmark-v1.0.1 |
edit-blank-name | edit | medium | hard-release-15 | ER3 | Missing name → guest; present empty string is not a guest | In `/app/repo/profile.py`, implement `display_name(user)`.
If the user has no name field, return `guest`. If they set a name, return that
name even when it is an empty string.
Do not change tests. You may run `python3 -m pytest /app/repo/tests -q`. You have 180 seconds. | In `/app/repo/profile.py`, implement `display_name(user)`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/edit-blank-name | 0.1.1 | benchmark-v1.0.1 |
testgen-tie-order | testgen | medium | hard-release-15 | TR1 | Rank by score; ties keep given order | `/app/repo/rank.py` implements `ranked(rows)`.
Each row is `{"name": str, "score": int}`. Return names ordered by score from
high to low. The rest of the contract is documented on `ranked`.
Write unit tests in `/app/repo/tests/test_rank.py`. Do not change `rank.py`. You have 180 seconds. | `/app/repo/rank.py` implements `ranked(rows)`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/testgen-tie-order | 0.1.1 | benchmark-v1.0.1 |
testgen-booking-touch | testgen | medium | hard-release-15 | TR2 | Adjacent bookings do not overlap | `/app/repo/booking.py` implements `overlap(a_start, a_end, b_start, b_end)`.
Times are minutes from midnight. Two bookings overlap if they share an instant.
See the docstring on `overlap` for endpoint rules.
Write unit tests in `/app/repo/tests/test_booking.py`. Do not change `booking.py`. You have 180 seconds. | `/app/repo/booking.py` implements `overlap(a_start, a_end, b_start, b_end)`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/testgen-booking-touch | 0.1.1 | benchmark-v1.0.1 |
testgen-zero-qty | testgen | medium | hard-release-15 | TR3 | Quantity 0 is valid; missing / negative error | `/app/repo/order.py` implements `line_total(qty, price)`.
`qty` is an int or `None`. Quantity must be a non-negative integer when present.
Missing quantity is an error. Line total is quantity times unit price.
Details are on `line_total`.
Write unit tests in `/app/repo/tests/test_order.py`. Do not change `order.py`. ... | `/app/repo/order.py` implements `line_total(qty, price)`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/testgen-zero-qty | 0.1.1 | benchmark-v1.0.1 |
repro-second-export | repro | medium | hard-release-15 | RR1 | Second `export_rows()` is empty (consumed iterator) | In `/app/repo`, `export_rows()` should return the full table every time it is called in a process. A second export currently comes back empty.
Write a minimal Python script to `/app/repro.py` that reproduces this failure: it must exit with a non-zero status on the current repository, and exit 0 after the defect is fix... | In `/app/repo`, `export_rows()` should return the full table every time it is called in a process. A second export currently comes back empty. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/repro-second-export | 0.1.1 | benchmark-v1.0.1 |
repro-nested-alias | repro | hard | hard-release-15 | RR2 | Shared nested customer dict aliases two orders | In `/app/repo`, attaching the same customer record to two orders must keep the orders independent. Renaming the customer on one order currently changes the other order too.
Write a minimal Python script to `/app/repro.py` that reproduces this failure: it must exit with a non-zero status on the current repository, and ... | In `/app/repo`, attaching the same customer record to two orders must keep the orders independent. Renaming the customer on one order currently changes the other order too. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/repro-nested-alias | 0.1.1 | benchmark-v1.0.1 |
repro-stale-quote | repro | medium | hard-release-15 | RR3 | Quote ignores in-process price updates | In `/app/repo`, `quote(sku)` should follow `set_price`. After a price change in the same process, quotes still show the old amount.
Write a minimal Python script to `/app/repro.py` that reproduces this failure: it must exit with a non-zero status on the current repository, and exit 0 after the defect is fixed. Do not ... | In `/app/repo`, `quote(sku)` should follow `set_price`. After a price change in the same process, quotes still show the old amount. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/repro-stale-quote | 0.1.1 | benchmark-v1.0.1 |
review-bare-except | review | easy | hard-release-15 | VR1 | Example passes; bare except swallows invalid input | Issue:
`parse_count(text)` must return the integer count. Invalid text must raise `ValueError`. Example: `parse_count("3")` equals `3`.
The candidate change is already applied in `/app/repo`. Decide whether this repository fully addresses the issue.
Write a single character to `/app/answer.txt`: `1` if it fully addre... | Issue: `parse_count(text)` must return the integer count. Invalid text must raise `ValueError`. Example: `parse_count("3")` equals `3`. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/review-bare-except | 0.1.1 | benchmark-v1.0.1 |
review-dead-helper | review | medium | hard-release-15 | VR2 | Helper example passes; live checkout ignores the helper | Issue:
Member checkout should charge 90% of list price. Example: a $100 member checkout is $90.
The candidate change is already applied in `/app/repo`. Decide whether this repository fully addresses the issue.
Write a single character to `/app/answer.txt`: `1` if it fully addresses the issue, `0` otherwise. You may i... | Issue: Member checkout should charge 90% of list price. Example: a $100 member checkout is $90. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/review-dead-helper | 0.1.1 | benchmark-v1.0.1 |
review-wired-helper | review | medium | hard-release-15 | VR3 | Live checkout calls the helper (label 1) | Issue:
Member checkout should charge 90% of list price. Example: a $100 member checkout is $90.
The candidate change is already applied in `/app/repo`. Decide whether this repository fully addresses the issue.
Write a single character to `/app/answer.txt`: `1` if it fully addresses the issue, `0` otherwise. You may i... | Issue: Member checkout should charge 90% of list price. Example: a $100 member checkout is $90. | Apache-2.0 | caojiajun777/small-ow-agent-bench | benchmark-v1.0.1 | https://github.com/caojiajun777/small-ow-agent-bench/tree/benchmark-v1.0.1/tasks/review-wired-helper | 0.1.1 | benchmark-v1.0.1 |
small-ow-agent-bench (catalog)
Frozen compact-shell system reliability for compact open-weight coding agents — not coding IQ.
- GitHub: caojiajun777/small-ow-agent-bench
- Tag:
benchmark-v1.0.1 - This dataset:
junjun77/small-ow-agent-bench
This dataset is not the Harbor task dump. Each of the 62 rows is catalog metadata plus the agent-visible instruction:
id,skill,difficulty,banktrap_id/trap(failure-mode label fromTRAPS.md)instruction/instruction_summary- license, GitHub tag, compact-shell version
It does not include hidden verifiers: tests/, solution/, foils/, gold file lists, mutants, or environment/repo. Those stay in Harbor tasks/ on GitHub.
Do not train on this 62-item bank and then report the same items as evaluation.
Compact-10 (v1.0.1)
Headline = five-skill macro mean. Micro = successes / 186. Artifact does not require finish; Clean does. Halt (Artifact=1, not clean) = 105. 12×62×3 = 2232 scored trials.
| # | Model | Artifact macro | Clean macro | Gap | Artifact micro | Clean micro |
|---|---|---|---|---|---|---|
| 1 | Qwen3.5-9B | 0.786 | 0.757 | 0.029 | 148/186 | 142/186 |
| 2 | Ministral-14B | 0.497 | 0.488 | 0.009 | 93/186 | 91/186 |
| 3 | Ministral-8B | 0.456 | 0.411 | 0.045 | 87/186 | 78/186 |
| 4 | Qwen3-14B | 0.404 | 0.333 | 0.072 | 76/186 | 65/186 |
| 5 | Gemma-3-12B | 0.313 | 0.114 | 0.199 | 60/186 | 18/186 |
| 6 | Granite-4.1-8B | 0.162 | 0.140 | 0.022 | 31/186 | 26/186 |
| 7 | Gemma-3-4B | 0.067 | 0.067 | 0.000 | 11/186 | 11/186 |
| 8 | Ministral-3B | 0.049 | 0.019 | 0.030 | 10/186 | 4/186 |
| 9 | Qwen3-8B | 0.034 | 0.018 | 0.015 | 6/186 | 3/186 |
| 10 | Llama-3.2-3B | 0.027 | 0.000 | 0.027 | 4/186 | 0/186 |
Upper-reference (not ranked with Compact-10)
| # | Model | Artifact macro | Clean macro | Gap | Artifact micro | Clean micro |
|---|---|---|---|---|---|---|
| — | Qwen3.8-27B | 0.863 | 0.845 | 0.018 | 162/186 | 159/186 |
| — | Qwen3.6-35B-A3B | 0.632 | 0.560 | 0.072 | 118/186 | 104/186 |
Full five-skill tables: results/leaderboard.md.
Regenerate:
python scripts/export_hf_catalog.py --write
- Downloads last month
- 33