Methodology

An operator’s recommended weapon, gear set and Potential come from one of two sources, and the Build tab states which one is showing: a measured ranking imported from prydwen.gg where it exists, or this site’s own arithmetic reading of the kit’s scaling rows where it does not.

The derived reading starts by finding the stat a kit actually multiplies. Its combat skills are scanned for a percentage-of-HP clause; if one is found, the build targets HP, otherwise it targets ATK. That single stat then drives which weapon and gear set score highest (see the next two sections), so a build recommendation is never chosen by hand: it is the output of the scaling stat plus the scoring rules below.

A weapon is ranked for an operator by scoring how squarely its own skill text names what that operator’s kit needs, restricted first to the one weapon type the operator can actually equip.

The score starts from rarity and base ATK, then adds points for each signal a weapon’s skills actually state: naming the operator’s attribute or damage type, Crit Rate or an ATK boost for a damage role, Ultimate Gain Efficiency for a non-damage role, team-wide wording for a Support role, or HP where the build targets HP. The highest-scoring 6★ weapon becomes the primary pick, the next-highest an alternative, and the best-scoring option at 4★ or below the budget pick. There is deliberately no simulated damage percentage on a derived row: no public source publishes one per weapon, and inventing a number to fill that column would be the most-trusted thing on the page. Where prydwen.gg has measured the operator, its own figures take over instead and are shown as measured.

A gear set is matched the same way a weapon is: its 3-piece set effect text is scanned for the same signals, an operator’s attribute, Crit stats, ATK or Skill DMG, Ultimate Gain Efficiency, team-wide wording, or HP, and the highest-scoring set wins. A set with no stated set effect cannot score and is never recommended blind. Ties between sets that score the same way are broken by the set’s own item level, since a higher-level set is strictly better gear at the same effect.

Placements on the tier list follow Prydwen’s Arknights: Endfield tier list, captured 2026-09-20 and reused with their authorization. Prydwen ranks a character into one of three categories (Damage, Enabler, Buffer), which this site reads onto its own four columns (Main DPS, Sub DPS, Support, Sustain) by the class and kit each operator already carries on this site. Prydwen currently assumes P0 for 6-star operators and P5 for 5-star operators, with Endministrator, a 6-star, read at P5 as a named exception; the Potential badge on a tile states which applies. Only 5-star and 6-star operators are placed, which is this site’s own scope rule and narrower than Prydwen’s own board, which also ranks 4-star operators.

This is a separate reading from the kit-rubric score shown on build guides and the home page (rarity, class fit, role coverage, the Potential a recommended build assumes, scaling rows, six stated utility effects, and how cheap the Ultimate is to keep up, summed to a total and cut at S 34, A 28, B 22, thresholds set from the highest score any operator in the roster actually reaches, 40 points as of the last review). That score is this site’s own, computed from the ingested kit data; it does not feed into or decide a tier-list band, and the two can disagree.

Which teams exist, their tier and their members follow Prydwen’s Arknights: Endfield team tier list, captured 2026-09-20 and reused with their authorization; a slot’s alternates are listed in the order Prydwen lists them. Prydwen’s own team cards carry no per-slot role text and no per-team note beyond their Changelog, so the role each slot plays and a team’s one-line premise are written by hand from what the game itself states about how a squad works: what each class is for, how an Arts Reaction fires between two operators of different elements, how a Physical Status applies and spends Vulnerable stacks, and the operator tags the wiki itself records (SP Recovery, Lift, Crush, and the rest).

What a team card does not claim is rotation order, SP economy or Potential thresholds: those depend on how a specific account plays the fight, not on what Prydwen or the kit data states, so they are left to a guide rather than asserted here. A team labelled off-meta still clears; Prydwen has it at T1 or below, usually because a later release overtook the carry as the fastest way to do it.

Every rate and guarantee on the banners page is Hypergryph’s own published number, read from the in-game Rules panel, not sampled or estimated from pull data. Endfield states its whole pity curve: a 0.8% base 6★ rate that holds for the first 65 pulls, then climbs 5 percentage points per pull from the 66th until the 80th pull, which is a guaranteed 6★.

Where a banner’s close date has not been published, the page says so rather than printing an estimate. Observed outcomes (how the pity curve actually plays out across real pulls) are a separate, sampled figure and live on Pull Stats, not on the banners page, so a published mechanic and a measured outcome are never mixed in the same table.

Events and redemption codes are swept by hand, cross-checked against a second source, and stamped with the date of that sweep rather than left to the page’s own render clock. The event slate’s primary source is endfield.wiki.gg’s Events table, cross-checked against Game8’s own listing; codes follow Game8’s page, checked against the wiki where the wiki carries the code at all.

A code whose stated expiry has passed moves out of the live list rather than staying at the top with a strikethrough, and a code with no stated expiry says so instead of guessing one. This is the “as of” discipline that runs across the whole site: a live-data page states the date it was last checked, and that date is what moves, never a build timestamp that would make every page look freshly checked on every deploy whether or not anything was actually re-verified.

Every marker on the interactive map is imported directly from Endfield Tools’ own interactive-map dataset, reused with the owner’s stated authorization, and is not independently re-surveyed or re-placed by hand. The map page states the date of that import, and the marker count, category count and per-region breakdown shown on the page are computed from the imported data itself rather than typed in separately, so the numbers cannot drift from what the map actually renders.

The found checklist is the one part of the map this site adds on top: clicking a marker saves its id in your own browser, never on a server, so the checklist reflects nothing about how the markers themselves were sourced or verified.

Blueprints are hand-picked from Endfield Tools’ community Factory Plans library against a written list of demand topics (a beginner base, a mid-game layout, a late-game layout, and the rest), not generated or auto-selected. Only plans carrying a real, validated in-game share code are eligible; the source also hosts “simulation” plans built in its own planner tool with no real share code, and those are excluded outright.

Where more than one plan answers the same topic, the one with the most likes at the time of selection is picked as the starting recommendation, and each pick’s reason states the figure it was chosen on. The owner’s stated authorization from Endfield Tools covers the interactive map specifically; it was not given for this separate, user-generated Blueprints section, so this selection is credited by link and attribution rather than presented under the same permission.

The pull statistics are a static, dated snapshot of EndfieldPlus’s public headhunting tracker, aggregated across every player who imported their own pull history into that tracker. That is a self-selected sample, not a random one: it skews toward players engaged enough to install a tracker and hand it their data, which is not the same population as the whole playerbase.

The page states the sample size (players and total pulls) per banner next to every figure it derives from it, and the site never queries EndfieldPlus at request time: the numbers shown are exactly what was captured on the stated date, aggregates only, with nothing that identifies an individual player.