Data Methodology

What is in the current dataset

The current snapshot contains 520 seed records across 6 map types and 9,845 populated point-of-interest placements. Of those records, 251 include a named special event and 101 carry the Old Gaol flag. The site currently labels this snapshot for game version 1.03.2.

Each record has a seed ID, map type, Nightlord, and a slot-to-POI mapping. Some records also include a special event, Old Gaol flag, marker data, or coordinates. A blank slot means the source record does not assign a POI there; it must not automatically be interpreted as a claim that no dynamic activity can occur at that position.

Sources and attribution

The 520 records that hold map type, Nightlord, special event and the slot positions were compiled and verified by Artimuz for this site, as were the marker files for Evergaols, Stonesword Keys, Townships and Scale-Bearing Merchants, and the boss data.

That does not mean 520 expeditions were played to find them. Research of this kind combines analysis of structured game data with observations that can be checked during a normal run, and both went into these records. What is claimed is that the records were assembled and checked for this project, nothing beyond that.

The building icons on this site are composites of the game's own symbols. The game has one symbol for a structure such as Marsh and separate ones for affinities and status effects such as Fire and Madness, while a player looking at the map sees a single structure carrying one of them, so the two were merged into one icon here. The artwork is FromSoftware's. Artimuz chose which pairs belong together and did the compositing. Of the 53 building icons, 33 are composites of this kind, and marking a slot means matching what you saw against one of them.

The map and the finder are translated into nine languages besides English: Português (Brasil), 日本語, Español (Latinoamérica), Español (España), 简体中文, 繁體中文, العربية, 한국어 and Deutsch. That covers the interface strings and the in-game term names shown on them: POIs, bosses, Nightlords and events. The editorial pages, this one included, are English only. Every translated term was looked up in the game running in that language and matched to the wording the game displays there, not to a machine translation or a community wiki.

thefifthmatt is credited here as an early source. His published catalogue was the reference when the tool started in 2025, and it is a large part of why the project got off the ground at all. kevins78 is credited for the Great Hollow and DLC-era material that informed the work at the time.

The website running in a visitor's browser does not access or modify the game process, memory, or files. That is a statement about what the site does, and it is separate from how the records above were gathered.

Transformation and matching

1. Source values are normalized into stable map, Nightlord, event, slot, and POI identifiers.
2. Choosing a map and optional Nightlord creates the initial set of compatible seed records.
3. Every POI mark is compared with the value stored for that slot. Records with conflicting values are removed; an unmarked slot does not remove anything.
4. When one candidate remains, the result view loads that record's full layout. If several remain, the useful next action is to mark a slot whose possible values differ between candidates.

This is deterministic filtering over the available dataset, not a confidence model that can repair an incorrect observation. If a church is marked as a fort, the correct seed may be filtered out. The finder therefore favors a few high-confidence marks over filling every visible slot quickly.

How checks and corrections work

A record can be checked by comparing the result layout with the corresponding in-game run. Reports sent through the site include enough context for Artimuz to locate the affected flow; reporters should also provide the seed ID, map type, slot, expected POI, observed POI, and game version whenever possible.

Reports are reviewed as discrepancies, not automatically accepted as corrections. Common causes include selecting an adjacent slot, confusing an elemental POI variant, using data after a game patch, or an actual source-record error. Confirmed changes should be documented in the public update history.

Probability limitations

Percentages on the event-hunting page are descriptive proportions within the records being compared. For example, an event appearing in 4 of 20 compatible records is displayed as 20%. That does not prove the game selects those 20 records with equal probability. Reliable real-world odds would require a sufficiently large, independently collected sample of actual runs and knowledge of the game's selection rules.

That page counts only records on a Shifting Earth, keeps the base and DLC pools apart and holds the DLC pool to the Balancers and Greg, so its denominators are smaller than the dataset totals quoted elsewhere here. A higher percentage in a smaller group can still point somewhere worth hunting, but check what is behind it first: with five records in the group, one of them is worth twenty points on its own. That is why the page prints the record count beside each best pairing, and says how many pairings share a top figure when more than one does. The pool sizes themselves are stated once in the notes under the page rather than repeated on every card.

Known limits

• A game update can make a previously correct record stale until the dataset is reviewed.
• Cataclysm is not governed by the seed. The same seed produces it in different places, so this finder cannot help with it, and a larger dataset would not change that. It underpins the Night Invaders strategy that the community rates as the strongest in the game, which makes it the biggest thing the site does not cover.
• Forts always hold a map at the top of one tower, and reading it in game reveals both the chests and the Scarabs. Only the chests are seed-determined, and this site already carries them: they sit on the result map behind the Chests filter. Scarab positions are not in any record and never will be, because they are random rather than seed-bound, so the talismans they drop stay behind the climb. On that half the in-game Fort map beats this tool outright.
• Missing source fields are shown as unknown or not recorded, rather than silently inferred.
• A result is only as reliable as the user's map type, Nightlord, and POI selections.
• Counts describe this dataset; they are not a substitute for official game documentation.