The short answer
The rarest seed on this site is 2016000 on Java, and the reason is arithmetic rather than opinion. We searched all 49 records for every structure and feature they name. Eight features appear in exactly one record, and 2016000 holds three of the eight — the only stronghold, the only desert temple and the only jungle temple in the entire set. No other record holds more than one. A further eight features appear in no record at all, including the trial chamber and the woodland mansion.
- The only record with three one-offs:
2016000— Java 1.20–1.21, spawn 0, 63, 0, stronghold at -21, 42, 18, desert temple at -320, 64, 180, jungle temple at 450, 63, -270. Six coordinates in one row, the most of any record. Stronghold, only record Two temples, only record Full xyz for all three Starts on the mainland, not an island - The only ancient city:
6390326396262027557— Java and Bedrock 1.18+, a mountain island with an underground ancient city and a warden in it. Only ancient city Island spawn No coordinate given - The only sugar cane:
-1276503895— a barren island with a few trees, a few stalks of sugar cane, an air-exposed diamond and a lava pool. Only sugar cane Diamond to portal without leaving Not a Java seed - The only plains spawn:
32— a small grass island, a few trees, sheep, and a cave system under it with a spawner. Only plains Two-character code No coordinate given - The only taiga, and the only spawn inside a dungeon:
1736813803934715730— you open your eyes inside a skeleton spawner room with taiga overhead. Only taiga Spawn inside a spawner room No coordinate given
The codes in this section — 2016000; 6390326396262027557; -1276503895; 32; 1736813803934715730; 39720048960
How we counted, so you can check the number yourself
Every figure on this page is a count of words inside this site's own records. None of it comes from the game, a generator, or anyone else's list.
What a record is. Forty-nine things: the twenty rows of the coordinate table on the home page, and the twenty-nine archive cards on the 2011–2019 page. A record is one row or one card. If a word appears twice inside one record it still counts as one record.
What counts as a hit. The word appears in the description text of a record. For a row, that is the description cell plus its coordinate cell. For a card, that is everything under its heading. Navigation bars, footers, headings, FAQ answers and prose elsewhere on the page are all excluded — the site name for the obsidian route page is not evidence that a record mentions obsidian, and it took a filter to keep that out.
What we are not measuring. How often a structure generates in Minecraft. That number would need a generator and a scan; we have neither. Rare on this page means rare in forty-nine records written by people who were looking for specific things. A witch hut scores seven here because seven different players went looking for one, not because witch huts are common in your world.
The one place this gets awkward. Obsidian reads as zero in the record text, but another page on this site describes three blocks of it in a blacksmith chest. We are listing it as unresolved rather than absent, because a count that quietly wins an argument against our own page is worth less than a count that admits the edge.
The eight features that appear in exactly one record
Sorted the way they sit in the set: three on a mainland village row, four on island rows, one on an archive card. The coordinate column is the important one. Three of these eight hand you a position you can walk to, and all three are on the same row.
| Feature | Records | The one record | Coordinate in the record | Tag |
|---|---|---|---|---|
| Stronghold | 1 / 49 | 2016000, Java 1.20–1.21 |
-21, 42, 18 | Only record, full xyz |
| Desert temple | 1 / 49 | 2016000, Java 1.20–1.21 |
-320, 64, 180 | Only record, full xyz |
| Jungle temple | 1 / 49 | 2016000, Java 1.20–1.21 |
450, 63, -270 | Only record, full xyz |
| Ancient city | 1 / 49 | 6390326396262027557, Java + Bedrock 1.18+ |
None recorded | Only record, no coordinate |
| Sugar cane | 1 / 49 | -1276503895, regional edition + Bedrock |
None recorded | Only record, no coordinate |
| Plains spawn | 1 / 49 | 32, Java + Bedrock 1.18+ |
None recorded | Only record, no coordinate |
| Taiga | 1 / 49 | 1736813803934715730, 1.2, archive card |
None recorded | Only record, no coordinate |
| Ice peaks | 1 / 49 | 39720048960, Java 1.18–1.21.2 / Bedrock 1.18–1.21.50 |
/tp -576 ~ 80 — x and z, no y |
Only record, partial coordinate |
Read the coordinate column as a promise, not a decoration. Three full positions, one target with a missing height, four blanks. That is the entire navigational value of the rarest layer of this site’s data.
The eight features that appear in none of the 49 records
These returned zero across every row and every card. Two of them are the ones players actually ask about.
| Feature | Records | Why it is worth naming |
|---|---|---|
| Trial chamber | 0 / 49 | The newest structure in the game and the one most likely to be searched for by name. Our records stop before it existed in a form anyone wrote down. |
| Woodland mansion | 0 / 49 | Thirteen years of records, no mansion. Nobody in this set was ever writing about one. |
| Igloo | 0 / 49 | Snowy terrain appears once, as ice peaks on a multi-biome island, and even that row does not name an igloo. |
| Coral | 0 / 49 | Seven records name a warm ocean, and not one of them describes what is in the water. |
| Spruce | 0 / 49 | The one taiga record describes the biome above a dungeon, not its trees. Tree species are essentially absent from this set as a field. |
| Savanna | 0 / 49 | Zero, in a set that names jungle eight times and badlands five. |
| Swamp | 0 / 49 | Zero. No witch hut record bothers to say what biome the hut stands in. |
| Mainland | 0 / 49 | The word never appears in a record, even though every island player’s first question is how far away it is. |
The pattern in the zero column is not about the game. It is about what people wrote down. Swamp is zero because nobody describing a witch hut mentioned the ground it stood on. Coral is zero because a warm ocean coordinate was treated as the whole fact. Trial chamber is zero because the records were written before it. Only the last of those is a fact about the world; the other seven are facts about our field list, and we would rather show you that than pretend the zeros mean something they do not.
What is not rare in these records, and why that is misleading too
The same scan run in the other direction. These are the features that come back crowded, and every one of them is crowded for the same reason: this site was built by people hunting them.
| Feature | Records | Where it concentrates |
|---|---|---|
| Ocean monument | 9 / 49 | Seven of the nine sit in the seven Java server rows — three kept, four rejected — where a monument was a requirement rather than a sight. |
| Tree | 9 / 49 | Six coordinate rows and three archive cards, and all three cards are from 2011. The one feature that reads the same in both halves of the set. |
| Pillager outpost | 8 / 49 | Almost entirely in the Java coordinate rows. One archive card, from 2019, names an outpost at spawn. |
| Jungle | 8 / 49 | Five archive cards and three coordinate rows. The most over-represented biome in the set by a wide margin. |
| Floating and overhang terrain | 8 / 49 | Seven 2011 archive cards plus one from 2012. An eighth hit is a 2019 card that mentions floating islands only in a sentence pointing at another page. |
| Witch hut | 7 / 49 | Every single one is a coordinate row. No archive card mentions a witch hut, because the hut did not exist yet for most of them. |
| Bastion remnant | 7 / 49 | Coordinate rows only, and in four of the seven it is the reason a seed was rejected. |
| Warm ocean | 7 / 49 | Coordinate rows only. Always a coordinate, never a description of what is in it. |
| Shipwreck | 7 / 49 | Six coordinate rows and one 2019 archive card. One of four features that show up in both halves, alongside tree, jungle and badlands. |
Nine out of forty-nine sounds like a lot until you remember the denominator is a curated set. If we had recorded fifty island seeds at random the tree count would be fifty, and the ocean monument count would be close to zero. The number that actually tells you something is this one: only 22 of the 49 records name any structure at all — sixteen of the twenty rows and six of the twenty-nine cards. More than half of this site’s data is terrain description with nothing built on it.
Why one row holding three of the eight is the real finding
Six records hold the eight one-off features. One of them holds three, and the other five hold one each. That distribution is the whole story of this page, because it means the rarest thing in the set is not a structure — it is a row that happened to be written by someone with the patience to measure everything in it.
Look at what that row actually contains. Spawn at 0, 63, 0. A village at -24, 64, 12. A stronghold at -21, 42, 18, which is under the village rather than out in the world somewhere. Diamond ore at -32, 11, 25. A desert temple at -320, 64, 180. A jungle temple at 450, 63, -270. Six positions, all three dimensions, in a single line of text. No other record in forty-nine comes close on height: the next richest row lists nine positions, and every one of them is a flat x, z pair with no y value at all.
Then the part we have to say out loud. That row is a plains village spawn. It is the only one of the six one-off records that does not start you on an island. The other four island rows hold the other four one-offs between them and give not one full coordinate; the archive card gives you a dungeon ceiling and no coordinate at all. So the honest summary is uncomfortable: on this site, the rare structures with a full coordinate are on the mainland, and the rare structures on islands are descriptions.
That is not a gameplay law. It is a record-keeping artifact, and the likeliest explanation is the simplest one — measuring coordinates takes effort, and the person who measured six of them was measuring a seed they were recommending as a speedrun start, not as an island. We are writing it down because it is what the data says, not because it is a pleasant thing for an island site to admit.
The judgment standard: is a one-off worth the trip?
Before you load a seed because a page called it rare, run these four checks against the record itself. They are the checks this page would fail on five of its own eight entries.
- Is there a coordinate, and does it have all three numbers? Three of the eight give full x, y and z. One gives x and z. Four give neither. A structure described but not located is a rumour; you will spend your first two hours walking in circles looking for a temple that may be a thousand blocks out.
- Does the record name the edition and the version? A one-off feature on an unlabelled record is untested by definition. The ancient city row is the good case — Java and Bedrock, 1.18 and up. The taiga card is the bad case: version 1.2, from 2012, and nothing that says whether the code still resolves to the same terrain.
- Is the feature above ground or below? Three of the eight are below the surface: a stronghold under a village, an ancient city under a mountain, and a spawner room you open your eyes inside. Underground means you cannot see it from the spawn point, which means the coordinate stops being optional and starts being the whole plan.
- Does the record say what is inside it? Almost never. The desert and jungle temples on
2016000are recorded as coordinates and nothing else — no chest, no loot, no note on what was in one. The ancient city is recorded as having a warden in it. A spawner room is recorded with skeletons already moving before you have a sword. Rare is not the same as useful, and in two of the eight the rare thing will kill you on day one.
The rule we would apply to our own page: if a seed’s only claim is a feature with no coordinate, treat the seed as a terrain recommendation and the feature as trivia. Load it for the island, not for the temple.
First-day strategy on a seed you picked for a rare feature
Take 2016000 on Java, because it is the only one here where the rare things are also reachable. The village is at -24, 64, 12, which is close enough that you can walk to it before the first night, and the stronghold is at -21, 42, 18 — directly below that village, not somewhere else on the map.
Java, 1.20–1.21: 2016000
Hour one — the village, not the temples. The desert temple at -320, 64, 180 and the jungle temple at 450, 63, -270 are both several hundred blocks out in opposite directions. Neither is a first-night destination. Take the village: beds, food, and a roof. The temples are day-three trips and they will still be there.
Hour two — the stronghold is not a day-one goal either, but the shaft is. A stronghold at y=42 under a village gives you stone brick, which is a building material and not a survival tool. What it does give you is a reason to dig down rather than wander, and a straight shaft from the village well area is easier to navigate than a cave system.
The diamond at -32, 11, 25. That is the coordinate on this row that changes your first day, because it is a known position at diamond level with a y value attached. Mining to a coordinate beats branch mining when you have one, and this row is one of only eleven table rows on the site that carries a coordinate at all.
What to skip. If the seed you picked is one of the four island rows with a one-off feature and no coordinate, do not spend day one hunting the feature. Survive the island first; the ancient city under the mountain or the spawner in the cave will still be under there on day five, and you will have a pickaxe by then.
Where the rest of the count lives
The eight one-off features are one slice of the same scan. Two neighbouring slices are worth your time if this is the question you came with. The coordinate page collects every record that has a position at all — eleven of the twenty rows, nine of the twenty-nine cards — which is the reason five of the eight one-offs above can only be described. The rejected page is the other end of the same problem: four seeds that had the structures and lost on distance, which is what happens when a rare feature is real but unreachable.
And if the rare thing you are actually after is a mushroom island rather than a temple, that count is here — four records out of forty-nine, which puts it in the crowded column rather than the rare one.
Running one of these with other people
If you want to check any of these coordinates with friends on Java or Bedrock, Shockbyte is a low-cost host that sets up in minutes — use code LAUNCH for 25% off your first month. If you sign up through that link, I may receive a commission at no extra cost to you. It does not change what you pay.
Frequently asked questions
What is the rarest seed in these 49 records?
Seed 2016000 on Java 1.20–1.21, because it is the only record that names a stronghold, the only one that names a desert temple and the only one that names a jungle temple. That is three of the eight features which appear in exactly one record out of forty-nine, and no other record holds more than one of them. The three coordinates sit in the same row: stronghold at -21, 42, 18, desert temple at -320, 64, 180, jungle temple at 450, 63, -270.
Which features never appear in your records at all?
Eight of them: coral, igloo, woodland mansion, spruce, savanna, swamp, trial chamber and mainland. They were searched for in all forty-nine records and returned zero. Trial chamber and woodland mansion are the two that matter most, because both are structures players actively hunt for and neither one appears anywhere in a record that goes back to 2011.
How rare is a real survival island seed?
We cannot answer that from our own data, and we will not borrow another site’s percentage to fake an answer. These 49 records are not a sample: they were written down because a player posted something worth writing down, so a rate calculated from them would describe our reading habits, not world generation. What this page can tell you is how often a feature appears inside those records, which is a different question.
Is a witch hut rare in Minecraft?
Not in these records: seven of the forty-nine mention a witch hut, which makes it one of the most common structures we hold. That number says nothing about how often a witch hut generates in a world. Our records were built by people who were hunting witch huts, so the field is crowded on purpose. Rarity inside a record set and rarity inside a world generator are two different measurements and only one of them is on this page.
Do you have per-chunk rarity data?
No. Nothing on this page is a probability, a spawn rate or a one-in-N figure. Every number is a count of how many of our own records say a word. We do not run a world generator, we do not scan seeds in bulk, and we have no way to tell you what fraction of worlds contain an ancient city. If a page gives you that number, ask what it scanned.
Which one-off features come with coordinates?
Three of the eight, and all three are on seed 2016000. The other five give you a description and nothing to walk to: the ancient city on 6390326396262027557 has no coordinate, the sugar cane on -1276503895 has none, the plains spawn on 32 has none, the taiga on 1736813803934715730 has none, and the ice peaks on 39720048960 give a teleport target with an x and a z but no y.
Can I get a rare structure on an island spawn?
Yes, four of the eight one-off features sit on an island spawn. The ancient city on 6390326396262027557 is under a mountain island, the sugar cane on -1276503895 is on a barren island, the plains spawn on 32 is a small grass island and the ice peaks on 39720048960 are reached by teleport from an island. The catch is that none of those four gives you a coordinate for the thing you came for, so you will be searching the island yourself.