Everything in one research loop
You should not have to remember which script owns candidate generation, which tool found a point, where a descent result went, or what still needs proving. Rank Hunter organizes the work around the curve and the question you are trying to answer.
The basic idea
Rank Hunter does not hide the evidence model, but it does hide unnecessary plumbing. A researcher can move from a promising family specialization to exact points, rigorous lower bounds, descent, geometry, and reproducible history without stitching together a pile of disconnected one-off scripts.
Discover
Start broad, specialize a published family, use a prepared Pipeline, or go straight after one retained curve. Heuristics help decide where to spend time; exact arithmetic still decides what gets accepted.
Launch search work through shared Pipeline infrastructure instead of separate ad-hoc engines. Family plugins can contribute exact specialization logic and search geometry without moving that mathematics into core.
Candidate pools preserve ranking provenance, screening state, and later search handoff so promising fibers do not disappear into terminal scrollback.
Route rational-point searches through compatible engines, including CPU/GPU ratpoints, completed-square cubic models, pointed quartics, and classical covering paths.
Nagao-style scores, root-number information, numerical geometry, local patterns, and search history can prioritize work while remaining explicitly non-rigorous where appropriate.
Pipelines
Pipelines are Rank Hunter's modular strategy builder. Choose what kind of population you are working on, then assemble the exact sequence of scientific stages you want Rank Hunter to execute and preserve.
Run against a plugin-backed family or an existing family candidate pool.
Build around a prescribed rational torsion target with compatible providers.
Start from General curve populations without a family dependency.
Turn one retained curve into the starting point for a deeper modular attack.
Current core catalog
63The builder validates stage contracts and ordering. A module advertises what kind of evidence or population it can produce; runtime success is never assumed.
Save a working strategy, duplicate it, change the target or budgets, and keep the scientific plan explicit.
Pipeline Runs keep execution state, derivations, candidates, runtime drift, and reproducibility context attached to the run.
Supported strategy modules preserve completed substeps so interrupted work can continue from the first unfinished checkpoint instead of blindly starting over.
Investigate
Rank Hunter keeps arithmetic, points, search history, geometry, research context, and next-step planning attached to the same retained object.
Escalate exact point discovery around a specific curve, preserve existing rigorous evidence, and keep failed or timed-out engines classified as inconclusive rather than destructive.
Review the stored model, arithmetic summary, rigorous evidence, heuristic signals, exact-point ledger, catalog context, provenance, history, and the next useful action from one curve-centered view.
Analyze canonical-height Gram matrices, generator correlations, eigenvalue strength, LLL-reduced bases, candidate residuals, and stored lattice runs. Numerical geometry can guide Target and Independence, but it never promotes rank by itself.
Use pointed quartics, exact quartic-to-elliptic maps, PARI reduction, and resumable searches inside the same evidence workflow used elsewhere in Rank Hunter.
Open plugin-owned tools such as Curve Explorer, Fiber Atlas, Leaderboard Compare, and Twist & Isogeny Lab without weakening core evidence boundaries.
Prove
Rank Hunter separates exact points, rigorous independence, lower bounds, upper bounds, saturation, and exact rank so that a promising computation never quietly becomes a theorem.
Verify the coordinates exactly on the stored elliptic curve.
Certify enough independent points to establish a lower bound.
Use supported descent/rank-bound engines under explicit time limits.
Record exact rank only when rigorous lower and upper bounds meet.
Resolve point dependence and certify new Mordell–Weil directions before increasing a rigorous lower bound.
Run bounded rigorous rank-bound methods, inspect structured engine evidence, escalate when stronger budgets are justified, and attempt exact-rank certification only when the current evidence supports it.
Record bounded saturation evidence separately. A nontrivial saturation index changes subgroup index/basis, not rank.
Organize
Rank Hunter treats execution history, evidence, retained curves, and reproducibility metadata as part of the research instrument rather than administrative leftovers.
Group related work under one durable research context.
Compose, save, run, inspect, duplicate, and reproduce modular scientific strategies.
Own queue, dispatcher, scheduler, lifecycle, results, pause, stop, resume, and retry.
Keep generation, ranking provenance, screening, and search handoff durable.
Retain the authoritative scientific record and research state for each curve.
Persist exact points, bounds, proof state, provenance, and reproducibility metadata.
Reference
Old curve lists, exported spreadsheets, certificates, notes, historical results, and reference datasets can stay available to Rank Hunter without being silently promoted into your authoritative local evidence. Browse it, compare it, seed new work from it — then let Rank Hunter reproduce what matters.
Think of Libraries as the clean room for everything accumulated outside Rank Hunter: old certs, spreadsheets, exported curve sets, source artifacts, historical observations, aliases, mapped fields, and provenance. Original files remain external research material. You can preview, map, project, search, compare, build Candidate Pools, or explicitly validate mapped points without silently promoting external rank claims into authoritative evidence.
Keep external reference sources such as ICARM or LMFDB distinct from evidence Rank Hunter has actually reproduced in your local research state.
Extend
Specialized mathematics belongs in plugins. Core owns the shared scientific contracts so Families, Features, and Workspaces can be powerful without turning Rank Hunter into a pile of private forks.
Exact specialization logic, maps, sections, theorem metadata, torsion structure, and search adapters.
Add focused behavior inside supported core research hooks without creating a whole new page.
Create self-contained exploratory pages while preserving core-owned persistence and proof boundaries.
Browse every currently published Family, Feature, and Workspace.
The short version
Search, investigate, and certify without losing the evidence trail between them.
