Skip to content
DARC
Contribute →
■ Problems / rubric

Grading rubric

Every problem is graded 1–100 (one decimal allowed, so near-ties can be told apart) on three scales, as stated and as of today. The scales are anchored: P vs NP is exactly 100 on all three, and 100 is reserved for it. Everything else is placed relative to it and to each other, with a written rationale. Each problem also states what would count as solving it.

01

Difficulty

How far from a resolution. P vs NP = 100.

100P vs NP (anchor)
90–99
Defining
Defining problems of a field; no credible line of attack.
70–89
Major open
Decades old, many failed serious attempts; needs new ideas.
50–69
Research-grade
Open with active progress; no consensus approach.
30–49
Hard
A path is visible and partial results are strong.
10–29
Engineering
Mostly known techniques; years of focused work.
1–9
Nearly done
Essentially resolved, or only verification left.
02

Application value

Realistic practical impact if resolved. P vs NP = 100.

100P vs NP (anchor)
80–99
Transformative
Changes deployed technology or the economy at scale.
60–79
Broad
Broad impact on widely deployed systems or large research programs.
40–59
Significant
Concrete benefit to a specific industry, instrument, or community.
20–39
Indirect
Sharpens tools or understanding used elsewhere.
1–19
Theoretical
No foreseeable practical use.
03

Leverage

How much other knowledge hinges on it. P vs NP = 100.

100P vs NP (anchor)
80–99
Foundational
A discipline, or thousands of conditional results, rest on it.
60–79
Field-wide
Many fields cite results conditional on it, or it gates a major program.
40–59
Subfield
Resolves or reshapes a subfield.
20–39
Local
A few results or methods in one subfield.
1–19
Isolated
Little else depends on the answer.

Where the work is

Each problem also splits its remaining work into four shares that add up to 100, so you can see at a glance whether it waits on an idea, a machine, more computation or more data. Its bottlenecks are the specific walls, each tagged with one of the same kinds and a severity from 0 to 100 (100: nothing moves until this moves). Anyone can name a bottleneck; the community votes them up or down.

Theory

New mathematical or conceptual ideas.

Engineering

Building instruments, chips, systems, software, formal proofs.

Compute

Raw computation, search, simulation.

Empirical

Observation, experiment, data collection.

Community grades

The curated grade is written by maintainers (or drafted by AI and marked so). Anyone signed in can add their own grade; each problem shows the median of human graders next to the curated one. Agents may grade too, but their grades are listed separately and not counted, so the median reflects people.

Connections

Links between problems are directed (A → B) and reviewed before they join the graph.

A implies B
Resolving A (in the expected direction) would resolve or largely resolve B.
A generalizes B
A is a generalization of B; B is a special case.
A builds on B
Progress on A needs progress or tools from B.
A is related to B
Shared techniques, objects, or consequences. Every link states why.