How the Wipe Report works

You paste a raid attempt. It reads every wipe out of the log and says which role the deaths point at, how sure it is, and exactly which measurements got it there. It is aimed at a role, never at a person, and on a messy night the honest answer is often that it cannot tell.

What it reads

Three things, all straight from Warcraft Logs: the night's fights (which were kills, which were wipes, how much boss health was left, how long each took), the roster of each pull — Warcraft Logs' own tank / healer / dps buckets — and the deaths in each wipe: who, when, and the ability that landed the killing blow.

The roster is fetched per pull rather than once for the night, because asked over a whole report Warcraft Logs buckets everyone who logged in all evening. A real 25-man SSC night came back as 205 actors and one tank.

The one idea the verdicts rest on

The collapse is not the cause. Every wipe ends with the whole raid dead. Those deaths are the wipe, not the reason for it, so the engine ignores the last fifteen seconds of the pull — or the last quarter of it, whichever is shorter — and reasons only about what happened before. Without that rule every wipe in history reads as a mass mechanic failure.

How a death becomes evidence

Each death before the collapse is looked up by the ability that killed it, against a vendored table of what every TBC raid ability means — avoidable damage, raid damage, tank damage, damage from an add, and so on. That is what turns "died to Arcane Orb" into "stood in something", and what separates a tank eating boss melee (normal) from a loose add meleeing your healers (nobody picked it up).

A death that the table cannot explain counts for nobody, and the page says so rather than quietly rounding it into the verdict. So does a death that is explained but proves nothing — a dps eating boss melee is what a raid that is already losing looks like.

The verdicts

The colour is the kind of answer, not the verdict: coral accuses a role, amber declines to — a pull collapse and an inconclusive wipe are painted the same — and grey is a pull that was not a wipe at all.

Naming a role takes a share of that role, not a head count: at least two explained deaths charged to it, worth at least 35% of the people the log lists in that role, and at least double the next role's share. The role tested is the one with the biggest share, not the one with the most deaths — a flat count asked the same of five healers as of seventeen damage dealers, and it let the larger role be named for being larger while the denser one was never looked at. A raid that loses three of its five healers is a healing story that five dead dps out-counts and does not out-weigh. On a log that reports no roster at all there is no share to take, so the older test stands there unchanged: three deaths of a kind, with a lead of at least two.

A role's own deaths never accuse that role. Where the ability table blames a death on the person who died — a fire pool, a frontal cone, a bomb they were meant to run out with — that death is counted toward nobody's verdict: not the share, not the margin, not the floor. One rule, and it reads the same for tanks, healers and damage dealers. Seven damage dealers dying in one fire pool is seven people standing in a fire pool; it is not a measurement of anybody's damage, and "where someone stood" is the one thing this page says it cannot see. The deaths stay on the card, named as what they were, with the ability's own explanation — this is the real line from a Shade of Akama wipe:

"7 dps died to Rain of Fire — An Ashtongue Elementalist drops a ten-yard fire pool for eight seconds: 3,238 a tick, one tick every two seconds, about 13,000 for staying in a whole one. It is instant and it does not follow anybody — the pool stays where it landed. Their own deaths, which count for nobody's verdict."

The verdict above them rests on the rest of the evidence or on nothing. One consequence worth knowing: where a role's whole charge was its own deaths, nothing is left to name anybody, so a pull that left the boss at 15% or less can fall through to the throughput read and be answered as pace — "the raid ran out of fight, not out of people" — instead of being blamed on the tank who died in a fire.

A role can be charged with more deaths than the log counted people in it — three deaths charged to a two-tank raid is 150% of the role. That is a tally of deaths charged to a role rather than of that role's members dying, and it is the strongest thing the count can say, so it names the role rather than being held back. It is also why the page writes "3 deaths at the 2 tanks" and never "3 of 2 tanks".

Named on one signal. A wipe the deaths could not separate can still be named by one clear signal in the last thirty seconds before the collapse, and only by one. There are three: a tank dying with a defensive cooldown never pressed; a tank dying with less than a quarter of the damage that killed him healed back, across a window of at least three seconds; and a raider pulling the boss off the tanks, holding it three seconds or more, dying for it, with two more dead within ten seconds. A tank death the ability table already blames on the victim is never one of them — dying to Illidan's Eye Blast with a cooldown up is a question about where someone stood, and this tool does not read positioning. A verdict reached this way is always medium and says so in its own second bullet — "Named on one signal: the raid was holding until this. Medium confidence, on purpose." Two different kinds of signal in the same window name nobody: each alone would point at a different role, so the verdict stays INCONCLUSIVE and one line reads both out. A cooldown counts as unused only where the log can prove it was there to press — the tool knows each cooldown's own length, drops one this tank already spent earlier in the pull, and requires the time out of combat plus the time into the pull to cover the cooldown, so a thirty-minute Shield Wall is not held against a tank ninety seconds into an attempt that began minutes after the last one. And the signal only speaks while the raid was standing: at most a quarter of the roster dead before it. The sentence it prints says the raid was holding until this, and on a pull with twenty-one of twenty-five already down, that sentence is false. It is false on a different pull too — one that reached a sliver of boss health — so no signal names anybody where the boss finished at 15% or less. A raid that got there needed twenty more seconds of damage, which is a pace story and not a moment, and the page says so instead: "A tank death in the last 30 s is not named here: the boss was at 8%, and that is a pace story." The death itself still prints, with its blow, its numbers and the cooldown nobody pressed. A role the deaths themselves name is untouched by this — it is one moment that cannot carry a pace problem, not the evidence.

The wall. When the last three attempts on a boss all ended within three points of boss health of each other, one teal line says so — "The wall: 65.1%, 64.7%, 65.1% on the last three pulls — the raid reaches the same place each time. Look at what happens there, not at who died." It accuses nobody, and it is deliberately rare: it printed on one wipe of the seventy-eight this was measured against.

Pace against your own kill. When the same night also killed the boss, one teal line — under the wall's, where that one prints too — compares the two attempts; this is the shape of that line, on the night's own Azgalor kill: "Pace: 60% of your kill's — at this rate the boss dies at 4:03 against the kill's 2:25." Both rates are whole-fight averages off a single measurement, the health the log saw the boss lose on that kill, so the projected clock is that average carried to the end — which is all "at this rate" claims, and nothing more. Any fight whose model carries more than one phase prints the percentage without it: a whole-fight average is a rate only on a fight that has one, and a model carrying more than one phase is not something this tool can say that about. The line needs that model to exist at all: on a boss this tool carries no fight model for it says nothing — not the clock, not the percentage — because "how many phases" is then a question with no answer, and a kill whose clock is mostly a phase the boss took no damage in would make the percentage itself wrong rather than merely unprojectable. On a night that never killed the boss the line does not appear at all — there is nothing honest to compare against, and another night's kill, in other gear with other people, is not it. And it appears only where the attempt left the boss at 25% or less — looser than the 15% that silences a signal above, because no verdict moves on this number.

Pull collapses and resets

Two of the shapes on a raid night are not a fight anybody failed at, and the report says which one it is rather than reaching for a role.

A pull collapse is a raid that fell over at the pull instead of during a fight: under sixty seconds, the boss still at 95% or more, at least 40% of the roster dead, the first death inside five seconds, and at least 60% of those deaths dealt by the boss itself or by something the fight model owns. The card reads "The raid fell over at the pull: 20 of 24 dead inside 0:43 with Rage Winterchill at 99%, most to Melee (7) and Death & Decay (7) — the boss's own damage, from the first seconds.", then the first raider the boss itself killed — a warlock's own Hellfire at 0:01 is not what that line is about — then the point of it: "Not a fight the log can judge: nobody held anything long enough to fail at it. A mis-pull, a raid not ready, or a wipe on purpose all look like this." The boss gets no verdict, no role is named, and the consumables bullet is left off — a raid wiping deliberately to reset Hyjal's waves is not unbuffed by accident, and the log carries nothing that separates it from a mis-pull. The roll-call table is still there to open; the verdict just does not point at it.

A reset is the other half of that shape: under forty-five seconds, the boss at 95% or more, and three dead or fewer — a bad pull angle, a tank not ready, someone still walking in. One line: "Reset at 0:16 with 1 dead — a false start, not a wipe. Counted for nobody." The row is dimmed in the list and drops out of the night's wipe count, so the number of pulls on the page still matches Warcraft Logs. A pull the report could not read at all is a different thing and stays in that count: it was a wipe, it just could not be judged.

Confidence, and what it is measuring

Confidence is about the margin, not about how bad the wipe was. When a role is named from the deaths, high means it owns four or more of the explained ones and leads the runner-up by more than double; medium means it cleared the bar without that much daylight. The throughput verdict is scored differently, because its margin is the boss's health bar rather than a death count: high under 10%, medium between 10 and 15. Every INCONCLUSIVE verdict is low, by construction — it is the engine saying the evidence did not separate. A pull collapse and a reset are both high: they are shapes the log measures rather than cases the engine argued. A verdict named on one signal is always medium, and never anything else.

What the log can now show, on a Black Temple or Hyjal wipe

For the fourteen Black Temple and Hyjal bosses, the report also reads four more things out of the log for every pull, against a fight model built for that boss: who the boss's swings actually landed on, which tank cooldowns went up and when, what everyone was carrying at the pull, and the boss's own casts. Nearly all of it is evidence rather than judgement — it adds measurements the existing rules can use, and every one of them can be absent without the wipe reading as anything worse than "not measured". The single exception is "named on one signal" above: a tank death or an aggro pull in the last thirty seconds can move a verdict that was INCONCLUSIVE, and nothing else can.

• Threat. The boss's target on every swing before the collapse. On a fight where the boss is supposed to stay on a tank, a swing that lands on someone else is a pull — "Franz pulled aggro at 2:10 and died to Melee 3s later", or, if they survived it, "held the boss for 4s — survived it." A fight with nobody tanking the boss at all is never read this way; there is no tank line to have left.
• Tank cooldowns. What a tank cast in the fifteen seconds before they died, set against what actually hit them and what healing landed in that window. A big hit landing with nothing pressed reads coral; a cooldown pressed and already expired reads amber.
• Consumables, at the pull. Every raider's buffs the instant the pull started. This is the only place the tool can see a flask, a weapon oil, or a well-fed buff — none of it is visible anywhere else in the log.
• The boss's own casts, and the phase. Used to place the wipe in the fight (a kicker like "WIPED AT 41% IN 'BEAMS AND SHRIEKS'"), to measure the enrage line honestly against the pull's length, and to open the short windows where a boss ability is expected to pull the boss off a tank on its own. It is also what names a kill the deaths table leaves blank: Azgalor's Doom kills twenty seconds after it lands, so a death carrying no ability at that distance from a Doom cast reads "Doom (Azgalor)" instead of "something the log does not name".

Every percentage printed anywhere on this page or the card — a wipe's own kicker, a role's share, the throughput margin — is rounded to the nearest whole number, page-wide. 99.51% reads as 100%; that is the rounding, not a different measurement. The wall is the one exception: its three percentages keep one decimal each, on purpose — three whole numbers can read as identical when the pulls that produced them were not, and the wall's entire point is showing that they were not.

Stood in it, and who fell without Tears

Two more per-raider measurements ride on the same fight model, for the same fourteen bosses, and neither one is a position claim — the log does not carry where anyone stood on the ground, only what hit them.

Stood in it. Every hit from a boss's own avoidable damage — the abilities the mechanics table already marks a raider's own doing, like Doomfire or Rain of Fire — is totalled per raider across the pull: how many times it landed, how much damage it added up to, and how long the hits kept coming, counting only while two hits of the same ability landed no more than three seconds apart. That last number is a floor, not a duration — a raider hit at 0:20 and again at 0:29 by the same fire is credited with nothing in between, because the fire was not touching him then — so the sentence says so out loud rather than reading like a measured stretch. A hit a raider's own shield ate whole still counts as a hit — he stood in it, and the shield paying for it does not erase that — so the count and the seconds include it exactly as they would a hit that landed clean. The number in brackets is damage taken, which is a different question: where the log books a fully-shielded hit as an absorb of its own, the whole hit is in that total; where it books it as a zero-damage hit with the shield noted beside it, the total adds nothing for it. The card names only the worst-off raider, and only past six seconds of it, the ability spelled the same capitalised way the death list already spells it: "Shnipa stood in Doomfire for at least 96 s across 40 hits (52,673) before dying to it at 3:31." That sentence is four numbers off the damage-taken stream and nothing else. It does not say where Shnipa was standing — the log has no position for him, and this page will not invent one either. On the card itself the same measurement runs wider than one sentence: a STOOD IN IT strip under the tank-cooldown one names the worst eight raiders by seconds, with everyone else folded under a tap — the same top-eight-then-<details> shape the consumables roll-call already uses below, so a busy pull does not spill twenty names onto a phone. A raider measured at zero seconds — one tick and no more — never appears on the strip at all: its own kicker is a claim of presence, and one tick does not prove that.

Who fell without Tears. Archimonde launches raiders off the platform, and the guild's own answer to the fall is Tears of the Goddess — an item click, open to any raider, that briefly slows falling speed. A raider counts as having fallen whether Falling is the death's own named cause or only what finished it — a raider whose own row already reads "finished by Falling" after a heavier hit is checked for the click exactly like one the log names for Falling outright. That raider is checked against their own latest click in the five seconds before landing, and two or more falls that found no click in that window print one amber line: "2 fell to Air Burst without pressing Tears of the Goddess: <name> at <time>, <name> at <time>." One raider missing the click is an accident, and naming him would be this tool's oldest failure — a single death read as a pattern; two is the raid not clicking it. The same check leaves a short note on that raider's own row in the death list, right beside the fall it answers: "— fell with Tears pressed at 2:40" when a click was found, "— fell without Tears" when none was.

Elune's Embrace, the cast behind Tears, is read off the same log stream as the tank cooldowns above, but it is not one, and the reference file that carries it says so: it belongs to no class, because an item any raider can carry belongs to no class and no race, and it is never listed among a tank's unused cooldowns for the same reason. Its cast was never provably captured on any wipe until this report started asking Warcraft Logs for it; that it appears by name in three separate nights' own logs is the corroboration this page can offer in its place.

The consumables roll-call

Per raider, at the pull, judged against that raider's own line in the guild's consumables guide — not one rule for the whole raid. The guide asks different things of different specs: a fury warrior is buffed with a flask or one elixir, a holy paladin's line names a flask and no elixir at all, a mage's names elixirs and no flask. Any flask meets the flask half, whichever one the guide happens to name. The elixir half is the items the guide actually lists, read by the slot they occupy — the game allows one battle elixir and one guardian elixir at a time, so two lines naming battle elixirs are a choice between them rather than a pair, and a spec whose guide names both slots wants one of each. A spec the guide names no flask and no elixir for is not checked. Well Fed is checked for everyone, and the main-hand weapon buff only for the specs whose guide names an oil or a stone — a protection warrior is never checked for one, because their guide names none. One amber bullet names who lacked what, by name, because the raid leader will ask: "Unbuffed at the pull — no flask or elixir: …; half a pair: …; no food: …; no weapon buff: …." It never accuses: a missing flask is a smaller margin, not a cause the log can prove.

The flask column has five reads. ✓ is buffed by the book. ½ is half a pair — one slot met and the other either empty or holding an elixir this spec's guide line does not name. ✗ is nothing recognised at all. ? is a flask or elixir this site's list has not seen. — is a spec the guide asks nothing of. Tapping a raider's name opens the row underneath it: Has is what the snapshot shows they were carrying, Guide is their spec's line word for word, and Missing is what is still wanted — "also: …" when the elixir slot is empty, "instead: …" when it already holds something that does nothing for that line, so the raider can tell whether to add one or swap one. A raider whose spec is asked only for a flask, and who drank elixirs instead, reads ½ with "instead: a flask".

If the pull has no pull-start snapshot at all — an older log without advanced combat logging turned on — there is no table to show: the whole roll-call section is skipped, and one amber line reads "Consumables could not be read: <reason>." That is not the same as "no flask"; the report has no idea what anyone was carrying, and says so rather than guessing at a table it cannot build.

Inside a readable table, a raider can still show a "?" in the flask cell. That is a raider clearly buffed by something — a flask or elixir aura is there — that this site's own list has not seen before; hovering the "?" names the actual buff. A raider like that is never counted as missing a flask, and the section carries its own note, "N on an elixir this tool does not list," so the gap is visible without being read as a miss. A short ignore list keeps the things that merely look the part out of that count — a Noggenfogger Elixir, a Teleport: Shattrath — and the recognised list itself grows the next time a real buff turns up in a captured log.

The night report card

A collapsed table above the wipe list, closed by default, one row per raider across every pull the night's judged verdicts carry findings for. It reads nothing new — every column is the same per-pull measurements described above, added up across the night — so it can carry the wipe list's own signals but nothing an accusing verdict, a roll-call row, or a tank-cooldown bullet has not already said once.

PULLS is how many of the night's counted pulls that raider showed up on at all, out of the total, dimmed once that is fewer than half of them — a small sample worth discounting, not a clean bill. Presence is read off any of four things: a roll-call row, a death, a stood-in-it measurement, or a threat finding — a raider who never died and never carried a problem column can still be present by roll-call or by a standing measurement of zero seconds. FLASK MISSES counts the pulls where that raider's own roll-call read a ✗ or a ½ — short of buffed, in the same reads the table above uses. STOOD IN IT is their own avoidable-damage seconds, summed across every pull that measured any. AGGRO counts the pulls where a threat finding shows that raider pulling the boss off the tanks. EARLY DEATHS is their own pre-collapse deaths, with the same floor every tally on this page already applies — a battle-rezzed raider dying twice counts once, and a wipe-protection save counts for nobody. FELL, NO TEARS is Archimonde falls of theirs that found no Tears of the Goddess click. A pull that never became a fight (RESET) or fell over at the pull (PULL COLLAPSE) is left out of every column, the same exclusion the night's own wipe count applies.

Tapping a raider's name filters the wipe list below to the pulls that carried a non-zero problem column for them — never just the pulls they were present for, which would turn "tap a name" into "show me a night this raider did nothing on". Tap the same name again to clear the filter.

What it cannot see, and will not pretend to

The log carries deaths, timings and killing blows, and, on Black Temple and Hyjal, tank cooldowns and the boss's own target. It never carries positioning or healing assignments, on any raid. So no verdict here quotes a spread distance in yards or tells a healer they should have topped someone off — those would be invented, and an invented number is indistinguishable from a real one once it is on the screen. Every accusing verdict says this out loud, in its own evidence list.

Outside Black Temple and Hyjal the four new streams above have nothing to read from — no fight model exists for an older raid, so those wipes read exactly as they did before. Even on a modelled boss, every stream is read one page at a time: a pull long enough to fill it would show a truncated notice rather than a silently short list, though nothing this site has captured has come close. And a phase the fight card names for a boss's own stages (Reliquary's essences, Illidan's flame phase) is shown on the card without ever being used to say where a wipe happened — only a phase the model can actually locate does that.

Treat a verdict as where to look, not as a ruling. It is a starting point for watching the video back.

Why the report link is not in the URL

Every other tool here puts the report code in the address bar so a link reopens what you were looking at. This one does not. A gear audit and an accusation are different things to hand around, and a link that reproduces a verdict about named people is a product decision we did not make lightly — so the code stays in the box you pasted it into, and /wipe is the whole address.

Ads, and what they pay for

There are two ad slots: one in the side rail on wide screens, one below the results. They are never put between a raider and their data, they hold their space so nothing jumps under your cursor, and the side one disappears on narrow screens rather than squashing the table. All of that describes what happens once an ad unit is approved. None is, so today neither slot is filled and neither one takes up any space at all.

They pay the bills. Simulating a full raid is a couple of minutes of computing time per report, and that costs real money. There is no paywall and no account, and nothing about the audit changes depending on whether an ad loads — block them and the tool works exactly the same.

Credit where it is due

The audits, the checks, and every choice about how a number is read, labelled and shown are this site's own. The raw material underneath has earned its credit: every dps number here is produced by wowsims' simulator — their simulator, their rotations, their talent builds — and the item database comes from them too. It is open source under the MIT licence, which is what makes a tool like this possible at all.

The best-in-slot lists and item scores are wowtbc.gg's. Where their score and their own pick disagree, this tool goes with the pick — they chose that item knowing what it does.

Game data and item art come from Blizzard Entertainment's official API. World of Warcraft and Burning Crusade Classic are trademarks of Blizzard Entertainment. This site is not affiliated with or endorsed by Blizzard, wowsims or wowtbc.gg.

Privacy

No account and no user database; the server keeps one line per request for 14 days, plus a cache of the public Warcraft Logs responses it fetches. The privacy page spells out exactly what is recorded and what is not.

← How this site works