Rank Hunter Documentation¶
Rank Hunter is a local research environment for searching, investigating, and certifying elliptic curves over \(\mathbf Q\). It combines candidate generation, family-specific search, rational-point discovery, exact point verification, Mordell–Weil work, rank bounds, quartics and coverings, pipelines, jobs, campaigns, external catalogs, and plugins in one application.
These docs describe how the released system actually behaves. They are organized for three audiences:
- Researchers using the UI — start with Getting Started and Using Rank Hunter.
- Researchers scripting or automating work — use the Command Line section.
- Developers extending or debugging Rank Hunter — use Plugins and Reference.
The public release documented here is Rank Hunter 0.9.2.
Start here¶
If this is your first time using Rank Hunter:
- Install Rank Hunter and verify the scientific runtime.
- Read What Counts as Proof? before interpreting ranks.
- Follow Your First Hunt from Candidate Pool to retained curve.
- Read The Interface for the page-by-page map.
- Use Troubleshooting when a job, runtime, or engine behaves unexpectedly.
If you are developing Rank Hunter or a plugin, also read:
The research workflow¶
The normal data flow is:
Family / generator
↓
Candidate Pool
↓
screening / Pipeline / Search
↓
retained Curve
↓
exact Point ledger
↓
independence / saturation / descent / coverings / lattices
↓
rigorous rank evidence
Each layer has a different role.
A Candidate is something worth spending time on. It is not automatically a retained curve and carries no rank claim by itself.
A Curve is durable scientific state: an exact Weierstrass model plus provenance and attached research results.
A Point is stored separately. Exact membership on the curve and independence are distinct questions.
A Job records work that was queued or executed. A Pipeline Run records stage-by-stage orchestration over a population.
A Campaign groups related research without changing the mathematics.
Evidence is deliberately conservative¶
Rank Hunter searches aggressively but promotes evidence conservatively.
| State | Meaning |
|---|---|
| Heuristic score | Ranking/scheduling signal only. |
| Exact point | Rational coordinates verified exactly on the stored curve. |
| Independent exact points | Exact points certified to contribute distinct Mordell–Weil directions. |
| Rigorous lower bound | Rank is proved to be at least this value. |
| Rigorous upper bound | A rigorous engine proved rank is at most this value. |
| Exact rank | Rigorous lower and upper bounds agree. |
| Timeout / error | Computation did not finish; mathematically inconclusive. |
Do not infer rank from raw point count, Nagao score, a numerical height matrix, an external leaderboard entry, or a timeout. See What Counts as Proof? and Rank & Evidence.
Where the main tasks live¶
| Goal | UI | Documentation |
|---|---|---|
| Generate/search candidates | Search, Candidates, Auto | Search, Candidate Pools |
| Build reusable workflows | Pipelines | Pipelines |
| Deep-search one curve | Target | Target Search |
| Inspect curves and points | Curves, Points | Curves & Points |
| Prove point independence | Independence | Independence |
| Search coverings/quartics | Quartics | Quartics & Coverings |
| Saturate a known subgroup | Saturation | Saturation |
| Study heights/lattices | Lattices | Lattices & Heights |
| Bound or close the rank | Descent | Descent & Rank Bounds |
| Manage long-running work | Jobs, Campaigns | Jobs & Campaigns |
| Extend Rank Hunter | Plugins | Plugins |
Operational model¶
The browser is a control surface. SQLite is the durable research record.
Closing a browser tab does not erase scientific state. Jobs may continue according to how they were launched and queued. The default database is rank42.db in the checkout unless another path is configured.
For a developer-level explanation of the UI shell, queue, plugin loader, evidence reducers, and workers, see Architecture.
For the important persisted entities and their relationships, see Data Model.