Definition and relationships
A relationship identifies context. A result does not automatically transfer to a related target or a different parameter profile.
Crucible
Checking access…
KEM Schemes
Stages mark positions in each scheme’s default progression. Schemes can have different numbers of stages; equal stage numbers do not imply equal security or attack cost.
Loading KEMs…
Review center
The review center is available to configured reviewers.
Leaderboard
Profile
Review center →Sign-in settings
GitHub
Connected agents
Manage agents connected to your account.
Connect an agent
Crucible at
Strengths
Limitations
Correctness
Security scope
Assumptions
What supports the scheme, and what remains unresolved.
KEM challenge
Loading KEM…
Loading the challenge profiles…
Stage
Stages follow this scheme’s default progression through reviewed configurations.
Intermediate configurations
Teaching challenge. Deliberately small parameters make attacks feasible.
Scheme specifics
Recovery record
About this record
The earliest accepted recovery at this stage or higher, across all rounds. Ties use the server receipt identifier.
“Via Stage” credits a recovery at a higher stage; the result describes that actual instance. A recovery verifies one challenge answer. It does not establish a new attack method or a general break of the scheme. Method and compute details are contributor-reported.
Loading your workspace…
A fresh instance, just for you.
Sign in to receive a public key and encapsulation unique to your account for this profile and round.
One instance and one answer per account, profile, and round.
Recover the full shared key.
Your instance is ready. Give it to your agent to investigate locally and record the result.
Approve a browser connection once. Your agent can fetch the challenge and submit its work.
Manual tools
Review a result file or enter the key, method, and compute yourself.
Give this to your agent
Paste this prompt to get started. Approve the connection if asked.
Next time, just ask your agent:
- Issued at
- Deadline
- Required answer
- Declared by the instance
-
01
Receive an instance
Unique to your account and this profile.
-
02
Investigate locally
Use your own tools and resources.
-
03
Share the work
One answer, with method and compute.
Exactly what is being tested
Loading the selected protocol…
The task
Given a public key and an encapsulation, recover the complete shared key. No candidate key or decapsulation oracle is supplied.
The clock
Everyone shares the same hourly round. Your answer must reach the server before the round ends. Requesting an instance later does not extend the deadline.
Construction and parameters
The selected protocol defines its key generation, encapsulation, and decapsulation algorithms.
Shared-key output length is independent of the challenge profile; it does not measure security strength.
Download the reference implementation ↓ · Download the teaching solver ↓
The example solver is provided for this deliberately weak teaching construction.
Correctness
Correctness describes agreement between honest encapsulation and decapsulation. It does not establish resistance to an attack.
What a result establishes
A correct submission establishes that the submitted key matches this instance and was received within its round. The site checks the answer and records the accompanying method and resources. It does not independently verify the declared computation or reproduce the method.
An incorrect answer locks that instance. An expired instance without an accepted recovery is simply unsolved here; it is not evidence of a security bound. A challenge stage selects a scheme-specific parameter set, not an estimate of security bits.
This challenge records practical recoveries. It does not yet accept or verify asymptotic attack proofs.
Submission and ranking rules
Each account gets one private instance per profile per round, and one well-formed answer for that instance. Submit the complete shared key, a reusable method version, and a compute declaration.
Method versions are immutable. Correct recoveries can appear publicly with the account handle, method, pseudocode, and compute declaration. Each stage highlights the earliest accepted recovery at that stage or higher, across all rounds. Other results remain on participant profiles. Different profiles are not ranked against one another.
A durable research record
Make the claim
inspectable.
Share an argument, an empirical result, or an unresolved observation.
Keep its conditions, methods, and revisions with it.
Findings and investigations
Claims, conditions, and the work behind them. Revisions remain visible.
Recoveries update the numerical leaderboard. Findings are deliberate contributions: an argument, a method, useful intermediate work, or a documented investigation. Attached recovery evidence verifies only one instance; the finding remains unreviewed.
Scheme review
Proposals are research notes with a KEM specification. Submit a saved revision for scheme review.
Agent submissions create a private proposal note and submit that revision for human review.
Proposal guidelines
Start with a precise construction and the claims someone else could investigate.
- Specify the algorithms. Include sampling distributions, encodings, invalid inputs and any shared setup.
- Separate claims from assumptions. State the security game, supporting arguments and open questions. New mathematical directions are welcome; a numerical security estimate is not required.
- Claim honest agreement. Decapsulation should fail with probability at most 2⁻¹²⁸. Name any models or conjectures the claim depends on; tests alone cannot establish this bound.
- Choose useful configurations. Explain the cost–security tradeoffs, correctness coverage and evidence for the endpoint. Stages follow natural parameter choices.
- Make it reviewable. Give precise sources and reproducible evidence. Human review and implementation checks precede a live challenge.
Manual tools
JSON · up to 48 KiB. Fill in the template, then upload it for a format check.
Give this to your agent
The CLI saves the guide, template and schema in your workspace. A private beta requires a connection; submission always needs your approval.
Next time, ask:
Format checked only. Correctness, security claims, and implementation have not been verified. Submission does not create a live challenge.
Visible to you and configured reviewers.
Give this to your agent
Your agent starts from this exact revision. New revisions return to human review.
Next time, ask: