Over 1000 posts asking for death recap

"

noribin#3026 a écrit :

"

DistributedAutomaton#5739 a écrit :

"

noribin#3026 a écrit :

I say its easy as a dev.

Nope. To get a basic understanding why, ask either Gemini or Claude, "how easy or difficult would it be to create a death log for a complex ARPG like Path of Exile 2?"

How you ask the question matters.
For example:

We want an estimate of a death recap in poe 2. We do not care about 100% exact information. Will this be an easy task?

Short answer: yes—wanting only an estimate makes it easy. You can build a fast, client-side “best guess” death log without touching server code. It won’t be perfect, but it can be useful and lightweight.

What becomes feasible with “estimates”

Client-only ring buffer (2–5 s): record recent incoming damage events (as exposed to the client), DoT ticks, buff/debuff changes, player inputs, and position samples.

Heuristic attribution: pick the most likely killer using rules instead of authoritative calculations.

Confidence scoring: communicate uncertainty (“High” if a single big hit; “Low” if many overlapping sources).

Tiny footprint: 1–3 MB RAM per character, negligible CPU if you sample/coalesce events.

Minimal data you need (client-side)

Time (ms), event type (hit/DoT), damage number(s) if available, damage tags (e.g., Fire/Cold/Physical), source handle/id if available, projectile/melee/ground flags.

Player snapshot every ~100 ms: life/ES, guard/ward, active guard skill (e.g., Molten Shell), flask states, major debuffs (shock, brittle, scorch, exposure).

Context hints the client already knows: map area, party size, latency estimate.

Heuristics that work well in ARPGs

Last-Hit Dominance: If a single hit within the last 250 ms dealt ≥ X% of max EHP (say 40–60%), attribute death to that source. Confidence: High.

Burst Window: If ≥3 hits in 350 ms reduce EHP by ≥70%, attribute to “Burst” and list the top source by damage share. Confidence: Medium.

Sustained DoT: If DoT ticks account for ≥60% of EHP lost over the last 1.5 s, attribute to the latest applied DoT tag (Burning, Poison, Bleed) and its applier if known. Confidence: Medium.

Ground Effects: If the player is inside a harmful ground volume for ≥500 ms and DoT ticks match that element, prefer “Ground (element)” unless a single hit threshold triggers. Confidence: Medium.

Ailment Spike: If a high-impact ailment (Shock ≥X%, Sap, Brittle) appeared ≤500 ms before lethal damage, append “amplified by Shock (X%).”

Guard Drop: If a guard/ward buff ended ≤300 ms before lethal, add “after Guard expired” tag.

Latency Reality Check: If RTT > 100 ms and the lethal hit timestamp is earlier than the client’s last life bar update, downrank confidence by one level.

Simple decision flow (pseudo)
window = events[-2500ms:0]
ehp_lost = sum(window.damage_to_ehp)

if last_big_hit.damage >= 0.5 * max_ehp or last_big_hit.damage >= 0.7 * current_ehp:
cause = last_big_hit.source_tag
confidence = "High"
elif burst_damage_in(350ms) >= 0.7 * max_ehp:
cause = "Burst (" + top_source_in_burst + ")"
confidence = "Medium"
elif dot_share(window) >= 0.6:
cause = latest_applied_dot_type + " DoT"
confidence = "Medium"
elif in_harmful_ground and dot_share(window) >= 0.4:
cause = "Ground (" + element + ")"
confidence = "Medium"
else:
cause = top_damage_source(window) # coalesced
confidence = "Low"

if guard_ended_recently: tags.add("Guard expired")
if shock_present: tags.add("Amplified by Shock")
if rtt>100: downgrade(confidence)

Output schema (compact & readable)
{
"time": "2025-11-06T12:34:56.789Z",
"area": "Crimson Fortress",
"latency_ms": 68,
"summary": {
"cause": "Fire Projectile",
"attribution": "Rare Fanatic Marksman",
"confidence": "Medium",
"pattern": "Burst"
},
"timeline_ms": [
{"t": -820, "event": "Ignite applied", "source": "Marksman"},
{"t": -270, "event": "Projectile Hit", "dmg_est": 1420, "tags": ["Fire","Projectile"]},
{"t": -40, "event": "Ignite tick", "dmg_est": 380},
{"t": 0, "event": "Death"}
],
"player_state": {
"max_ehp": 5100, "ehp_before": 1600, "flasks": {"ruby": "cooldown"},
"buffs": ["Molten Shell (expired -310ms)"], "debuffs": ["Shock 18%"]
},
"contributors": [
{"source": "Fire Projectile", "share": 0.62},
{"source": "Ignite", "share": 0.38}
],
"notes": ["Guard expired 0.31s before death"]
}

How accurate is this in practice?

Often “good enough.” You’ll correctly flag most one-shots and many burst deaths.

Weak on edge cases: overlapping ground DoTs from multiple mobs, hidden monster mods, penetration/crit rolls you can’t see, or desyncy moments.

Mitigation: show confidence, list top alternatives (“Possible: Chaos DoT (low confidence)”).

Performance & implementation tips

Coalesce events: combine repeated ticks from same source each 50–100 ms to cut volume.

Sample inputs at 60–100 Hz, states at 10–20 Hz.

Use IDs, not names, in the buffer; resolve to strings only on display.

Bound memory: cap the ring buffer by time and count (e.g., 2,000 events or 5 s, whichever first).

Debounce spam: ignore micro-hits below 1% max EHP unless inside a burst window.

Nice UX touches

Show a one-line summary (“Killed by Fire Projectile (Rare), burst in 0.33s — Medium confidence”).

Offer “Why?” with a collapsible timeline and tags (Shock, Guard expired, Ground).

Let players copy a JSON blob for bug reports/posts.