# Problem sets
`leetctl` ships eleven curated problem lists. They are plain TOML under [`data/sets/`](../data/sets),
compiled into the binary, so filtering by one needs no network call and no LeetCode Premium
subscription.
```sh
leetctl sets # what's available
leetctl list --set blind75 --stat # your progress through a set
leetctl pick --set blind75 --difficulty medium # random pick from a set
```
## What's in the box
| `blind75` | Blind 75 | 75 | [neetcode-gh/leetcode](https://github.com/neetcode-gh/leetcode) | MIT |
| `neetcode150` | NeetCode 150 | 150 | same | MIT |
| `neetcode250` | NeetCode 250 | 250 | [neetcode.io practice page](https://neetcode.io/practice/practice/neetcode250) | NeetCode, identifiers only |
| `neetcode-all` | NeetCode All | 450 | [neetcode-gh/leetcode](https://github.com/neetcode-gh/leetcode) | MIT |
| `top-interview-150` | Top Interview 150 | 150 | LeetCode study plan | LeetCode, identifiers only |
| `top-100-liked` | Top 100 Liked | 100 | LeetCode study plan | LeetCode, identifiers only |
| `leetcode-75` | LeetCode 75 | 75 | LeetCode study plan | LeetCode, identifiers only |
| `google` | Google | 488 | [hxu296/leetcode-company-wise-problems-2022](https://github.com/hxu296/leetcode-company-wise-problems-2022) | MIT |
| `facebook` | Meta / Facebook | 371 | same | MIT |
| `amazon` | Amazon | 592 | same | MIT |
| `microsoft` | Microsoft | 363 | same | MIT |
### The company lists are a 2022 snapshot — read this before trusting them
LeetCode's live company tags are a **Premium** feature and are not available to this tool. The four
company sets come instead from a community-collected snapshot of which problems carried each
company's tag in **2022**. Treat them as "problems that were being asked around 2022", not as a
current signal.
They are membership lists, not ranked frequency lists: the upstream data carries an occurrence
count, but this tool samples uniformly and sorts by problem id, so that ranking is not stored. Each
file records `source_as_of = "2022"`, and `leetctl sets --sources` prints it.
### Where the NeetCode lists come from
`blind75`, `neetcode150`, and `neetcode-all` come from NeetCode's MIT-licensed data file,
[`.problemSiteData.json`][ncdata] — 450 entries, flagged for those first two lists only.
**`neetcode250` is not in that file.** It exists only in the neetcode.io site itself, whose problem
table is compiled into an Angular bundle rather than served as JSON (`neetcode.io/api/problems`
returns the site's HTML shell). The generator therefore scrapes the content-hashed bundle name off
the practice page, fetches the bundle, and reads the problem object literals out of it:
```js
{problem:"Two Sum",pattern:"Arrays & Hashing",link:"two-sum/",...,neetcode250:!0,ncLink:"two-integer-sum/"}
```
`link` is the LeetCode slug and is what gets joined on. `ncLink` is NeetCode's own slug for the
same problem, and is deliberately ignored — it is a different naming scheme entirely
(`two-sum` vs `two-integer-sum`, `contains-duplicate` vs `duplicate-integer`).
That bundle also carries 973 problems and its own `blind75` / `neetcode150` flags. The generator
checks those against the MIT data file and **fails if the two sources disagree**, because mixing
sources is only safe while they agree; today they match exactly. It also asserts the lists still
nest (blind75 ⊆ neetcode150 ⊆ neetcode250), which `src/sets/mod.rs` re-checks at test time.
`neetcode-all` stays at the MIT file's 450 — every problem NeetCode publishes a *solution* for. The
site's own catalog is larger (973), so "All" here means "all solved", not "everything on the site".
[ncdata]: https://github.com/neetcode-gh/leetcode/blob/main/.problemSiteData.json
### Premium problems
Some sets contain LeetCode Premium problems (e.g. *Encode and Decode Strings*, *Walls and Gates*).
They stay in the sets — they are genuinely part of those lists — but show as locked and cannot be
fetched without a subscription. Counts, measured against a current problem cache:
| `blind75` | 6 |
| `neetcode150` | 7 |
| `neetcode250` | 7 |
| `neetcode-all` | 9 |
| `google` | 80 |
| `top-interview-150`, `top-100-liked`, `leetcode-75` | 0 |
Filter them out with the `L` query flag:
```sh
leetctl pick --set neetcode150 --query L
```
## File format
One file per set. Every entry carries both the frontend id and the slug; `fid` is what problems are
actually filtered on, and `slug` is there so the files stay readable and diff meaningfully when
regenerated.
```toml
# @generated by scripts/gen_sets.py — do not edit by hand.
slug = "blind75"
name = "Blind 75"
description = "The original Blind 75 list, as tracked by NeetCode."
source_url = "https://github.com/neetcode-gh/leetcode"
source_license = "MIT"
source_as_of = "2026-08-17" # vintage of the data
generated_at = "2026-08-17" # when it was last fetched
problems = [
{ fid = 217, slug = "contains-duplicate" },
{ fid = 242, slug = "valid-anagram" },
]
```
Curated lists keep their source's curriculum order. The company lists are sorted by `fid`, because
their upstream order carries a frequency ranking this tool does not use — sorting makes regenerated
diffs reviewable.
`source_as_of` is a plain date rather than a `stale` boolean: nothing has to decide when a set stops
being current, and the field cannot itself go out of date.
## Regenerating
```sh
python3 scripts/gen_sets.py
```
Standard library only, no virtualenv. It fetches every source, resolves each problem, rewrites all
eleven files, and prints a per-set count.
**Generation fails rather than shipping a short set.** If any slug does not resolve, or a set does
not come out at the size recorded in `EXPECTED_COUNTS`, the script exits non-zero and writes
nothing for that set — otherwise a "Blind 75" holding 74 problems could ship unnoticed. A genuine
upstream change means updating `EXPECTED_COUNTS` deliberately, after reviewing the diff. The same
counts are asserted again by the Rust tests in `src/sets/mod.rs`.
**Slugs are the join key, fids come from LeetCode.** Sources disagree about ids — the company CSVs
carry only a URL, and one NeetCode entry has a malformed id field (`135-candy` where every other
entry is zero-padded to four digits). So the generator pulls the authoritative slug → frontend id
index from `https://leetcode.com/api/problems/all/` and joins on that. A slug that does not resolve
fails the run with an explanation, and is never guessed at.
LeetCode has renamed seven problems since the 2022 company data was collected (`implement-strstr` →
`find-the-index-of-the-first-occurrence-in-a-string`, `friend-circles` → `number-of-provinces`, and
five others). `RENAMED_SLUGS` in the generator maps them across; each target was checked against the
live index by frontend id rather than inferred from the old name. Without that map, 13 entries drop
out of the company sets.
## Adding a set
1. Add a source and an entry to the relevant table in `scripts/gen_sets.py`.
2. Run it, and commit both the script change and the generated `data/sets/<slug>.toml`.
3. Add its size to `EXPECTED_COUNTS` in the generator, and to the matching table in
`src/sets/mod.rs`'s tests.
4. Register the file in `REGISTRY` in `src/sets/mod.rs` — this is what makes the slug valid on the
command line and in shell completions.
5. Add a row to the table above.
Sets must come from a source that permits redistribution. The four sources shipped today are
MIT-licensed or fetched from LeetCode's own public endpoints, and only problem identifiers are
stored — never problem text. Each set records its source and licence, and `LICENSE` carries the
upstream notices.