Play D&D every week and it could be three years before anyone's character dies
By Jordan Cole, founder of LoreifyPublished

Going down is an ordinary night at the table. Dying is not. In 16,077 sessions recorded with Loreify from November 2025 through October 2026, the notes show someone on the party's side knocked unconscious in about one session in four, and a death saving throw in about one in seven. Three failed death saves came up about once in 76 sessions. By the rules of fifth-edition D&D that means death, though the notes do not always count by the rules.
So check the notes against themselves. In fewer than half of those sessions did the list written at the end of the fight also mark the character dead. Count only those and a death is on record in about one session in 159. The honest answer is a range, from one session in 76 to one in 159, and it describes tables that record their games, counted from what their notes wrote down. At one session a week, that is three failed death saves about once every year and a half, and a death on record about once every three years, for the whole table and not for any one character.
Critical Role has spent the year on that thin line. Brennan Lee Mulligan's Campaign 4 killed Teor Pridesire, Travis Willingham's paladin, in June 2026, and Nerdist reported that Mulligan had asked the whole cast to keep a backup character ready. In October 2026, Nerdist described a later episode that left one hero banished after several near-deaths. Baldur's Gate 3 players know the feeling from every downed companion.
Down is common, dead is rare

About seven in ten sessions had a named monster or enemy in the notes. In those, someone was knocked unconscious in about one session in three, a death save came up in about one in five, and three failures in just under 2%. Nearly every three-failure session had a fight on record.
The notes hold 3,154 death-save entries, one per character per session. More than half recorded no failed save at all: the rolls went the character's way, help arrived first, or the notes did not catch the roll. About one entry in 13 reached three failures.
Of 2,258 campaigns, 178 (8%) had a three-failure entry in the year. Short campaigns flatter that figure, so here it is by length.
| Sessions recorded in the year | Campaigns | Any three-failure entry |
|---|---|---|
| 1 to 9 | 1,730 | 3.8% |
| 10 to 19 | 347 | 17.9% |
| 20 or more | 181 | 28.2% |
Even among those with 20 or more sessions, about seven in ten had no three-failure entry in what they recorded that year.
Two cautions. The share moved as the way the notes are written changed: ten of the twelve months fell between 0.9% and 1.4%, while June and July 2026 ran at 1.8% and 2.3%. And a looser test gives a higher number. An end-of-fight status that reads "dead" turned up in 2.7% of sessions. But that list takes in allies and pets: where a campaign had a saved character list, only about a third of the "dead" statuses matched a listed character, against nearly two thirds of other statuses. The strict measure has a blind spot too: a death that never reaches a death save, such as a blow big enough to kill outright.
No deadly opening stretch
We expected deaths to bunch at the start of a campaign, when characters are fragile. There is no clear early cluster. Counting each campaign's sessions in the order they were recorded, three failures came up in about 1.1% of sessions 1 to 3 and between 1.3% and 1.5% of every later stretch. The first recorded session alone came in at 1.55%, the highest of any stretch, and the next two at well under 1%. We would not build a rule on that dip: session one is barely above the later stretches, and ranking by the order tables set themselves gives a flatter line.
The first recorded session is not always a campaign's first, and nothing in the data says what level the characters were. What the numbers support is narrower: risk follows the fight, not the session count.
Three failures is not the last word
Where a campaign recorded at least three more sessions, about half of the three-failure entries were for a character who appears again in a later end-of-fight list without a death word in the status. That is not proof of a revival. It cannot tell a Revivify from a namesake, or from a fight the notes miscounted.
Critical Role blurs the same line. Nerdist describes one of its characters cut down in the fourth episode and brought back among the living without being alive. Whether that counts as a death depends on who is counting.
So how worried should you be? As a player, expect to hit the floor and expect to get up. In the entries (about a quarter) that named one of three common kinds of help, it was a healing spell or feature such as Healing Word or Lay on Hands more than half the time, and a potion about a quarter. As a Dungeon Master, a table that has gone 20 sessions without three failed saves is not soft. It is typical. A table that loses a hero every few sessions is playing a much harder game than most, and that is worth saying out loud early. If the roll itself is hazy, our guide to death saving throws covers it.
How we checked the numbers
What was counted. Completed sessions that belong to a campaign and have structured notes, recorded with Loreify in the twelve months from November 2025 through October 2026: 16,077 sessions from 2,258 campaigns. The evidence ends on the last day of October 2026. November 2026 is left out because the month is not finished. Sessions outside a campaign, or without structured notes, are left out.
What it is. The notes are written from a transcript of the session. They record what came up, not a ruling on what happened, and the way notes are written changed during the year. These are tables that record their games with Loreify: mostly English-speaking, and mostly but not only playing D&D fifth edition. No field says which game a table plays, so a table whose game has no death saves still counts in every total. The notes of 7.9% of sessions carry no death-save field at all; among the notes that do, three failures came up in 1.43%.
The measures, with the exact shares. "Knocked unconscious" (23.8% of sessions) means an end-of-fight status entry marked as having been unconscious, or reading downed, unconscious, knocked out or dying. A death saving throw (13.8%) means at least one death-save entry. "Three failed death saves" (1.32%, 212 sessions) means an entry recording three or more failures; an entry whose failure count is not a number counts as fewer than three. Both lists can include allies and pets as well as player characters. "Marked dead in the same session" (0.63%, 101 sessions from 89 campaigns) means a three-failure entry whose character also appears in that session's end-of-fight list with a death word and no revive word. In the other three-failure sessions the character was listed with a revive word, listed with neither, or not listed. The looser "dead" status leaves out undead, revived and nearly dead. A fight (68.8% of sessions) means the notes named a monster or enemy; in those sessions the shares were about 32%, 19% and just under 2%.
The three years. The headline figure is arithmetic: one session in 159 with a death on record, at one session a week. The looser "dead" status count, which takes in allies and pets, would make it shorter.
Entries. Of the death-save entries, 55.8% recorded no failed save and 7.8% recorded three or more. "Appears again" (49.1% of three-failure entries, against 76.0% of the others) means the same first word of a character's label turns up in a later end-of-fight list of the same campaign with no death word in the status. One of three kinds of help was named in 750 of 2,907 entries with fewer than three failures (entries that named another kind, or only who helped, are left out): a named healing spell or feature in 55.7% of those, a potion in 26.4%, a Medicine check or Spare the Dying in 17.9%.
Session number is a session's place among its campaign's completed sessions, by recording date. By stretch the shares were 1.06%, 1.47%, 1.43% and 1.34%; by the order tables stored, 1.23% to 1.42%. Sessions 2 and 3 together came in at 0.72%. Where a saved character list existed, 36.4% of "dead" statuses matched a listed character, against 63.9% of other statuses.
Privacy and the reporting floor. Labels were compared by the query and never read or printed; no names of people, characters or campaigns were used. Every share is a share of at least one hundred sessions from at least fifty campaigns.
The news. Critical Role facts are from Nerdist (June 2026, October 2026 and October 2025) and Wargamer, read on November 20, 2026. We found no report of a player-character death in the campaign after June 2026.