Genshin Impact
Data checked ·
Model version · onebst-conservative-hard-proof-2026-10-11
Results estimate a plan under your targets, resources, deadlines and model assumptions. They do not guarantee actual pulls. Follow in-game rules and official announcements.
Published base rates, hard-pity limits and rate-up rules are rule inputs. A complete per-pull soft-pity table is generally not a published rule. Genshin and Star Rail estimates combine published rules with statistical soft-pity models; statistical references are not official confirmation. The Genshin weapon model uses a 77-pull cap as a statistical modeling choice, rather than replacing the official description.
ZZZ uses verified base rates, hard pity and rate-up guarantees in a simplified model, without a complete soft-pity curve. A published consolidated probability is not a fixed probability for every pull. This approximation is not the complete official mechanism.
Pity is separate across banner types. Carry-over, guarantees and special banner states follow each game configuration. Genshin Capturing Radiance depends on the toggle and initial counter; weapon Fate Points do not carry into the next phase. Check your own game history.
Preview targets with unannounced rank, phase or dates may use explicitly labeled planning assumptions. Being selectable does not mean that the rank, lineup or deadline is officially confirmed.
Targets share one budget and are attempted in order. Each target does not receive the same balance again. Stage results describe progress through that order; average pull cost is not a guaranteed cost.
Your game server determines deadlines, daily resets and claimable income days. Display time zones only convert specific times. Extra income is credited under your selected amounts and claim dates; put already claimed rewards in your balance to avoid double counting. Projected dates and future rewards remain assumptions.
Added top-ups use the selected region, quantity and configured regular rewards, excluding first-purchase doubling. This is a scenario comparison, not a purchase recommendation. Refund exchanges enter the simulation only when enabled; do not manually count the same refund again.
Complex plans use seeded Monte Carlo simulation with actual sample counts, model versions and precision states. A run-time budget can reduce samples for long plans. Sampling intervals describe simulation uncertainty, not errors in rules, income or date assumptions.
Every sampled run succeeding does not prove certainty. Without a hard-pity proof, finite all-success samples alone do not establish a guaranteed 100%. The proof uses official 90 / 80 caps, entered guarantees and current path state to check the shared budget at every deadline. Only proven prefixes display 100%. Named 4-star / A-rank targets have no targeted hard guarantee. The conservative bound assumes accurate inputs and planned income, without relying on incidental drops or random refunds. Check the sample count, result state and assumptions together.
Check the author before the claim. A banner notice from the official site or verified game account can support the lineup and times it names. A character introduction may disclose only an identity, element or Path; it does not automatically establish rarity, phase order or a rate-up slot. Official community domains also host player posts, so the domain alone is insufficient evidence.
Announced, projected and awaiting notice describe different evidence states. Projected dates support planning scenarios, not promises. Missing four-star or A-Rank lists remain unannounced. Historical version selectors query the archive; they do not generate a separate web page for every version number. Dedicated guides need relevant records, sources and a useful purpose, while earlier records remain accessible through the history table.
The data-check date records when this site last maintained that game’s records. An announcement date belongs to the original source; the two dates are not interchangeable. Revising page wording does not mean all game data was checked again. The update log distinguishes data changes from checks with no change and describes the records actually maintained, rather than claiming a complete announcement history.
To report a discrepancy, use Contact & corrections with the game, server, version or phase, page URL, displayed value and supporting official notice. An in-game rules screenshot can help explain a rules issue. Do not send a UID, password or token. The maintainer checks the evidence and its scope before changing records and explanations; an unresolved report must not be described as a completed fix.
Saved plans retain their original inputs instead of being rewritten by a guide update. When revisiting a plan, check whether the target remains valid, its deadline changed or rewards were already collected, then decide how to revise it. This site does not publish prediction-accuracy statistics. Without a retained original prediction and the later announcement, there is no historical score to reconstruct honestly.
The links below provide game rules and statistical references. Data check dates come from current game configurations. Recorded announcements, confirmed lineups and projected content are labeled separately. Maintenance is manual, without a promise of real-time updates.
“The gacha models this toolkit can implement are limited.”
Data checked ·
Model version · onebst-conservative-hard-proof-2026-10-11
Data checked ·
Model version · hsr-onebst-hard-proof-2026-10-11
Data checked ·
Model version · zzz-official-base-hard-proof-2