Verdict

A (95). Nothing in this key needs fixing at the level of decisions -- you covered 84% of the damage that needed an outside heal, cleansed everything reachable, and nobody died. The two things left are both habits carried over from the Altar key an hour earlier: Holy Power overcap at 19%, and 9% of your combat time spent doing nothing.

Analysis

What you did well

• **Throughput 100.** Of the 164,660,898 damage that needed an outside heal -- that is, after removing what players self-healed and what was overkill on an already-dead target -- you covered **84%**, plus 10,070,141 absorbed. That is the top of the scale.
• **Dispel 5 of 5.** Every damaging dispellable debuff that landed came off. Nothing leaked.
• **Zero deaths, and the timer agrees.** 23:43 against a 30:00 par -- 20.9% under, a two-chest key.
• **Response 88.** In the twenty worst eight-second stretches of party damage, you put back 50% of the 94,330,271 that landed. Up from 46% in Altar of Fangs.

What was outside your control

• **Living Bomb, Earthbound's Imprint and Cinderbolt chose their targets.** 42, 38 and 7 hits on you respectively, all excused -- these are casts that named you, not places you stood. Your `mechanics` is 100.
• **Inferno was 15% of everything the party took** (511 hits, 40.3M, all five players). Excavating Blast, Thunderous Stomp and Chillstorm add another 12%. Over a quarter of the party's damage intake in this key landed on everyone at once and is not attributable to anybody's positioning.
• **The group's interrupts were the weak axis, not yours.** 117 of 176 stoppable casts stopped, 66% -- Cinderbolt alone got through 48 times of 100. You have no interrupt on the bar, so this is not scored against you, but every leaked Cinderbolt is damage that arrives at your healing.

Two habits to carry out of this key

**Holy Power overcap, 19%.** You generated 691 and lost 130 of it to a full bar. It was 20% in Altar of Fangs the same night, which makes it a sequencing habit rather than a bad key -- roughly forty spenders across the two runs that were generated and never paid out. This is the single largest pure-throughput leak on your sheet and it costs nothing to fix but attention.

**Activity 86 -- 91% engaged.** You were idle for about 9% of the time you were alive and in combat, against 98% engaged in Altar of Fangs. On a 22:29 combat clock that is roughly two minutes of nothing. Some of that is real -- this key had 6 pulls against Altar's 10, so there is more downtime between packs -- but the axis is measured only while in combat, so it is not counting the walking.

**Utility 79 -- 42 used of the ~75 the kit offered.** Not a fault at this grade, but it is the axis with the most room left. Arbiter counts what your talents could have produced given their measured cooldowns.

Compared with the Altar of Fangs +12 an hour earlier

Same key level, same night, same character; a useful controlled pair.

  axis          Altar +12    Ruby +12
  throughput       92          100
  response         77           88
  dispel           89          100
  activity         98           86
  overall          91           95

The healing axes all moved up and activity moved down. The reason is in the deaths: Altar had two, Ruby had none, and healing that arrives after a player is dead counts for nothing on either `throughput` or `response`. The Altar analysis has the detail -- both deaths turned on where Holy Shock went during a burst window.

Rating went 2750 to 2776 across the pair.

What to change

1. Spend Holy Power before generating into a full bar -- 19% here, 20% in Altar, about forty lost spenders across the night.
2. Push the group on Cinderbolt interrupts: 48 of 100 leaked, and every one is damage that lands on you to heal.
3. Nothing else in this key needs changing. It is the reference run to compare the next Ruby Life Pools against.

Verdict

A (91). The key was never in danger of the timer -- you finished 26:41 with only 30 seconds lost to deaths. Neither death was a healing failure in the sense of you being too slow or out of mana; in both, the one instant heal you had went to yourself or the tank while a DPS was falling. Both dying players also pressed no defensive of their own.

Analysis

What was outside your control

Three things this key charged nobody for, so you should not read them as mistakes.

• **Paralyzing Shots and Triple Shot picked you.** Twinfang Harrower applied Paralyzing Shots to you at 0:27, 6:03, 6:30, 21:23, 21:46 and 22:10 -- 54 ticks of Triple Shot and 34 of Paralyzing Shots landed on you across the key. Arbiter excuses these under "the enemy chose the targets": a hostile cast that names a player is proof of the enemy's choice, not of your positioning. Your `mechanics` axis is a clean 100 as a result.
• **You take more damage than any DPS in the group, and that is the build working.** 29,077 per second, against 20,735 for Rolan, 24,933 for Ophelastra and 27,186 for Totums. Blessing of Dawn redirects a share of the party's damage onto you by design. Self-healing is therefore not a leak in general -- see the next section for the narrow case where it was.
• **A third of the party's damage landed on everyone at once.** Unstable Totem (22.9M across all five), Laced Edge (18.3M), Toxic Surge (12.5M), Synchronized Venom (11.1M). Arbiter does not grade these because unavoidable raid damage and a puddle all five stood in look identical in a combat log. They are listed so you know the tool saw them, not because they are anyone's fault.

What you did well

• **Dispels: 22 of the 24 you could physically reach.** Thirty dispellable debuffs landed. Arbiter scores you against what one dispeller could actually get to, not the raw count -- the other eight overlapped a debuff you were already cleansing, and no single player can be in two cleanses at once. 89 on that axis is close to the ceiling the run allowed.
• **Activity 98.** You were engaged 94% of every second you were alive and in combat. There is no idle time to recover here.
• **You survived a key that was actively hunting you** -- zero deaths while taking more damage per second than every damage dealer in the group.

The one thing to change: where Holy Shock goes in a burst

Both deaths share a pattern, and it is the only actionable thing in this key.

**Totums, 3:19.** He took 897,493 in 2.4 seconds -- Septic Spatter for 666,130, then Blood Sacrifice and Unstable Totem on top. He received **no healing at all** in that window. What you were doing:

  3:14   Eternal Flame  ->  Totums          (5 seconds before the burst)
  3:15   Judgment       ->  Primal Serpent  (damage)
  3:18   Holy Shock     ->  yourself
  3:19   Beacon of Virtue -> Ophelastra     (the instant he died)

You had touched him. Then the burst started and your instant went to your own health bar.

**Rolan, 19:24.** 1,546,614 over 4.7 seconds; he received 100,666, almost certainly Light of Dawn splash rather than anything aimed at him.

  19:20  Light of Dawn     ->  (AoE)
  19:21  Beacon of Virtue  ->  yourself
  19:22  Divine Protection ->  yourself
  19:23  Hex Lord's Doom   ->  (trinket)
  19:24  Holy Shock        ->  Doryhunky (the tank)

Four and a half seconds is a long window. Nothing targeted went to the man dying in it.

**The honest version of this finding:** you are the second-most-damaged player in the group, so self-healing is usually correct, and the tank at 90,224 per second is a real claim on your attention. The narrow thing worth changing is **Holy Shock specifically** -- it is your only instant, it is the only tool that answers a 2.4-second burst, and in both deaths it went somewhere that was not going to die. Everything else in those timelines is defensible.

**And it was not only you.** Both players pressed **no personal defensive** in their own death windows. A 897k burst in 2.4 seconds on an Elemental Shaman who used nothing is not a heal that was available to be cast.

Holy Power: one press in five is thrown away

You generated 817 Holy Power and lost **165 of it to overcap** -- 20%. Overcap means the resource was generated while you were already at the cap, so it evaporated: the generator was pressed but paid nothing.

This is the largest pure-throughput leak on the sheet and it is not a reaction-time problem, it is a sequencing one. It was 19% in Ruby Life Pools too, so it is a habit rather than a bad key. Roughly forty extra Holy Power spenders were available to you across these two runs and were not cast.

(Refused presses -- the ones the global cooldown ate -- are **not** part of this and are not a fault. They cost nothing. Only the generated-and-lost Holy Power is real.)

The Idol is finally being used correctly -- this is the good news

Hex Lord's Dooming Idol was pressed **13 times in this key and 12 in Ruby Life Pools**, at a median of **18 stacks**, and the ramp hit its 30-stack cap and wiped unspent **zero times in either key**.

That is the whole game with this trinket. It is throughput-linear -- each stack is worth 39 Intellect for 30 seconds no matter when you cash it, so pressing at 18 stacks and pressing at 30 average out to the same +195 Intellect. Stack count at press is a shape dial, not a throughput dial. The **only** way to actually lose value is to let the ramp touch 30, where the whole pile is destroyed and re-ramps from one.

Zero resets means zero loss. For comparison, the three keys on 2026-08-21 -- same trinket, before the stack bar existed -- captured 38%, 63% and 8% of what was available, and one Altar key reset at cap five times with three of its four presses cashing a single stack.

The TollPips stack bar added on 2026-08-22 is what changed, and these two keys are the evidence that it worked. Nothing to fix here; it is worth knowing because it means the Holy Power overcap in the previous section is now the *only* throughput leak left on the sheet.

Response 77: what the number actually says

A "worst window" is one of the twenty eight-second stretches in which the party took the most damage. `response` asks what share of the damage that landed in those windows you put back as healing.

You put back **46% of the 92,345,692** that landed in them. In Ruby Life Pools the same night you put back 50% and scored 88. The scale is absolute -- it is not a comparison against the other players -- so this is a real difference between the two keys, and the gap is roughly the two deaths: healing that arrives after someone is dead is not counted, and neither is healing aimed at the wrong person.

Fixing the Holy Shock habit in the section above is the same fix as raising this number. They are the same behaviour measured two ways.

What the group leaked

For context rather than blame -- none of this is scored against you.

• **Interrupts: 84 of 95 stoppable casts stopped, 88%.** Strong. Envenom got through 8 times of 44 and Mass Envenom twice of 15.
• **Envenom left up seven times, 53 seconds in total.** This is the dispel that mattered; it is a sustained damage debuff and every second of it is damage you then have to heal.
• Both deaths were on trash, not bosses, and 77% of the party's damage was taken on trash. Pull 9 -- the 65-enemy pull at 17:55 -- ran 5:28 and took 97.7M by itself.

A number in the table above is inflated -- do not quote it

The "dispellable debuffs left up" line reads:

  Paralyzing Shots   1x left up,   10s total,   8,181,300 damage

That 8,181,300 is **not** what the ten-second leak cost. It is Paralyzing Shots' total damage across the entire key, to every player who ever had a counted instance of it. Verified by summing the per-player intake: 3,659,553 (you) + 3,213,752 (Totums) + 1,093,920 (Rolan) + 214,075 (Ophelastra) = 8,181,300 exactly.

The same inflation applies to the summary line "14,698,097 damage came from the ones left up". Both figures come from `derive.py`'s `hurt[name]`, which sums a spell's whole-key damage rather than the damage inside the windows it was actually left up.

The real cost of that one leak is roughly ten seconds of ticks on one player -- on the order of a million, not eight. The dispel *grade* is unaffected; only these two damage figures are wrong.

What to change

1. In a burst window, Holy Shock goes to the player who is dropping -- not to yourself and not to the tank. It is the only instant you have and it was misallocated in both deaths.
2. Spend Holy Power before generating into a full bar. 20% overcap here, 19% in Ruby -- about forty lost spenders across the night.
3. Ask Totums and Rolan to press something. Neither used a personal defensive in their own death window.
4. Fix derive.py's dispel-leak damage: hurt[name] sums the spell's whole-key damage instead of the damage during the left-up windows, which overstates every 'damage left up' figure the report prints.

Verdict

The key timed: 27:29 against a par the log bounds to 32:34-43:26. Nine deaths, a wipe on Lithiel Cinderfury, and 3:16 of the 27:29 lost to pulls - 2:56 of it to that wipe.

WHOSE FAULT: not the party's, and this analysis previously said otherwise. The wipe at 22:07-22:19 happened because extra mobs were pulled that are not supposed to be pullable during that phase. The log agrees. Across 21:37-22:20 the damage came from five distinct sources at once - Wild Imp (976 damage events), Defiled Golem (305), Lithiel Cinderfury (102), Infernal (53) and Shivan Punisher (21). That is a boss plus a swarm plus two extra elites, and it killed five people in twelve seconds. An earlier version of this note made 'four of the five pressed no defensive' the headline correctable. It was the wrong reading: the pull was already lost, and a Shield Wall does not change that.

The defensive habit is still worth naming, but on the OTHER cluster. At 3:08-3:47 a Street Sneak pack killed three people in thirty-nine seconds - a survivable pull - and Diddymytomes and Zydrae both died with nothing pressed while Heartstop Poison sat uncleansed for twenty seconds.

GRADES ON THIS KEY ARE NOT COMPARABLE TO ANYTHING GRADED BEFORE TODAY. Four model changes landed because of this run, three of them fixing things that were marking players down for the tool's blind spots. Last section.

Analysis

Where the 27:29 went

21:46 fighting (81%), 3:07 routing, 1:48 wipe recovery, 0:45 death penalty, 0:10 pull timer. Ten pulls, 158 enemies, 258,846,377 damage taken.

  Kystia Manaheart          2:03
  Zaen Bladesorrow          2:00
  Xathuux the Annihilator   2:02
  Lithiel Cinderfury        0:43   WIPE
  Lithiel Cinderfury        2:45   kill

Bosses took 46% of the damage and six of the nine deaths.

The group interrupt figure reads 32% and should not be read. 174 of the 210 'stoppable' casts in this key are Felfire Burst, cast 251 times by Wild Imps and Unleashed Imps - disposable adds a party AoEs rather than kicks. On everything else the group went 33 of 36: 92%. The three that leaked were two Fel Rages and one Scathing Review.

Dispels were 19 of the 21 one presser could reach. The two left up were Heartstop Poison, 20 seconds and 4,080,270 - the same debuff in the 3:08 cluster.

Ryco - Prot Warrior, C (66), up from a D that was the tool's fault

WHAT WAS NOT YOURS

• The D. Mitigation scored 12 on '10% uptime of the 18% this kit can hold'. That 18% was Shield Wall alone - 8.0s over a 43.4s recharge - because Shield Block and Ignore Pain were missing from the grader's defensive table entirely. Measured off the log you held Shield Block for 951.6s (59.0%) and Ignore Pain for 896.8s (55.6%). With both counted the axis reads 75: 86% uptime of the 100% the kit can hold, 88% of worst windows covered. The grade went D (53) to C (66).
• Felfire Burst, 76 hits against a party median of 12.5, was your entire over-exposure line and is now excused. The imps name their target in the cast and they named you, because you are the tank. Mechanics 92 to 100.

WHAT YOU DID WELL

• Activity 92 - 98% engaged while alive, highest in the party.
• Rage: 108 wasted of 15,062 generated. 1%.
• 24 interrupts by raw count, the most in the party, plus Disrupting Shout and Shockwave.
• 86% mitigation uptime is a tank playing the spec properly, and nothing in the report said so until today.

WHAT TO CHANGE

1. Damage 44 - 105,083 dps, 55% of the party average - is now the softest number in your line, and at ilvl 294 in a party with a 312 mage some of that is gear.
2. Intimidating Shout is in your loadout and was never pressed. On a key with an imp swarm that is a real AoE stop.
3. Your first death at 3:38 took 3,403,043 over 7.5 seconds and you did press Shield Wall; the party put 1,783,722 into you and it was not enough. That one is the pack, not you.

Hyporock - Holy Paladin, A- (86)

WHAT WAS NOT YOURS

Cutpurse, and you were right to push back on it. 71 hits for 3,121,092 against a party median of 33.5 was your whole over-exposure line, and the tooltip settles it: Effect #1 is a Jump Charge with 100 yards of range - the mob travels to whoever it picked. The log agrees independently, because the cast names its victim: `Row Hooligan -> Ryco`, then Zydrae, then you twice. Where anybody stood is irrelevant. Mechanics 83 to 100, and a new rule now excuses any spell whose caster named the player.

That also retires the finding from the Altar of Fangs key that this was 'the second key running where the measurable thing about your positioning is melee range'. One of the two was never positioning.

WHAT YOU DID WELL

• Dispels 19 of 21 reachable; utility 100 with 36 uses of the ~71 the kit offered - seven Blessings of Sacrifice, seven Aura Masteries, four Blessings of Protection, three Lay on Hands.
• Response 85: you put back 49% of the 77,782,694 that landed in the party's worst windows, the highest of the twelve keys measured on the rebuilt axis.
• Survival 86 on a key where the party died nine times.

WHAT TO CHANGE

1. Throughput 67 - you covered 70% of the 191,923,620 that needed an outside heal, where the scale wants about 82% for a 100. This is the axis with the most room in your line and the one worth watching.
2. You died at 22:17 having taken 2,054,129 over 12.5 seconds with nothing pressed. The pull was lost regardless, but Divine Shield was cast zero times all key and twelve and a half seconds is a long decline.
3. Holy Power: 100 wasted of 654 generated (15%).

Diddymytomes - Arcane Mage, B- (76)

WHAT YOU DID WELL

• Damage 100: 219,848 dps, 124% of the median, at 45.8 casts a minute. Comfortably the best damage in the key.
• Mechanics 100 - nothing avoidable, on both the old model and the new one.
• 16 Counterspells, 37 Shimmers, 10 Alter Times.

WHAT TO CHANGE

1. Survival 39 is the worst score on the board: three deaths, more than anyone. 3:08 (1,813,244 over 2.2s, nothing pressed), 22:07 (1,052,215 over 6.9s, nothing pressed), 25:41 (1,455,675 over 10.3s, Prismatic Barrier pressed).
2. Ice Block was never pressed in the whole key. The 3:08 death is the one that matters - that pull was survivable and two other people died in the same thirty-nine seconds.
3. Mana floor 13%, the only player near the bottom.

Zydrae - Unholy DK, C (66)

WHAT IS STILL CHARGED, AND WHY

Zydrae is the only player left with over-exposure on this key: 2,010,849. Chaos Barrage, 19 hits against a party median of 9, is the real one - and note that Ryco's 20 hits of the same spell WERE excused, because the casts named him and did not name you. That is the new rule working as intended: the mob picked the tank, and the damage you took from it was not aimed at you.

Murder in a Row, 16 hits for 841,180 and '(peers 0)', I checked by hand: nobody else took a single tick. That is the genuinely ambiguous case - a bleed on exactly one player reads identically whether a mob leapt on you or you stood somewhere alone. The axis attributes half rather than all for that reason. Unproven, not a finding.

WHAT YOU DID WELL

• 21 interrupts off 27 Mind Freezes - a 78% hit rate on a 15-second cooldown, second only to the tank.
• 182,009 dps, exactly on the median, at ilvl 300 - the lowest-geared damage dealer matching the middle.
• 52.7 casts a minute, the highest in the key.
• The only person in the wipe who pressed a defensive.

WHAT TO CHANGE

1. Anti-Magic Zone is in your loadout and was never pressed. It would not have saved the over-pull, but the 3:08 cluster and the Xathuux fight are both AMZ situations.
2. Runic Power: 483 wasted of 8,076 (6%), the highest waste fraction in the party.

Nexein - Shadow Priest, B (79)

WHAT YOU DID WELL

• Mechanics 100 and survival 86 - nothing avoidable, one death, in the wipe.
• The '32 stops vs 14 expected' line is worth unpacking: only 8 are kicks. The other 24 are crowd control that stopped a cast, which the model counts and which is the correct answer to an imp swarm. 28 CC casts, five times anyone else.
• 166,836 dps at ilvl 291 - the lowest item level in the party - plus 8,858 HPS of off-healing and 24 Power Infusions.

WHAT TO CHANGE

1. Desperate Prayer was never pressed, and you died at 22:12 having taken 825,978 over 6.0 seconds with 10,930 of healing received - effectively none. Again, that pull was lost anyway.
2. Mass Dispel is in your loadout and unused, on a key with two Heartstop Poisons up for 20 seconds and 4,080,270 damage. You are the second dispeller the healer did not have. (Purify Disease is probably vacuous - check the dungeon's dispel types before treating it as a finding.)
3. Your off-healing is now visible to the grader in a way it was not before: it used to sit in the HEALER's denominator, so every point you healed lowered Hyporock's throughput score.

What this key changed in the grader

Four changes, and three of them were the tool marking somebody down for its own gaps.

1. SHIELD BLOCK AND IGNORE PAIN WERE NOT IN THE DEFENSIVE TABLE. `achievable_uptime` iterates `PERSONAL_DEFENSIVES` for BOTH halves of its ratio, so a Protection Warrior was scored on Shield Wall alone. Every tank in the corpus before this key was a Blood DK or a Vengeance DH, whose main defensives were already listed - which is why two months of keys never exposed it. Ryco: mitigation 12 to 75, D to C.

2. A CAST THAT NAMES ITS VICTIM IS NOT ABOUT POSITION. `forced_single_target` needs one victim and two casters; `forced_multi_target` needs simultaneous applications. Cutpurse has neither - it lands on one player per cast and spreads across all five. But the enemy names the player in the cast, and a ground effect names nobody, so the distinction is already in the log. Kept per (spell, victim) rather than per spell, so a nuke aimed at one player that splashes the people standing next to them still charges the bystanders.

   The cost is that the mechanics axis is now close to saturated: 25 of 30 player-keys on the 2026-08-25 night score 99 or better, against 19 before today. It is measuring something narrower and truer - ground effects, persistent areas, environmental hazards and melee, the things that log no player destination because there was nobody to name - but it needs re-banding before the number means much.

3. THE HEALER'S DENOMINATOR STILL COUNTED OTHER PEOPLE'S HEALING. `healing_need` subtracts self-healing, because that is health the healer did not have to restore. A Shadow Priest healing the TANK is the identical claim and stayed in for another two months. On this key the four non-healers put out about 24k HPS between them, so a healer who covered everything actually left to cover could not have scored above roughly 84%.

4. `RESPONSE` WAS AN EASINESS DETECTOR POINTED BACKWARDS. It divided the healer's output in the party's worst windows by the healer's OWN average output. Both ends move with the key: on a hard key the healer is near-saturated the whole time, the baseline rises, and the ratio compresses toward 1.0 however well they played.

   Across 23 healer-keys the old ratio ran r = -0.63 against the party's damage-per-second, -0.50 against key level, and -0.49 against the healer's own HPS - it went DOWN the more they healed - while managing r = -0.09 against deaths. An axis named for rising to emergencies had no measurable relationship with anybody dying, and it carried 17% of the grade. The RESPONSE_MIN_CONTRAST guard added last night was this same bug caught in its most extreme form.

   It now uses an absolute denominator: of the damage that landed in the party's worst windows, how much did this healer put back. Across two independent batches its correlation with difficulty is +0.07 and -0.03, against -0.25 and -0.30 for the old one, and it separates keys more sharply. Deliberately NOT claimed: that it predicts deaths better - it ran -0.42 on one batch and +0.92 on another, which at n=8 and n=4 means nothing.

   Both healer bands remain cut on ONE healer's range, 23 keys of Hyporock. That is the largest open weakness in the healer half of this model and no amount of internal consistency fixes it.

What to change

1. Diddymytomes: press Ice Block. Zero casts across three deaths, and the 3:08 one was on a survivable pull.
2. Zydrae: Anti-Magic Zone was never pressed. Not for the over-pull, but the 3:08 cluster and Xathuux both wanted it.
3. Nexein: Mass Dispel on the Heartstop Poisons - 20 seconds up and 4,080,270 damage, on a key where the healer could only reach 19 of 21.
4. Group: the 32% interrupt figure is 139 leaked imp casts. On everything that was not the swarm you went 33 of 36. Do not chase that number.
5. Arbiter: re-band `mechanics` - 25 of 30 player-keys now score 99+, so it no longer discriminates. Same remedy as mitigation and dispel needed before it.
6. Arbiter: both healer bands are cut on 23 keys of one healer. Feed a second healer through before trusting throughput or response across players.

Verdict

The key was never in trouble. 25:53 against a par the log bounds to 31:02-33:25, so it cleared with five to seven minutes spare. All three deaths landed after 19:42, on the last two pulls, and the entire death penalty was 15 seconds of a 25:53 run. Nobody lost this key and nobody came close to losing it.

This run was graded twice. The first pass gave Hyporock a C+ (70.8), and 3,474,194 of their 4,512,475 of 'over-exposure' - damage the axis says comparable players did not take - was Rav'i's Triple Shot. That is a debuff the boss applies by name to three of the five players the instant the cast finishes, and it then ticks for ten seconds. There is nothing to stand out of and nothing to dodge. It went out six times and picked Hyporock all six.

The two Altar of Fangs +11s of the night before settle it: Triple Shot picked them twice out of six in each, their mechanics axis read 95 and 94 against this key's 54, and their over-exposure was 385,031 and 621,072 against 4,512,475. Nothing about how they played changed by a factor of seven.

So the grader was fixed rather than the number explained away - `derive.forced_multi_target`, below. Regraded, the mechanics axis reads 89 instead of 54 and the grade is B- (75.9). That is the grade of record; the C+ is gone.

Ryco's A- was honest on both passes, and the interrupt half of it is the best single performance in the key.

Analysis

Where the 25:53 went

23:38 fighting (92%), 1:48 walking, 0:15 death penalty, 0:19 pull timer. Thirteen pulls, 176 enemies, 308,300,201 damage taken by the party.

  Rav'i              2:52
  The Writhing Coil  2:44
  Zul'jan            3:18

Bosses accounted for 22% of the damage taken and two of the three deaths; trash took the other 78%. The two pulls that cost time were the last two: pull 12 at 19:11 (25 enemies, one death) and pull 13 at 22:27 (Zul'jan, two deaths).

The group stopped 72 of 85 stoppable casts (85%). Envenom leaked 12 times of 53 and Mass Envenom once of 7 - 'leaked' meaning a cast that could have been kicked and was not.

Hyporock - B- (75.9)

WHAT WAS NOT YOURS

Triple Shot: 64 ticks for 5,170,894, from six selections you had no input into. The axis no longer charges it, and the report now names it under 'not charged, the enemy chose the targets' rather than dropping it silently.

Unstable Totem (5,613,157 over 113 ticks) and Bloodletting (1,086,007) log with no source and hit the whole party at roughly equal rates - Ryco took 5,744,583 and 972,355 of the same two. Neither is attributable to anyone.

WHAT YOU DID WELL

• Dispels, 22 of the 22 a single presser could reach out of 33 that landed. Three of those 22 are Blessing of Freedom taking Paralyzing Shots off, which the grader could not see until this pass - Freedom removes a debuff without ever logging a dispel, so all three read as debuffs left up.
• Utility 94 - 30 uses against the ~77 presses your kit offered across the key.
• The rez chain on Zul'jan. You died at 23:10; Letmetaunt had Raise Ally on you by 23:13. Sassme died at 24:39 and you had Intercession on them by 24:41. Two battle rezzes on the last boss, both inside three seconds, and the key still timed.

WHAT TO CHANGE

1. Paralyzing Shots is yours alone, and it is the steep one. Envenom is a Poison, so Kloe's Remove Corruption reaches it - and Kloe took 5 of the 12 Envenom cleanses to your 7. Paralyzing Shots is Magic; a Balance druid cannot touch it. Two were left up for ten seconds in total and cost 4,750,266. That is 475,027 per second of uptime against Envenom's 93,339. When both are up, take the Shots and leave the Poison to the druid. Freedom is a second answer to it and you already use it that way.

2. Zul'jan's melee killed you, and only you and the tank were in it. Axegrinder landed on exactly two people all fight: Letmetaunt nine times, you three. It was the largest single slice - 562,428 of 1,117,124 - of the 2.1-second burst that killed you at 23:10. You pressed Shield of the Righteous twice and Divine Protection inside that window and still died, and as the only healer there was no external to save you. The defensives were not the problem; the standing position was.

3. Throughput 64 means you covered 69% of the 191,887,943 that needed an outside heal - 41% of the party's intake was self-covered or overkill and never yours to catch. That is below your own archived median of about 77, but it is an ordinary bad-ish number rather than the outlier. The outlier was mechanics, and mechanics was Triple Shot.

Ryco - A- (84.1), and one thing that costs a life

WHAT YOU DID WELL

• Interrupts, 100. 21 stops against the 14 the model expects of the role, out of 72 the group landed - second only to Kloe's 22, and on a key where Envenom got through twelve times, the ones that landed were carrying weight.
• Mechanics 95: 1% of your intake was damage comparable players did not take. The single flag, Toxic Surge 20x against a party median of 9.5, is 524,281 out of a 41,604,579 intake. It is noise.
• 194,842 dps, on the median for the key, while doing the most interrupting.

WHAT TO CHANGE

1. You pressed nothing when you died. At 19:42 you took 959,820 over 2.3 seconds - Venom Burst 517,277, Toxic Beam 287,357, Toxic Surge 155,186 - and used no defensive at all. Die by the Sword was cast once in the entire key, at 25:35, so it was certainly available at 19:42. Rallying Cry, Spell Reflection and Storm Bolt are in your loadout and were not pressed once all run. Hyporock put 191,157 into you inside that window and it was nowhere near enough on its own.

2. Rage: 709 of 2,905 generated was lost to the cap - 24%, the highest waste fraction of the three DPS.

3. Activity 80 - you were engaged 92% of the time you were alive and in combat, against 97% for both Sassme and Kloe. That is roughly two minutes of standing still that the other two DPS did not have, and it is the cheapest of the three to fix.

What this key changed in the grader

Two fixes came out of this run, and both were things the tool could not see rather than things it weighed wrongly.

1. `derive.forced_multi_target`. Its sibling `forced_single_target` already excused a debuff the log proves an enemy chose a player for, but it demands one victim and two casters, and the commonest unavoidable debuff has neither - a boss picking k of the party per cast. The new rule wants three or more bursts of simultaneous applications whose target set MOVES: a frontal cone also stamps several players at one instant, and eating a cone is position, but a cone keeps hitting whoever stands in front while a selection rule spreads across the party. Triple Shot's six bursts of three covered all five.

   Where the check sits was half the fix. It was consulted only inside the `med == 0` branch, which a single-target debuff always reaches because every peer sits at zero - but a debuff dealt to three of five leaves the other two with a healthy median, so the player it favoured was charged the difference and never came near the test.

   Across the 30 player-keys of 2026-08-25 it moved three: Cameraop's Altar of Fangs mechanics 77 -> 100 (they were the six-of-six target that night, for 2,141,359), Ghoultard 70 -> 77, Doryhunky 98 -> 100. It is a narrow rule, not a general amnesty.

2. Blessing of Freedom counts as a removal. Freedom and Blessing of Protection take a debuff off without ever logging SPELL_DISPEL - grepping every log on disk for one returns zero rows - so the dispel scan, which is gated on that event, could not see them at all. They behave like an immunity: the aura just goes away, within milliseconds of the cast. Detected now as cast-then-removal, with the hostile-DEBUFF check and a 250 ms window doing the work; widened to a second the same scan starts collecting the target's own buffs expiring.

   It moved this key's dispels from 19/30 to 22/33. The three extra opportunities are the tell: those instances were being deleted by the four-second flicker filter, because Freedom ended them so fast and nothing had recorded a remover. That is the same shape as the fast-cleanse bug of 2026-08-22, arriving through a different door.

   Learned into a separate `breakable` table, never into `dispellable`. The two are different claims - a root only Freedom can take off is not something a cleanse reaches, and writing it there would start charging every later healer for not pressing a button they do not have.

What to change

1. Hyporock: when Paralyzing Shots and Envenom are both up, cleanse the Shots. It is Magic and Kloe cannot reach it, and it is 5x steeper per second of uptime than the Poison.
2. Hyporock: get out of Zul'jan's melee. Axegrinder touched nobody but you and the tank, and it was half the burst that killed you at 23:10.
3. Ryco: press one defensive when the health bar moves. Die by the Sword was available at your death and was used once in 25 minutes.
4. Arbiter: the other 21 archived keys were graded before both fixes and are not directly comparable to this one. Regrade the corpus before reading any trend across dates.

Verdict

The key ran 2:21 and contained exactly one pull. The group engaged eleven mobs at 0:20 — four Ravenous Descendants, five Primal Serpents, a Ritual Chieftain and a Twinfang Harrower — and killed one of them. Eleven deaths followed in fifty-nine seconds, and the log stops with nobody having re-entered combat.

It was much closer than the grades suggest. The pack held 95.9M health between the eleven; the group removed 72.3M of it. When the last player died a Primal Serpent was on 87,213 health — 1.2% — and two Ravenous Descendants were on 10.8% and 11.5%. Fourteen more seconds at the output the group was already producing at 0:40 would have cleared three of the eleven and broken the pull.

Those seconds went on the run-back instead. Four players were down at 1:11; the tank held alone for thirty-one seconds while the others released and returned one at a time into a fight nothing had reset, and six more deaths followed. Party output fell from 17.4M per ten seconds at its peak to 3.0M, and the pack moved from 28% remaining to 25% across the whole second half.

Hyporock's D is an artifact of two things, and the healing is neither of them. The party took 363k damage a second — twice the 183k/s of the Altar of Fangs +11 that timed three hours earlier the same night — and the `response` axis was broken on a key this short. Last two sections.

Analysis

The two minutes, in order

21:04:15 the key starts. 0:20 the group pulls; the mobs are already in log range at the key start, so this is the pack at the instance door.

0:25 — a Ritual Chieftain casts Unstable Totem. There is no SPELL_CAST_START for it anywhere in the log, so it is instant and cannot be kicked. It pulses for the rest of the fight: 134 ticks, 6,451,833 across the party, 19% of everything the group took, from that one cast.

0:46 — first death, Nostick, to Ravenous Claws.
1:05 — Hyporock: 952,276 over 1.6s (Toxic Breath 692,207 then Piercing Hiss 260,069), 53,495 healing received.
1:11 — Julpool, Nostick and Iengry die in the same second, taking the party to four down. The Twinfang Harrower dies in that same second: the only enemy killed all key.

1:11 to 1:42 — the tank is the only player alive. He takes 3,050,769 over the last 13.5s, with 1,659,467 healing arriving from Hyporock and Iengry, and dies at 1:42 with no defensive pressed in that window.

1:29, 1:30, 1:39, 1:44, 1:45 — five more deaths as players return. 1:53, the last damage event. 2:21, the log ends.

Nothing died, so the damage never stopped

This is the run in one number. The group dealt 74.1M into the pack and removed 72.3M of 95.9M health — 75%. One mob died.

What was still standing when the last player fell:

  Primal Serpent          87,213 of  7,075,486    1.2%
  Ravenous Descendant    933,334 of  8,647,816   10.8%
  Ravenous Descendant    998,621 of  8,647,816   11.5%
  Ravenous Descendant  1,510,552 of  8,647,816   17.5%
  Primal Serpent       1,290,239 of  7,075,486   18.2%
  Ravenous Descendant  2,291,293 of  8,647,816   26.5%
  Primal Serpent       2,922,821 of  7,075,486   41.3%
  Ritual Chieftain     5,468,935 of 12,578,639   43.5%
  Primal Serpent       3,774,454 of  7,075,486   53.3%
  Primal Serpent       4,389,662 of  7,075,486   62.0%

Every mob left alive is another Piercing Hiss and another Fetid Spit, so a pull where nothing dies gets harder as it goes rather than easier. The Twinfang Harrower — 13.4M, the largest pool in the pack — was focused and killed. After it went down at 1:11 the damage spread, and nothing else was taken across the line.

Mob health here is measured rather than estimated: the advanced block on every SPELL_DAMAGE carries the target's current and maximum health.

Where the 34 million came from

Four sources are 72% of everything the party took:

  Ravenous Claws   7,068,092   (5,555,469 of it on the tank)
  Unstable Totem   6,451,833   (134 ticks, from ONE cast at 0:25)
  Piercing Hiss    5,925,558   (5 of 12 casts leaked)
  Fetid Spit       5,040,512

Unstable Totem is the one worth acting on, because it is the only one with a single point of failure. It is instant, so it cannot be interrupted; it belongs to the Ritual Chieftain, and it stops when the Chieftain does. The Chieftain finished on 5,468,935 health — about eight seconds of the group's peak output — while doing 19% of all incoming damage for every second it lived. It also landed two of the eleven killing blows.

Piercing Hiss is the interrupt line: seven stopped, five through, 5.9M, and five of the eleven killing blows including both of Hyporock's deaths.

One dispel finding, and it is small: Paralyzing Shots was left up once for seven seconds and 986,075 damage came out of it. Hyporock cleansed one of the two.

The run-back cost more than the wipe

There is no resurrection of any kind in this log — no SPELL_RESURRECT, no Raise Ally, no Rebirth, no Soulstone. The players who died are casting and dealing damage again within seconds: Nostick died at 21:05:01, cast Glide at 21:05:05, and then dealt damage on 231 events before dying again at 1:11. They released and ran back, which the pull's position at the instance door makes fast. (That four-second turnaround is the weakest number in this report and the one worth checking against what people remember; everything else in this section follows from timestamps alone.)

The tank did not die at 1:11, so nothing reset, and each returning player arrived alone into a pack of ten. Party damage dealt per ten seconds:

  0:30-0:40   12,310,258
  0:40-0:50   17,352,825   <- peak
  0:50-1:00   10,701,639
  1:00-1:10    9,612,911
  1:10-1:20    6,845,830   <- four players down
  1:20-1:30    3,565,021
  1:30-1:40    3,742,407
  1:40-1:50    3,008,384

The second half produced a fifth of peak output and six of the eleven deaths. At +11 the timer charges five seconds a death, so the eleven cost 55 seconds of a 2:21 key, and the six that came after the fight was already lost cost 30 of them.

The counterfactual is not free: had nobody run back, the tank dies alone and the pack resets to full. But the pack did not need to reset — it needed about fourteen seconds of output the group had already demonstrated, and returning one at a time never assembled that.

The healer's D is the axis and the intake, not the healing

Hyporock's measured output per ten seconds, effective healing with overheal beside it:

  0:20-0:30   1,328,162   (30% overheal)
  0:30-0:40   3,139,373   (33%)   <- 314k HPS, the party's worst window
  0:40-0:50   1,779,946   ( 1%)
  0:50-1:00   1,797,508   ( 2%)
  1:00-1:10     948,835   (25%)   <- dies at 1:05

11,245,985 effective across 93.7 seconds of combat is 120,021 HPS, against 91,687 HPS per combat second in the Altar of Fangs +11 that timed at 17:53 the same night. Overheal for the key was 21% against 49% in that timed run: almost nothing was wasted, because there was nowhere for a wasted heal to go.

Every major cooldown was spent, and spent early: Avenging Wrath 0:23, Blessing of Sacrifice 0:24, Divine Protection 0:26, Aura Mastery 0:35, Divine Shield 0:45, Lay on Hands 0:47. Aura Mastery at 0:35 lands inside the 0:30-0:40 window, the single worst ten seconds the party had. That is correct timing, not a panic dump.

So two axes here are describing the run rather than the player. `throughput` at 6 is coverage of 47%, against a band cut on keys where the median is 73%; at 363k/s incoming — twice the timed run's 183k/s — 47% is close to what a Holy Paladin at ceiling produces. `response` at 0 was a bug, now fixed, and removing it moved the grade from F (45) to D (56).

What is left that is actually Hyporock's: 80% activity and 33.6 casts a minute against 97% and 43.3 in the timed key, most of which is the ~25 seconds spent dead or running; Holy Power overcapped 12 of 41 generated (29%); and Blessing of Protection never pressed on a pull where four players were eaten by melee — though with 1:34 of combat in the whole key, one run is not evidence of a habit.

Two things this key changed in the tool

First, this run did not exist. `logfile.find_runs` closed a key only on a CHALLENGE_MODE_END or on the next START, so a key whose log simply stops — which is exactly what abandoning one looks like — was dropped from the listing entirely. It is always the LAST run in the file, which is always the one being asked about. Fixed, with a regression test.

Second, `response` scored 0 for a healer who opened at 314k HPS with 1% overheal. The axis is a ratio of the healing rate inside the party's worst windows to the rate across the whole fight, and `party_spikes` returns twenty eight-second windows — a ceiling of 160 seconds. On 93 seconds of combat the windows cover the fight end to end, numerator and denominator are the same seconds, and the ratio is pinned at 1.0 whatever the healer did. 1.0 sits below the band, so it scored zero. It now drops out under rule 3 and says why. Because 160 seconds is a hard ceiling the guard can only fire below 320 seconds of combat, so no completed key in the archive moved.

What would falsify this

Worth being explicit, because this is built entirely from one log:

• The four-second death-to-Glide turnaround is the weakest claim here. If the group had a combat resurrection the client did not log, the whole run-back section changes shape. Ask the people who were there.
• `affix 158` is unnamed. Xal'atath casts "Xal'atath's Bargain" (461870) at 0:04 and a Voidbound Emissary spawns at 1:33 and buffs the pack at 1:52 — after the last death at 1:45 — so the seasonal affix contributed nothing to this wipe. The label is printed as a raw id on purpose rather than guessed.
• No over-exposure was reported for anyone, so none of the "(peers 0)" traps apply here.
• The death count has no external cross-check: an abandoned key writes no CHALLENGE_MODE_END, so the usual arithmetic against the game's own timer is unavailable. Eleven is the raw UNIT_DIED count, all eleven flagged as real deaths rather than unconscious.

What to change

1. Kill the Ritual Chieftain. Its Unstable Totem is instant, unkickable, and did 19% of all damage taken in the key from a single cast at 0:25; it stops when the Chieftain does, and the Chieftain finished on 5.5M — about eight seconds of the group's peak output.
2. Kick Piercing Hiss. Five of twelve got through for 5.9M, and it landed five of the eleven killing blows, including both of Hyporock's deaths.
3. On a wipe, release together and come back together. Six of the eleven deaths were players arriving alone into a pack of ten while the tank was still holding it; they cost 30 seconds of timer and moved the pack three percentage points.
4. If the pull is going badly, finish something. Three mobs were under 12% when the last player died and a Primal Serpent was on 1.2%; every one that dies removes its own Piercing Hiss and Fetid Spit from the rest of the fight.