Night at the Fortress Gate

How it works, and what you are actually deciding

Fourteen turns until dawn. Each turn you aim the torch at one lane — or call on a companion first — and the night moves on. A threat that closes the distance while its lane is dark and unanswered costs that lane ground; enough of that and the lane is through. Nothing here is a roll: every strike is exact, every torch beam is exact, and the only thing that is genuinely uncertain is which of the three lanes needs you most right now.

Five companions, one verb each. The dog tells you whether anything is close on a lane, without needing the torch there. The bowman can loose blind at an unlit lane, at half his lit chance. The builder braces one lane for the rest of the night, once. The runner buys a lane one more turn in the dark. The captain will tell you, once, which lane is genuinely worst right now. None of them is a second torch — each is a way to answer for a lane you are not looking at, never a way to look at two at once.

Six things were measured, and it took five rounds of fixes before the numbers were honest

1,500 seeded nights per figure, against the shipped rules. Two of those rounds turned out to be bugs in the measurement itself, not the game — worth saying plainly, because a broken measurement still prints a plausible-looking number.

Reading had to beat a fixed rotation, and it took three attempts. At first, the torch's reach and the three-lane rotation period matched almost exactly, so cycling the torch lane by lane with no thought at all cleared 88% of nights — reading a sign bought nothing. Narrowing the torch's reach exposed a second fault: the policy measuring "responsive" play remembered what a lane held forever once checked, so "clear back at turn three" was still trusted at turn twelve — worse than a blind rotation that revisits every lane for free. Even after both of those, it was recording "I aimed the torch here" as permanently safe regardless of what it actually found, so it locked onto blindly re-torching one lane for most of the night. Fixed by making every reading decay honestly. Shipped: blind rotation clears 31.5% of nights; reading it well clears 76.7%.

Every companion has to be worth calling — the hardest of the six, because the test measuring it was quietly broken: it removed a companion's action from a turn, but not the information that companion had already fed into the plan, so "what if I never had a captain" measured as exactly no difference, always. Fixed, and it then surfaced two real gaps in the game: the dog was answering "will anything ever come" instead of "is anything close," and calling the dog changed nothing you could actually see on the lane card. Both are fixed now — the dog scents what's near, and says so. Shipped, cost of skipping each one: dog +0.07, captain +0.04, runner +0.07, bowman +0.19, builder +0.28 lanes lost.

Lightning has to pay without deciding the night — and it does, by a real margin: using a flash's information is worth about 0.1 lanes over ignoring it. But a night with no lightning at all still clears clean 68% of the time on the torch and companions alone, against 77% with it. A real boost, not the deciding factor.

No lane may be safe to ignore every night — the road falls 0.4% of nights, the ditch 12%, the postern 13%. It is still the strongest lane, honestly, just not unconditionally.

The gate holds completely sometimes, and does not others, under the same good play — 77% clean, 0% lost outright, under the full kit; blind play clears 0% clean and loses everything 80% of the time.

The fortress master never claims a lane held if it did not, or fell if it did not. Enforced in the code rather than hoped for.

It runs in your browser and stays there

No account, no sign-in, no network. Your watch becomes a short code you can share, and the link opens on what got through and what did not. There is no code on this page that could send anything anywhere: pages in this corner are forbidden to make requests at all, and a test enforces it rather than anybody remembering.

The gate, the torch and the rain are drawn with a 3D engine — the torch is a real cone that sweeps to wherever you aim it, and lightning is a genuine flash across the whole scene rather than a number on a meter. If your browser will not give the page a canvas, everything is described in words instead and nothing about the watch changes.

Argue about it in the forum

The board’s forum has a room for this. Post which lane went through, and be told you should have called the dog on the postern first.

This is fan-made, and unofficial

This is an unofficial fan game inspired by The Last Kingdom — the television series and the novels of Bernard Cornwell it is based on — naming the work, its setting and its premise under nominative fair use. It is a free, unofficial fan tribute, made for fun. It is not affiliated with, endorsed by, or connected to Carnival Film & Television, Netflix, or Bernard Cornwell and his publishers.

The gate here is not any particular gate and nobody in it is named. A watch, a torch, rain and a night that goes on too long are the ordinary furniture of anybody who ever stood guard. Every line of text here was written fresh for it. No dialogue, novel text, script text, scene, still, logo or artwork from any television series or book is reproduced anywhere in it. Iron Realms — the strategy game the rest of this site is about — is an original fantasy world with its own map, its own houses and its own art, and has no connection to the world of The Last Kingdom.

The rest of the corner

The Last Kingdom wing has Lord of Bebbanburg, Hold the Wall, Holmgang, The War for the Kingdoms, Saxon or Dane, Arseling to Ealdorman, The King’s Shadow, Storm the Coast, The Burh Holds, the trivia rounds and the character pages. The fan tributes hub has the Westeros, Supernatural, King of the Hill and Fringe wings alongside it, and the free tools that are not about anybody’s fiction. If you would rather play something with a map in it, Iron Realms is free and runs in a browser tab.