Capture Completeness
We are rebuilding the New World game server from packet captures of the live game. This page tracks every gameplay action we know about, whether the captures we have prove how it works, and which new captures would fill the gaps.
Where things stand
What we still need
Progress by area
Every tracked action in each area, by state. Pick an area to see what it still needs.
Full evidence table — all tracked rows with their technical proof grades, filters and work queues
How to read the technical table
1st-person proven — a local input or request was captured together with the authoritative server result it produced. The strongest evidence.
3rd-person — the action was watched being performed by another player. That capture can never contain the other player's input, so it proves the output only.
Output only — the server output is well-attested but the local input that triggers it has not been resolved yet.
Route only — a typed client request exists on the wire with no paired authoritative result identified yet. A transport ACK does not count.
Correlated / candidate — an input classifier matches repeatedly but is not yet unique or controlled enough to call proven.
Bounded negative — scanned and absent, within a recorded scope. This means "our captures don't contain it", never "the game doesn't have it".
Not linked yet — a checklist item that no reviewed link folds onto an evidence row. The scanners may hold evidence for it under a route or selector name; the board cannot claim it until someone writes the link.
Not in any capture — an action known from the game datafiles whose authored animation appears in no capture in the corpus. A real capture gap.
Captures column — how many distinct capture sessions contain the evidence. A capture merely declared to cover an activity is scheduling metadata, not proof.
Conformance column — a different question: did our server answer a replayed retail input with the same wire output. Scored per retail gesture window and per tier, each shown separately: census = per-channel message counts, tape = the owner state/selector tape, outcome = cooldown / hit / projectile results, rpc = typed request → response. "not run" means the tier was not attempted; a census pass alone never means the behaviour matched. It never changes the proof grade.