1. Green rounds — the basics

• When: early phases → with each Judgment of Eternity cast (skill 80629, ~4.4 s cast; markers appear ~1.36 s into it). Last phase → at +3 s, +33 s, +63 s (then +93 s…) after the 81071 channel starts. solid
• Markers are placed one by one, 80 ms apart ("slots"). solid
• Slot 1 = the Fixated player, at any distance. If nobody holds Fixated, slot 1 stays empty and the first marker lands ~80 ms late. solid (165/165 rounds with a fixated player had them first.)
• Count = min(3, floor(players up / 3)), downed/dead players don't count. solid — 338/338 rounds with a fixated player (excluding the fight-start round) match (2026-10-02 full set of 141 CM logs: 757/758; the 1 exception is a round 2.6 s before a wipe ended the log). Every short round in the user's own logs is explained by downs/deaths.
• Fight start (first round, ~3 s / ~6 s raw): always only the fixated player (48/49 logs). The others are still at the entrance ~4,400 units away. The exception (224233) had 6 players already hitting the boss and got 3. In CM players standing at the boss's feet at that moment (EI showed 4 units — actually unloaded positions) were not picked. solid
• Who gets the non-fixated greens: players near Vloxx are strongly favoured (in 76 picks: rank 1 = 24, rank 2 = 22, rank 3 = 14, rank 4 = 11, 5–7 = 5) but not strictly the closest — looks like a distance-weighted pick. likely
• At +63 s, the Probability Distribution (PD) targets of that same instant don't get greens (except the fixated player, who can get both). Seen in every +63 s round. This is about who, it does not explain the count. solid
• Failed green (not enough players stacked in the 240-radius circle) → skill 80378 "Judgment of Eternity (3-people green failed)" damage + 23296 (4,000 fixed) on everyone in it, and −3 Ascension stacks each. solid

2. Fixated

• Buff 34508, lasts 60 s. Removed early by stealth (every non-down/non-death early removal coincided to the ms with a stealth start, 16+ cases), down, death. solid
• Never applied during a split; a Fixated applied before a split carries through it. solid
• Boss phases: after Fixated ends, a new one is applied usually ~7.6–8.1 s later (EI logs, gaps not spanning a split: expiry median 7.6 s, 40/46 within 3–10 s). The second Fixated of every fight is at ~67–68 s (EI) / ~71 s (raw). solid
• After stealth the boss often misses a cycle: EI Spear phase — 8 of 14 stealth losses were re-applied >12 s later vs 1 of 9 after expiry. Overall stealth ≈ 50 % late, expiry ≈ 10 % late. likely (re-application trigger itself not identified: boss cast ends explain only 43/84 applications — REJECTED as the trigger.)
• Last phase: Fixated is only applied on ticks at +2.5, +22.5, +42.5, +62.5 s (0.5 s before green round +3 s and before each PD tick), only if nobody holds it:
  • +2.5 s tick never skipped: 13/13 applied when nobody held it (11 after expiry, 2 after a down, 0 after stealth = untested).
  • +22.5 / +42.5 / +62.5 ticks skip sometimes (≈ 40 %), no rule found (not time since loss, not cause, not holder state, not stealth at the tick). After a +22.5 skip, +42.5 applied 6/6. open
  • The previous holder can be re-fixated immediately (233450, 230802, 220703).

3. ★ The way to get fewer greens: the "+2.5 s Fixated" timing bug likely → strong

If nobody holds Fixated when the last phase starts, the +2.5 s tick gives it to someone. It lasts exactly 60 s, so it ends 10–30 ms after the +62.5 s tick. That tick still sees a holder → gives Fixated to nobody → the +63 s green round has no fixated player and comes out 2 greens short:

4 of 4, no counter-example. Every other log where +2.5 s applied had the holder go down before +62.5 s (normal +63 s round). The group in 005256 / 010435 / 3273c (Scourge Too Strong, Dps Is Optional, Lahyde, …) appears to do it on purpose. Recipe: (1) nobody fixated at last-phase +2.5 s; (2) that Fixated holder doesn't stealth or go down for 60 s; (3) +63 s → 1 green. For (1) the last pre-last-phase Fixated must be applied ≥ ~58 s before the last phase starts, or be removed (stealth during Split 3 — untested, and stealth makes the boss skip its next chance ~half the time in boss phases). In the user's 10 raw last-phase fights, 9 still had the pre-last-phase Fixated at +2.5 s (applied only 30–45 s before).

Related "stale fixation" cases (one green lost at +33 s): 050224 (Fixated stealthed off at +15 s, +22.5 tick skipped → 2 greens); 210120 +63 s (holder downed, +62.5 skipped, 0 instead of 1). Not consistent: other no-fixated rounds were full (214712, 234618, 213946, 223543 at +33 s). open

4. Probability Distribution (spread + puddle)

• Fires at last-phase +23, +43, +63 s (every 20 s); same count rule (max(1, floor(up/3))); fixated player first. 63/65 rounds match. Never came up short with everyone up. solid

5b. Ascension's Sacrifice — 2-people green "shackles" (Staff phase) solid

Skill 81076, cast twice in the Staff phase (~38 s and ~85 s raw time). Marker = agent effect 37976 (EI GUID A47987D0864223429B261683B6452826, mechanic 2Green.Slct). Data: data/ascension_sacrifice_2green.csv (script scripts/asc_sacrifice.js, 62 casts / 186 picks from the user's raw CM logs).

• 3 markers per cast, 0.68 s apart, at +2.03, +2.72, +3.40 s after the cast starts.
• Marker 1 = the Fixated player: 59/59 casts where someone was fixated (3 casts with nobody fixated → someone else).
• Markers 2 and 3 are not proximity-based: rank by distance to Vloxx is uniform (mean 4.7 vs 4.75 random, ranks 1–10 all common); same for distance to the previous target and to the fixated player. Looks random among players up (excluding those already marked).
• Not baitable by positioning. Only "bait": whoever holds Fixated at that moment gets the first shackle.
• Deeper search for picks 2/3 (124 picks, chance of "picked player is the top candidate" = 11.8 %), all at chance level (scripts asc_sacrifice_features.js, asc_sacrifice_links.js): distance to boss (at pick, at cast start, 2 s before), angle to boss facing, distance to previous target / fixated target / nearest Aspect / arena centre, crowding, HP %, Ascension stacks, toughness/healing/concentration/condition ranks, subgroup (all-same-subgroup 16 % = random), squad order, instance id, profession, damage on boss (5/15/60 s/total), damage taken, time since last boss hit, times previously targeted by 3-greens / PD / shackles / Fixated, being hit/targeted by any boss or Aspect skill or effect in the 11 s before. Per-player rates 0.8–1.3× expected. Best single feature 17.7 % (noise among ~50 tests). Buff 80951 (5 s) + skill 23296 (5,000 dmg) + 81076 hit land on the target at the same ms as the marker = consequences. Conclusion: indistinguishable from uniform random among up players (excluding already-marked) in logged data. likely
• Re-run on the full raw set (2026-10-02, 128 CM raw logs: 264 casts / 127 fights / 519 picks 2–3): marker 1 = fixated 242/242 (22 casts with nobody fixated). Best feature 14.1 % vs 11.8 % random (≈ +1.6 SE, noise among ~50 tests); the earlier "damage on boss 5 s" 17.7 % regressed to 12.3 %. Monte-Carlo randomness tests (scripts/asc_sacrifice_randomness.js, 5,000 sims): every statistic p = 0.25–0.80; the small-sample "too even across players" hint (p = 0.02 on 62 casts) vanished (p = 0.36). A per-player bias of ≳1.3× would now be detectable — none found. solid
• 12 NM logs with 33 shackle casts are in logs/raw_nm/ (not analysed).
• Line of sight REJECTED (scripts/asc_sacrifice_los.js, positions in data/ascension_sacrifice_positions.csv): pick rate vs expected by absolute 300×300 map cell (25 cells, chi-square 8.7, random ≈ 25), by direction × distance from Vloxx (16 bins, chi-square 5.3, random ≈ 15) — every area 0.76–1.32× expected, no dead zone. All candidates are within ±30 height of Vloxx (flat arena), so nothing to hide behind.

5. Ascension Orbs & Aspects

• Orbs are agents (max HP 14,940). They spawn in groups of 3, 0.2 s apart, 200–450 units around an Aspect, ~2.0 s (1.5–2.6 s) after that Aspect's breakbar is broken — 86/86 confirmed breaks dropped orbs (full set: 250/251, the miss = break 2.1 s before the log ended); deaths without a CC dropped orbs 4/91 times (probably a simultaneous CC elsewhere). CC drops orbs, death doesn't. solid Bigger groups (6/11) = overlapping breaks or Vloxx's own breakbar.
• Unpicked orbs disappear 120.1 s after spawning (10/10; full set 14/14 that nobody picked; team-change event 5 s after vs 2 s for picked ones). Picked orbs: median ~4 s after spawn. Picking one gives the player Ascension. solid
• Aspects (Staff, Spear, Sword) respawn exactly 40.0 s after dying, each on its own timer, in every phase incl. last phase (Staff 64, Spear 58, Sword 22 deaths; full set Staff 199, Spear 156, Sword 36 — all 39.97–40.01 s). solid

5c. Arena coordinates — Aspect spawn points solid

• Each Aspect always spawns at the exact same spot (670 spawns in 128 raw logs, zero variation; all at z = −12395):

• Spawn spacing: Staff↔Spear 1,649 · Staff↔Sword 1,247 · Spear↔Sword 1,200 (world units).
• Conversion: map coordinate = arcdps world position × 0.0254 (inches → meters, no offset; second value increases upward on the map). Derived from the user's marker tool screenshot (arrow 292.731, 408.926; circle 267.190, 389.968) whose layout around the boss only fits this mapping; not yet confirmed in game by placing a marker on a spawn likely.
• arcdps squad markers: statechange 53, skill = marker index (0 arrow, 1 circle, 2 heart, 3 square, 4 star, 5 spiral, 6 triangle, 7 x), position = src as two floats (x, y) + value as float z; (Infinity, Infinity) = marker removed.

6. Hypotheses that were tested and REJECTED (don't redo)

• Range limit ~600 (last phase) / ~775 (early, = 600 + boss hitbox 150 + player 24). Looked perfect on EI data (211/211), broken by the user's raw 2026-10-01 logs (201030 +33 s: 3 greens with everyone 714–898 away; 213946/223543 +33 s: 3 greens with only 2–4 players within 600). The fit was an artifact of EI's 300 ms positions. Greens in the last phase are not range-limited in any way found. (Fight-start 1-green is due to everyone being 4,000+ away / not engaged, not a 600 range.)
• "Random draw, stop at first out-of-range/invalid pick" model.
• Reservation of soakers around each green; minimum spacing between greens; distance to arena centre or any other point.
• Boss HP threshold (~6.5 %) — looked monotonic on 8 rounds, nothing happens at that HP in game; superseded by §3.
• Stealth / dodge / any player cast at the green moment (dodging players still get greens; nobody dodged in 220703).
• Threat via damage on Vloxx (direct or condition); being in combat (everyone enters at log start).
• Greens on clones/minions (raw logs: greens only ever on players); EI hiding duplicate greens (EI mechanic has 0 ICD).
• Ascension stacks, Empowered stacks, adds alive, Probability Distribution targets as a cause of the count.

7. Open questions / suggested next tests

1. Have the fixated player stealth during Split 3 (before the 81071 channel starts) → check in the log whether the +2.5 s tick still applied (fixation_last_phase_ticks.csv, tick 2.5) and whether +63 s had 1 green.
2. Rule behind skipped +22.5/+42.5/+62.5 s ticks.
3. The 050224-type one-green-short +33 s rounds.
4. Exact weighting for non-fixated green targets.

Anything new should be checked on raw .zevtc (exact positions, every effect) — EI JSON misled us once (range).

The goal behind the research

Gather general, evidence-backed knowledge about how Vloxx CM works — how each mechanic picks its targets, its timers, what adds do and when they respawn, where things spawn — so the group can plan the fight from facts instead of guesses. Every claim is tied to log evidence and a confidence tag, and disproved ideas are kept so they aren't re-tested.

Topics so far: 3-people greens (Judgment of Eternity), Fixated, Probability Distribution, 2-people green shackles (Ascension's Sacrifice), Ascension orbs, the Aspect adds (respawn, CC, spawn points), arena coordinates. A notable practical result: the "+2.5 s Fixated" timing bug that makes the last-phase +63 s green round come out 2 greens short (FINDINGS §3).

Important IDs (arcdps build EVTC20260929, game build 207890/207970)

Working rule (from the user): whenever something new is learned, save it here (FINDINGS.md / README / data / scripts) in the same turn, then rebuild the doc page: node scripts/build_docs.js. New FINDINGS sections (## N. Title) appear automatically; to give one its own sidebar category, add an entry to CATS in scripts/build_docs.js.

Datasets (data/)

Times are seconds from log start. Raw logs start ~3 s later than EI (EI shows the first Fixated at 0 s, raw at 3.0 s). "Last phase" = from the start of Vloxx's channel skill 81071; its rounds are at +3, +33, +63 s (greens), +23, +43, +63 s (Probability Distribution) and Fixated ticks at +2.5, +22.5, +42.5, +62.5 s.

Getting more logs

• dps.report: https://dps.report/getJson?permalink=<id> → EI JSON (positions every 300 ms, buffs, mechanics). No raw .zevtc download.
• GW2 Wingman (https://gw2wingman.nevermindcreations.de), API documented inside the site bundle:
  • /api/getRndLogs?bossID=28106&sampleSize=-1&onlyKills=AllTries&IncludeEnglishLogs=on&IncludeFrenchLogs=on&IncludeGermanLogs=on&IncludeSpanishLogs=on&returnFullLogs=true → all attempts with summaries (~140 MB)
  • /api/getFullJson/<log id> → full EI JSON; /api/getMetadata/<log id> → summary.
  • Wingman does not flag CM for Vloxx. Use HP: CM = 84,939,840 (EI's own rule: >70 M CM, >85 M Legendary CM); NM = 42,469,920. Estimate from summaries: squadDmg / (1 - hpLeft/100) ≈ 85 M (CM) or ≈ 42.4 M (NM).
  • CM released Tuesday 2026-09-29 23:00 CEST. On 2026-10-01 Wingman had 14 CM attempts that reached the last phase (all in logs/), 1 CM kill.
• Raw logs are always better than EI JSON: exact positions, every effect/buff event, agent effects (EI doesn't export). New raw logs: the user's folder C:\Users\Sanqu\Documents\Guild Wars 2\addons\arcdps\arcdps.cbtlogs\Vloxx (28106).

How to run

cd vloxx-research
node scripts/build_dataset.js      # ~15 s, rebuilds data/*.csv from logs/
node scripts/key_checks.js         # prints the headline conclusions recomputed from the CSVs

scripts/evtc.js parses .zevtc directly (zip + EVTC rev 1). scripts/load.js turns raw EVTC and EI JSON into one common structure (greens, PD, fixation, stealth, downs, deaths, positions, orbs, aspects).

Folder layout

README.md            this file (context + how to continue)
docs/index.html      static documentation page (categories, search) GENERATED from FINDINGS.md + README.md + data/ —
                     rebuild with `node scripts/build_docs.js` after every change to the notes
FINDINGS.md          every conclusion, its evidence, and hypotheses that were tested and REJECTED
logs/raw/            the user's own raw arcdps logs (.zevtc), CM, 2026-09-30 (5 that reached last phase) + all of 2026-10-01
logs/ei/             Elite Insights JSON of other squads' logs (gzipped). dps.report links + Wingman imports
logs/sources.json    log id -> URL (dps.report / Wingman)
logs/wingman_vloxx_all_attempts_metadata.json.gz   Wingman summary of all 3,398 Vloxx attempts (as of 2026-10-01)
data/*.csv           datasets extracted from all logs (see below)
scripts/             parser + loaders + dataset builder + checks (Node.js, no dependencies)

Status / open questions (see FINDINGS §7)

1. Untested: does stealthing off Fixated during Split 3 (before the last phase starts) let the +2.5 s tick apply a fresh Fixated? Boss-phase data says stealth makes the boss skip its next re-application ~half the time.
2. Why later last-phase ticks (+22.5/+42.5/+62.5) sometimes skip with nobody fixated — no rule found.
3. 1 unexplained short round: 050224 +33 s (Fixated stealthed off at +15 s, +22.5 s tick skipped, 2 greens instead of 3).
4. Which players get the non-fixated greens — biased to near the boss, not strictly nearest; exact rule unknown.

Datasets (data/)

Logs: 128 raw CM .zevtc · 17 EI JSON · 12 raw NM.

Scripts (scripts/)