Glossary

An acknowledgement lets the sender stop recovering an item the receiver has accepted. At the game level, receipts identify complete scenes, action results or projectile records. Gap maps preserve which items are still missing rather than treating every earlier item as received.

QUIC also has transport acknowledgements. Those concern packet delivery and loss detection; the application may still lack a fragment needed to decode a scene. Keeping these layers distinct prevents an incomplete update from becoming the next trusted baseline.

An action is identified within its session, player occupant, life, input schedule and sequence. Retrying the record keeps that identity. Delivery receipts have their own identity because one action can produce several contacts.

This lets the game recover missing results without turning duplicates into new outcomes. It also prevents an old queued action from borrowing the identity of a later life or a new person using the same player slot. A visually similar effect is not proof that two records are the same action.

The air step is SHADOWFRAG’s second jump. Press Jump again while airborne to extend or redirect a route; simply continuing to hold the first jump is different from a fresh press.

A shadow warp restores the air step, making chains such as leap, air step, warp and another air step possible. The visible flip explains the action, while the gameplay collision hull keeps its ordinary orientation. An air step follows the movement rules and momentum limits; it is not unlimited flight.

Application payload counts bytes belonging to SHADOWFRAG’s encoded records at the measurement’s stated boundary. Repeated records can still count as payload even when their information is a retry rather than a new action.

The receiver’s IP traffic also includes QUIC framing, authentication, acknowledgements and network headers. A 33% payload reduction is therefore not automatically a 33% reduction in wire traffic, CPU work or latency. Local models, sounds and particle geometry are drawn from installed assets rather than streamed as full visual state.

For samples of 2, 2 and 8 ms, the arithmetic mean is 4 ms. The slower sample affects the mean, while the value alone does not show the distribution or its tail.

A mean completed-frame time describes the fixture’s frames. Precision Dojo’s mean time alive describes eliminated practice ninjas’ lifetimes, including time spent aiming elsewhere; it is not your reaction time. Neither should be compared with a different metric merely because both are averages.

With both layers intact, ordinary damage follows armor, then shield, then kill across separate hits. Armor gives one extra layer; it does not create a second shield or a health bar.

Armor availability follows the match’s pickup rules, and Shogun has no armor pickups. Shadow arrival kills bypass armor and shields. A cosmetic chest plate or model change does not independently grant protection: the server’s equipment state determines the layer.

An audio device needs a continuing stream of samples. The callback mixes already prepared voices into the next output segment under the device’s timing constraints.

SHADOWFRAG’s callback performs no file I/O, decoding, allocation, network access or sound generation. A worker prepares and validates bank replacements; the running game adopts them using its existing audio synchronization. Playing sounds retain their sample data while later triggers can use the replacement. Keeping that work separate supports consistent playback without claiming that every possible device or listening setup is already qualified.

The game triggers a cue when an eligible event occurs, then selects its prepared recording. Multiple takes give repeated actions variation without making every clip a different gameplay rule.

The current installed bank has 185 edited clips across 106 active cues. Reserved design slots are not active triggers. Rail readiness cues follow the owner’s real cooldown, while hit sounds confirm accepted contacts. The native mixer plays prepared samples; it does not call a sound-generation service during a match.

Mixing determines how prepared voices combine. Panning places sound across the output channels; distance attenuation reduces a spatial sound with distance. Ducking temporarily lowers selected sounds so another remains understandable, while crossfading blends between old and new audio.

In SHADOWFRAG, tutorial speech gently lowers effects and ambience while private hit confirmations remain clear. Environment changes and replacement ambience crossfade. Much filtering and level preparation is baked into the clips offline, and the native callback retains its bounded playback work rather than loading or decoding new files.

Authoritative state is the shared result the server has validated under the game rules. A client requests movement or an attack; the server checks identity, timing, equipment, cooldowns and target eligibility before committing the outcome.

Prediction makes a local action respond promptly. It remains a provisional view that can be corrected by the server’s answer. This separation helps keep one consistent fight, but server validation alone is not proof that every form of cheating is impossible. Demos add inspectable evidence with their own limits.

A server needs upload capacity for traffic sent to everyone; a client needs download capacity for its own incoming traffic. Data-rate counters show what crossed a defined boundary during a sample. A configured per-client budget caps application sending but does not discover the real path’s available capacity.

More bandwidth alone does not guarantee lower latency or a larger match: CPU work, queues, packet scheduling and client generation can still fail. Always distinguish application payload from on-wire bytes and state whether a number is per player, busiest client or aggregate server traffic.

A useful benchmark records what ran, the build and content, hardware, duration, settings and measurement boundary. Render fixtures, simulation tests and two-host protocol campaigns each isolate different costs.

An isolated fast renderer and a fast server cannot be added together to certify a full match with rendering, input, audio, networking and AI. Retained failures and background contention are part of interpreting the evidence. A performance target is also different from a measured result that achieved it.

SHADOWFRAG bots move, fight and pursue objectives through the game’s controllers and cooldowns. Shogun adds squad assignments, route planning, purchases and capture commitments on the server.

A bot brain is different from a synthetic protocol client used to generate network traffic. The latter can send real movement and attacks without rendering, mixing audio or running full tactical AI. A test with 128 protocol clients therefore does not establish a 128-bot Shogun match. Bots are included in ordinary player-cap accounting.

The bow records a draw, then releases an arrow when you let go. More draw changes its accepted launch speed. Unlike a rail shot, the arrow still needs time to reach a moving opponent, so aim ahead of where that opponent is now.

Draw, release and cancellation are distinct actions. Losing a network packet must not invent a release, and cancelling a draw must not fire an arrow. Yumi, Bamboo and Oni change the model, not the bow’s rules. Precision Dojo uses the same draw and projectile handling as the game.

Take the opposing flag and return it to your base to score, but your own flag must be home. A carrier may therefore need teammates to recover the friendly flag before completing a return.

Chi, air steps and shadow warps remain available while carrying. The objective does not itself require one particular weapon or shield policy; those are selected separately in standard setup. The full map and objective indicators help locate flags without revealing enemy players as radar contacts.

Chi-jump is the movement ability bound to Left Shift by default. Unlike an ordinary directional leap, its launch follows your aim: look along a low arc to cross distance or higher to reach a ledge.

Chi has one use per flight and a cooldown. It refreshes on landing or when a shadow warp arrives beside a wall. It does no damage, and carrying a flag does not remove it. Its aim, momentum and gravity still follow the shared movement controller; the visual arc is not a preselected cinematic path.

Your client reads input, predicts your movement and draws the world. The server receives requests from participants, advances shared rules and confirms positions, damage, equipment and objectives. The server is also called the authority.

A dedicated server runs the match without a local player’s rendering workload. Local Practice still exercises a real client/server connection, while Precision Dojo is a separate offline session within the client. A private host leaving ends that hosted match; a server and client need compatible game builds.

Without prediction, every leap would wait for a network round trip before visibly beginning. SHADOWFRAG applies your inputs locally through the movement controller while sending them to the server for validation.

The resulting predicted state feels immediate but is provisional. If the server’s accepted result differs, reconciliation restores that result and reapplies newer inputs. Projectile effects can also be predicted visually; drawing an arrow or beam does not confirm a damaging hit. Prediction cannot remove physical connection delay or make an invalid action valid.

The Council plans one or two Community Shoguns per region to lead local events and represent players in conversations with the development team. Relevant experience, region and time zone help the team review an introduction.

The title does not grant an in-game weapon, character class or authority over Shogun capture rules. It names a contribution to the community. Read the Council role description.

A presence mask says which fields are actually encoded. A varint uses fewer bytes for suitably small integers; ZigZag maps signed integers so small positive and negative values can both encode compactly. These are representations, not permission to round away aim or movement state.

The compact-wire work omits only exact known defaults, preserves nonzero values and negative zero where required, and retains full gap receipts. Its scene encoding does not depend on a previous unreliable packet. Smaller records can still take CPU work, so byte savings do not imply an equal latency or capacity gain.

A hash is calculated from specified input. Matching expected digests helps check that participants or tools are using the intended map, package or binary. A gameplay fingerprint similarly identifies the defined shared gameplay contract.

The map atlas retains geometry hashes so an image can be traced to its actual layout. Test manifests retain build and content identities for reproducible comparison. The digest only identifies what was hashed: it does not establish balance, safety, performance or identity trust by itself.

A cooldown limits how often an ability or weapon can act. SHADOWFRAG also shares weapon timing so changing weapons does not bypass an attack’s wait. Reloads, ammunition and one-use movement readiness add their own constraints.

A readiness sound follows the owner’s actual cooldown reaching zero. It is feedback about the rule, not the rule itself. Cooldowns differ by action: the net’s long recharge and chi’s per-flight readiness should not be read as interchangeable timers.

CPU means central processing unit. It executes the game’s ordinary instructions, including movement rules, AI planning, input services and transport work. Those jobs can run on different threads and compete with other programs.

A CPU percentage, elapsed time and measured per-stage duration are different quantities. The client generator’s reduced CPU use is a result for synthetic protocol clients, rather than a full-player rendering benchmark. Machine model, active background work and workload must remain attached to comparisons.

Congestion control limits outstanding transport traffic in response to path feedback so a sender does not simply emit unlimited bytes into a busy connection. CUBIC uses a cubic growth function as part of that control.

SHADOWFRAG currently uses CUBIC with PicoQUIC and keeps Quiche/CUBIC selectable. Congestion control differs from a configured application byte budget and from pacing individual packets. It cannot remove physical distance or guarantee a shot’s response time. The algorithm’s standard is RFC 9438.

A renderer can reject objects or effects outside its view before constructing or submitting their geometry. SHADOWFRAG’s movement feedback uses bounded local pools, short lifetimes and view culling so unseen dust need not be built.

This is a presentation optimization. The server still maintains whatever state combat and objectives require. Rendering visibility, network relevance and historical hit eligibility are separate policies, even when each tries to avoid unnecessary work.

Uppercase B means bytes and lowercase b means bits. In decimal units, 1 kB is 1,000 bytes, 1 MB is 1,000,000 bytes and 1 Mbps is 1,000,000 bits each second. Thus 2.5 MB/s is 20 Mbps.

Binary units use different labels: 1 KiB is 1,024 bytes and 1 MiB is 1,048,576 bytes. Check the stated counter boundary too: a payload rate excludes transport overhead, and a server total cannot be compared directly with a per-player rate.

UDP carries individual datagrams. QUIC runs over UDP and its datagram extension lets an application send message boundaries without QUIC retransmitting a missing datagram. Later data need not wait for that missing message to be restored.

This does not mean bypassing security or congestion control. SHADOWFRAG can still repeat inputs and acknowledge durable records at the game layer. QUIC packet delivery and complete-scene acceptance remain distinct. See the QUIC datagram specification.

A referee can align compatible tracks to inspect committed actions, rejection reasons and available timing. The authority track supplies server decisions, while participant tracks can provide each client’s reported observations. Their different provenance must remain visible.

Absent tracks or incomplete files are gaps in evidence. A modified client can lie before committing its report, and server identity needs independent trust. Client feedback intervals use that client’s own monotonic clock; subtracting unrelated clocks on separate machines does not establish latency.

Demolition asks teams to plan around sites rather than only accumulate kills. Attackers carry a bomb toward one of two sites, while defenders try to stop the plant or defuse after it is planted.

The server owns planting, interruption, defusing, detonation and round outcome. Clients display progress rather than supplying their own elapsed duration or winner. Map routes and rotations matter, and the selected weapons and protection still change how a site is approached.

With the default controls, WASD chooses your movement direction and Space starts a leap. A sideways leap can move you out of a firing line while your mouse keeps the target centered.

Holding Jump chains ordinary leaps; a fresh press near landing can use the landing timing window. Horizontal reach, jump height and time in the air are separate properties: the shorter-leap balance pass reduced distance without stretching the vertical arc. Default bindings can be changed in the game.

A duel is a head-to-head fight between two players. A compact arena makes it easier to learn an opponent’s routes, attack timing and use of cover.

It is not inherently an unshielded rail-only format. SHADOWFRAG separates map, objective, allowed equipment and protection, so those choices still define how the duel plays. A small duel also produces a different networking workload from a crowded team fight.

A match is not just one independent connection multiplied by its players. If every participant must learn about every other participant, there are N × (N − 1) directed source-to-recipient pairs: at four players there are twelve.

Real traffic also depends on relevance, shared encoding, events and equipment. The expression is therefore a useful model term, not a complete prediction of bandwidth or CPU. It helped the archived small-match cost model, but does not justify extending that fitted model to unmeasured 64-player encounters.

Field of view, abbreviated FOV, describes how much a camera takes in. Along with aspect ratio and viewpoint, it changes how world shapes appear on screen. Weapon-model framing also affects how much the first-person arms or bow occupy the image.

SHADOWFRAG offers per-weapon aiming and weapon presentation settings, plus separate normal and zoom sensitivity. Precision Dojo uses your saved settings. A presentation change can expose more of the aiming area without changing weapon damage, projectile speed or the server’s hit geometry.

Context matters for this abbreviation. In a performance counter, FPS means frames per second. In descriptions such as Movement FPS or competitive FPS, it means first-person shooter: a game aimed from the character’s viewpoint.

The two meanings do not create a performance claim. Calling SHADOWFRAG a Movement FPS describes the role of movement in combat; it does not promise any particular frame rate on your machine.

A scene that is too large for one bounded message is split into fragments with an identified frame and part layout. The client validates their lengths and identities, assembles the complete scene and accepts it together.

A lost fragment can leave useful packets arriving without a fresh scene being installed. Newer state may supersede an incomplete older assembly rather than building a queue of stale fights. FramePart is the traffic table’s technical name for this category, not a video-frame upload.

Frame time is usually expressed in milliseconds. At a perfectly even 120 rendered frames per second, the interval is about 8.33 ms; at 240 it is about 4.17 ms. Real frame times vary, so an average alone can hide slow frames.

The measurement boundary matters. SHADOWFRAG’s retained completed-frame renderer fixture waits for GPU completion but excludes live input, HUD, audio and networking. Its reciprocal is not a live-game FPS promise. Frame time is also separate from the fixed simulation step and from a click becoming visible on a display.

Frame rate is a rate of images. Over a measured interval, the number of frames divided by elapsed seconds gives average frames per second. A more even sequence generally feels smoother than one with occasional long pauses at a similar average.

Inspect frame-time percentiles as well as the average, and check whether a report counts rendered, completed or displayed frames. A recording labeled 60 frames per second describes its capture format, not necessarily the game’s performance. A renderer-only benchmark does not establish the rate of a full live match.

In free-for-all, opponents are individual competitors rather than members of two objective teams. The map, equipment and protection rules still affect the fight.

This matters when reading the capacity study: the confirmed 128-client Highlands workload was synthetic FFA on that landscape. Using the Shogun Highlands map does not turn it into a Shogun match or measure the mode’s complete economy, tactical feed and bot planning. The ordinary game limit remains 64 players including bots.

FXAA means fast approximate anti-aliasing. It examines the rendered image to reduce the appearance of jagged edges, rather than changing gameplay geometry or collision.

In the retained renderer optimization, a scene at native resolution with FXAA disabled can use a direct path. Scaled scenes, previews and FXAA still use the offscreen targets they need. The measured 8.7% mean-frame improvement belongs to the matched ultrawide fixture; it is not a guaranteed effect of disabling the filter in every scene.

GPU means graphics processing unit. The CPU prepares rendering work and the GPU turns it into image data using geometry, textures and graphics programs.

The GPU can still be working after the CPU has submitted a frame. A completed-frame fixture waits for that work to finish, but the compositor and display may add further stages. A shot’s submitted gameplay reference therefore does not claim that its image physically reached the monitor. Resolution and effects also change the GPU workload.

A hash condenses bytes into a fixed-size digest. A chain includes earlier digest information in later records, so changing retained history breaks the expected links. SHADOWFRAG demos use SHA-256 chaining, server sealing and small periodic client commitments.

These checks support integrity and identify gaps. They do not create trust in an unknown server key, prove what appeared on a monitor or prevent a modified client from lying before it commits. Identity and the meaning of recorded data still need separate review.

A headless test exercises selected systems without normal interactive graphics. It is useful for repeatable rule, encoding, replay and networking checks that do not require a display.

Passing it is evidence for the tested code paths. Native UI, audio-device behavior, GPU presentation and a person’s experience may require separate checks. A headless or synthetic workload should not be presented as a full-client benchmark merely because it runs real game logic.

Remote players are drawn from timestamped state. For rail and melee, SHADOWFRAG can reconstruct the permitted target geometry associated with the action’s submitted scene, while independently validating the attacker’s origin and current damage eligibility.

History is bounded. An expired reference is rejected rather than silently exchanged for a newer target pose. A later respawn cannot be hit at the previous life’s position. This does not rewind every part of the match or erase peeker’s advantage. Arrows, stars, grenades and nets instead launch and collide in the current world.

The public scoreboard uses server-confirmed outcomes. An accepted damaging attack enters the attempt count; it becomes a hit when it damages an opponent. A released arrow can therefore be an attempt before its eventual impact is known.

Armor breaks, shield breaks and melee contacts count. One grenade damaging several opponents still contributes one successful attack. Self or friendly damage, nets, movement abilities, cancelled bow draws, rejected attacks and warmup shots do not award a weapon hit. Precision Dojo also reports practice accuracy, but comparisons still need the same weapon and practice conditions.

For rail and melee, a reference connects the input’s chronology with a canonical scene and permitted owner-camera basis. SHADOWFRAG selects from scenes successfully submitted to its bounded CPU renderer queue. Later rendering cannot casually rewrite a reference already assigned to an action.

The server still requires issued scene tokens, complete-scene receipts, matching life and warp identities, origin history and strict age bounds. Submission does not prove GPU completion, compositor presentation or what a person physically saw. An unavailable or expired reference is explicit; the game does not substitute a current scene to rescue it.

A visible model and a gameplay collision shape serve different jobs. The model explains who is moving and what they are doing; the hit geometry gives combat and movement a consistent boundary to test.

SHADOWFRAG animates flips and leans around the unchanged gameplay hull. Cosmetic headwear and weapon models do not change it. Different checks can also use different tolerances: shadow frags require physical body overlap, rather than borrowing a weapon’s aiming tolerance. A replay overlay shows reconstructed gameplay geometry, not proof of exact monitor pixels.

For a hitscan weapon, the game tests the shot’s direction against eligible targets and cover at the relevant moment. There is no arrow-like flight interval to lead. SHADOWFRAG’s rail uses this kind of trace.

The client may draw the shot immediately, while the server’s confirmation arrives later. A historical hit check can reconstruct a bounded earlier target scene, but cannot make every delayed shot valid. Projectiles use a different current-world flight and collision path.

HUD means heads-up display. Its groups explain immediate status while leaving the world visible. The HUD editor can move, anchor, scale or fade those groups, with previews for different aspect ratios.

The minimap and full tactical map answer a wider navigation question: where routes, objectives, friendly markers and supplied camps are. They do not provide enemy radar. Highlands keeps navigation available because of its scale; other maps have their own policies. HUD layout is a local preference, while objective state still comes from the server.

Immutable means unchanged after publication. A service constructs a state, then hands readers a stable version rather than letting them observe half an update.

SHADOWFRAG uses this separation so rendering can consume scenes without owning the input, prediction and network-result schedule. The network encoder can also share prepared actor bytes while preserving recipient-specific timing. Here publication means an internal software handoff, not publishing a website or releasing a game build.

Multiplayer commands belong to identified execution steps. The server enforces arrival bounds and the client also has bounded admission rules for delayed native events. The exact checks depend on the stage and action.

Late input and late action counters remain visible even if a workload passes other numerical screens. Recovery, a larger queue or a new epoch does not grant a late-fire grace period. An explicit rejection is preferable to quietly using a newer origin or view that the original action did not own.

The client and server need to agree which simulation step owns each input command. An epoch identifies that schedule and its command numbering. It is part of action identity, not simply a wall-clock date.

After a serious service hitch, coordinated recovery can grant a new schedule. Already assigned actions keep their original identity and outcome. A stale grant superseded by newer authoritative state must be ignored. Recovery does not relax damage checks or transplant an old shot into a later valid slot.

A physical input travels through the device, operating system, game input service, simulation or server decision, rendering and display. A software timer can observe selected stages, but cannot automatically prove the entire physical path.

SHADOWFRAG separates input acquisition, prediction and incoming results from rendering so a display wait does not stall those services. Its shot reference follows a submitted-view timing contract. That engineering and the software feedback observations still require separate hardware-event and display measurements before making a physical latency claim.

A handoff is a recovery transition in the agreement that places commands on server time. It is more significant than one packet arriving unevenly and is distinct from the small landing buffer used by jump controls.

The current direct-LAN capacity screen declared zero handoffs as a condition. Some higher-population runs connected and completed but needed two, so they did not earn a confirmation. A handoff count is therefore a diagnostic about continuity under that workload, rather than a player count or an ordinary frame-time value.

Instagib makes a single clean hit decisive instead of reducing a large health bar over several hits. SHADOWFRAG began around ninja instagib and later added projectiles, protection choices and objectives.

Under ordinary damage rules, armor is removed before a shield, and a damaging hit with neither layer remaining kills. A classic unshielded rail duel therefore behaves differently from a match where kills recharge shields. Shadow frags have their own arrival-kill rule. The word alone does not identify the selected weapon, objective or protection policy.

Interpolation has two related uses on this site. Snapshot interpolation reconstructs motion between timestamped gameplay states. In the cost lab, a fitted model estimates traffic between tested configurations inside a declared range.

The archived cost model covers 8–12 players, up to four bow holders and up to four shadow holders under its particular action schedule. It must not be extended to 64 players, unmeasured weapons, CPU or memory. Fitting observed samples is not enough on its own; the revision was checked against fresh configurations and seeds.

If successive updates take different amounts of time to cross the connection, their arrival spacing changes. That variation is jitter. It differs from a constant delay and from packets that never arrive.

Interpolation can smooth ordinary variation with a buffer, while input scheduling and bounded recovery address other parts of the path. Those mechanisms have limits and can add age or extra work. The capacity study retains separate delay, loss and jitter cases because combining them can behave differently from testing each alone.

A jump buffer briefly remembers an eligible press made near a landing, so the next jump can begin when the movement rules allow it. A landing timing window also supports deliberate momentum-preserving hops.

These are bounded movement rules shared by client and server. They do not add a grace period for firing late network actions, and they are unrelated to the input buffer used to schedule commands across a connection.

Last Ninja / Scroll gives players both a survival goal and an objective to contest. A player can win by remaining the last survivor, or by keeping the scroll for 60 continuous seconds.

The continuous hold is important: reaching the scroll once does not settle the round, and a dropped objective changes the pressure on the remaining players. The scroll carried in The Crimson Path’s story is part of that tutorial’s mission, rather than evidence that it is running the same multiplayer rules.

Latency can describe one-way delivery, a round trip, input processing, server response or a physical action becoming visible. Those intervals contain different work and cannot be used interchangeably.

Distance, routing, queueing, scheduling and recovery all can add delay. SHADOWFRAG separates measured path RTT, scene age, software weapon feedback and rendering. A LAN test does not guarantee the same result on the Internet, and software timing alone does not measure physical mouse-to-display response.

A numeric player slot can be reused after someone leaves. Occupant generation distinguishes those people. Life or spawn identity changes after a respawn, and warp identity marks a discontinuous position change within a life.

Snapshots, shot references, effects and projectile ownership retain the identities they need. An old shot cannot hit a later life at the previous position, and interpolation or collision must not invent a path through a teleport gap. Existing projectiles also keep their launch-era appearance rather than adopting a replacement occupant’s look.

A loadout can specify rail, bow, stars, shrapnel and melee, with net and shadow access controlled separately. Standard match setup separates equipment from map, objective and protection choices, while Shogun uses its own purchasing rules and excludes railguns.

Some development experiments assign fixed equipment cohorts or holder quotas. Those allocations define a workload; changing them changes the scenario. They are separate from the next-round cosmetic appearance roster and cannot raise the configured player cap.

Map Studio combines terrain, materials, reusable assets, placement and validation. An author can walk a route with the actual movement controller, inspect collision or navigation overlays, then compile the editable document into a package the game can load.

A playtest starts a separate local host and client while preserving the editor document and camera. Asset identities and revisions keep later library changes from silently altering placed objects. Structural checks help find blocked passages, but human play still has to judge whether a route is enjoyable and balanced.

The cost lab estimates how traffic changes when another bow or shadow holder joins a declared small-match setup. More players also mean more viewers, so the fitted additional bow traffic grows with population.

This is conditional model behavior inside measured bounds. It does not include every CPU, memory or burst cost, and is not a hosting guarantee. Newer active/unused encounter pairs also depend on deaths, supply and actual throw counts. Adding isolated feature differences together cannot certify an untested combined fight.

The median splits ordered observations around the middle. For 2, 2 and 8 ms, the median is 2 ms while the arithmetic mean is 4 ms. This makes it useful for describing a center without letting one extreme observation dominate it.

A median over runs and a median over individual clients are different summaries. Neither replaces a tail percentile or explains which participant had the largest delay. Check the grouping when reading paired benchmark results.

Melee asks you to reach striking distance before attacking. The server validates the permitted origin and historical target reference, with a short reach and aiming tolerance. One swing hits one target and removes one armor, shield or life layer under ordinary damage rules.

Drawing a blade motion across the screen does not make every touched pixel a hit. Katana, Ninjato and Kama are cosmetic blade designs with the same rules. A swing respects cover, teams and shared weapon timing; using it cancels a held bow without releasing an arrow.

Elapsed-time measurements need a stable ordering. A monotonic or steady clock measures intervals independently of changes to the calendar time displayed by the operating system.

Its numeric origin is not automatically shared with another computer. Timing a request and reply on one client is valid for that interval; subtracting a server timestamp from a client timestamp requires an explicit synchronization model and uncertainty. SHADOWFRAG distinguishes original event time, acquisition time, server phase estimates and eventual feedback rather than treating them as one physical clock.

FPS here means first-person shooter: you aim from your ninja’s viewpoint. Movement FPS describes a shooter where choosing and combining routes matters alongside aiming.

In SHADOWFRAG, directional leaps, air steps, chi-jumps and shadow warps can change a landing or open a firing angle. Those tools remain central when the objective changes from a duel to carrying a flag or reinforcing a Shogun settlement. It is a description of the game’s emphasis, rather than a separate match mode or a promise about frame rate. Explore the game.

Native demos retain enough state and events for the renderer to reconstruct a match. The viewer offers pause, seeking, playback speed, player follow and free camera. Opening playback leaves the live match and cannot send recorded input back to a server.

Competitive review combines server evidence with participant tracks collected after the match. Ordinary clients record only what they received. Missing information stays visible rather than becoming invented inventory or position evidence. Reconstructed playback does not prove exact monitor pixels or automatically expose an external cheat overlay.

At native resolution, the scene and the intended output use matching pixel dimensions. A lower or higher render scale draws the scene at a different size and requires scaling for the output.

Changing pixel count affects image detail and GPU work, while field of view changes how much of the world the camera includes. SHADOWFRAG’s native direct path can skip a composition pass only when its conditions hold; scaled rendering and FXAA still need their appropriate targets. An ultrawide image is also a distinct shape, not simply a faster or slower preset.

A navigation graph represents places and valid connections for a route planner. Bots use it to choose how to reach an objective, then apply real movement commands along the route.

This is why a visual map inspection is insufficient: cover beside the southern Highlands bridge disconnected nearby sampled routes even though its center was physically walkable. Repairing the graph connection restored a practical crossing. Planning frequency and work are bounded, so the AI does not search the whole world every rendered frame.

The net is bound to F by default and does not require switching weapons or interrupting a drawn bow. It travels visibly, stops at solid cover or its first eligible enemy, and leaves armor and shields intact.

The slow affects horizontal movement while preserving jump height, steering and weapon use. Shadow warps remain available. Another net cannot stack or refresh an active slow. Death does not reset the throwing cooldown. Net contact has its own confirmation and is excluded from damaging-weapon hit percentage.

Netcode includes sending input, predicting local movement, accepting server corrections, reconstructing other players and confirming combat. The transport library carries bytes, while the game’s own rules decide what those bytes mean.

Good local response does not prove that remote state is fresh, and a fast server tick does not prove that a result reached your client quickly. SHADOWFRAG measures these stages separately. No single ping, tick-rate or bandwidth number describes the entire experience. Explore the netcode and its retained measurements.

Networking wraps data in packets with the information needed at each layer. SHADOWFRAG game records are encoded inside QUIC traffic carried by UDP and IP. Their boundaries need not match: one delivery can carry several records, while a large scene needs multiple fragments.

Losing a packet can therefore affect several records or leave a scene incomplete. The game needs identities, fragment validation and appropriate recovery rather than assuming each received packet is one new accepted action.

Ninja stars let you make four quick throws, then require a reload. Each star has travel time and server-checked contact, so a fast sequence still asks you to lead or track a moving target.

Magazine and reload state belong to the game rules. Switching weapons does not create ammunition or bypass the shared weapon timing. A star-heavy networking fixture produces frequent projectile launch and terminal records; its measured cost depends on the encounter rather than a universal price per star user.

Every Shogun settlement has two linked outposts. An attacker needs both to open the settlement, giving a smaller group useful work away from the main court.

An outpost phase takes eight uncontested seconds. Neutralizing an enemy owner and then claiming the point are separate phases; extra bodies do not accelerate either. Outposts enable attacks and supply, but do not count as settlements for territory ticket drain or settlement income.

Sort measured values from smallest to largest. The 99th percentile, p99, lies near the point where 99% are at or below it. For example, a 30 ms feedback p99 says roughly 99 out of 100 measured feedback times were no more than 30 ms; some can be higher. Histogram buckets and calculation methods affect precision.

The current capacity report takes the largest p99 calculated separately for each client. That is a worst-client percentile, not a percentile pooled across all clients and not the maximum observed sample. Population, duration, sample count and workload remain essential context. p99.9 uses the corresponding 99.9% point.

p99.9 is the 99.9th percentile. It looks farther into the slow end of a distribution than p99: about 999 out of 1,000 measured values are at or below it, subject to the calculation and histogram resolution.

The capacity screen uses p99.9 server tick time alongside p99 and other criteria. Rare-event interpretation needs enough samples and independent repeated runs; displaying more decimal places does not create that evidence. A server tick percentile also says nothing by itself about client state freshness or physical input response.

Pacing spaces traffic over time according to transport and rate-control decisions. It differs from congestion control’s limit on outstanding traffic and from a configured application byte budget.

SHADOWFRAG’s native adapter must honor each backend’s contract. Prepared Quiche packets carry their scheduled send time, while a successful nonempty PicoQUIC preparation is already ready under its native pacing. A previously cached wake must not delay that ready packet. An adapter timing histogram is meaningful only for the particular stage it measures.

A network packet can disappear or arrive too late to be useful. The application and transport must decide what to repeat, what to acknowledge and what can expire.

SHADOWFRAG repeats recent movement commands, tracks complete-scene receipts and recovers durable action and projectile results. Cosmetic cues can have a shorter useful lifetime. A QUIC counter labeled considered-lost records the transport’s inference; it is not a direct count of physical network drops, because reordering and delayed acknowledgement can influence that inference.

Different information needs different recovery. Recent inputs can repeat until acknowledged. A newer complete scene can replace an old partial one. A projectile launch or terminal record needs its own receipt so eventual outcomes remain known.

Retries create traffic and work, and can worsen queues if unmanaged. SHADOWFRAG bounds recovery and keeps duplicate identities, with separate timing for projectile lifecycles and action or hit results. Receiving another copy must not execute movement, apply damage or play confirmed feedback twice.

The current implementation allocates stable pages for player and projectile state. Growth preserves numeric identities and borrowed references; reused storage is reset before exposure. Simulation, history and presentation follow the constructed extent.

This removes work from ordinary smaller matches without changing movement or combat rules. It does not promise that every operation scales only with active players: a sparse state using a high numeric slot can still involve a larger prefix. Storage capacity also does not authorize a higher playable population.

A before-and-after pair can use the same map, camera, equipment schedule and machine. An active/unused feature pair keeps declared permissions while changing whether the feature is exercised. The exact control matters: changing the workload or cohorts can introduce other differences.

Background activity and run order can still affect results, so repeated clean pairs and retained failures improve interpretation. Deaths or replenishment can change how many actions actually occur. A measured traffic difference is conditional on that encounter and cannot automatically become a fixed per-user price.

PCM means pulse-code modulation. In the game’s format, each waveform sample is a 16-bit value and 48 kHz means 48,000 samples per second per channel. Mono stores one channel. This sample rate is unrelated to simulation ticks or display refresh.

Sound Studio converts supported WAV or MP3 imports to the prepared runtime format and keeps their originals. WAV is a file container and MP3 is a compressed audio format; naming an imported file does not mean the callback decodes it during play. Prepared PCM consumes memory in exchange for predictable playback work.

A QUIC implementation is library code that handles the transport’s connections, packets and delivery mechanisms. SHADOWFRAG’s adapter connects PicoQUIC to its own scheduling and gameplay protocol.

Current launches default to PicoQUIC/CUBIC after retained direct and impaired comparisons; Quiche/CUBIC remains selectable. This does not establish that one library is faster on every machine or connection, nor change combat validation or the 64-player limit. The upstream project describes its scope in the PicoQUIC repository.

SHADOWFRAG applies the same 64-player bound across setup, planning, startup and admission. Bots occupy places within that total. A setting imported from a larger stress fixture does not authorize a larger ordinary match.

Three clean 128-client Highlands FFA confirmations establish a conditional wired-LAN research result with declared equipment and synthetic clients. Dense higher-count encounters and full Shogun AI need separate evidence. Research can inform a future limit; it does not silently change the current one.

Position age measures freshness of the world state reaching clients. The current capacity campaign reports it independently from server tick work and weapon-request feedback, because each can fail while another looks healthy.

The published aggregate selects the largest individual-client p99. For example, a 25 ms worst-client p99 does not mean that every position for every client arrived within 25 ms, or that the scene was drawn on a monitor within that time. It remains conditional on the test’s clock estimate, workload and sampling.

Precision Dojo runs in the client without starting a server or contacting an online service. Foundations isolates aiming problems; Aerial Mastery puts you and opponents into flight, with optional return fire and automatic chi relaunch.

Your sensitivity, bindings, field of view and weapon rules apply. Cadence changes the exercise pattern rather than scaling hitboxes or projectile speed. The dedicated training space is Stillwater Court. It is separate from the multiplayer map named Dojo, historically Cloud Dojo, and does not add another map to the multiplayer roster. Mean time alive measures eliminated practice ninjas’ lifetimes, not reaction time.

Projectiles occupy the world over time rather than resolving as an instant rail trace. Their launch speed, gravity where applicable, lifetime and collision rules determine where they can go.

SHADOWFRAG’s client predicts the visual flight for immediate response; the server approves the launch and checks contact with cover and moving players. It checks the interval between simulation steps, not only isolated positions. Projectile travel time is intentional gameplay time and is separate from the delay in receiving launch confirmation. Nets travel as projectiles but do no damage.

The launch record gives the client the approved origin, velocity, timing and identity needed to reconstruct flight. A terminal record says the projectile hit or expired. The detailed path can be drawn locally instead of streaming every position.

Either record can arrive first under reordering. Retaining the terminal identity prevents a delayed launch from resurrecting a finished projectile. Lifecycle recovery differs from the action’s launch-acceptance and hit-result feedback, and every path still leaves collision and damage with the server.

A qualification needs declared criteria and evidence appropriate to the claim. The current capacity screen checks server work, state age, feedback, generator timing, client validity and input handoffs, then requires clean repeated confirmations.

Passing establishes the declared research envelope, not every map, mode, machine or connection. Full rendering, audio, Shogun AI, hardware-event latency and Internet conditions need their own integrated evidence. The game’s shipping player cap is a separate configured policy. A missing measurement is unavailable, not a zero or a pass.

Input, encoded records, native transport output and renderer submissions can each wait in separate queues. A stage accepting an item does not prove the next stage sent, decoded or displayed it.

SHADOWFRAG bounds queues and gives urgent results, fresh scene fragments, durable records and cosmetics bounded service turns. Backpressure means a later stage cannot currently accept more work. Keeping everything forever can increase age and memory; replaceable stale scenes and durable outcomes need different policies.

QUIC provides encrypted connections and reliable streams over UDP. Its datagram extension also carries messages without requiring retransmission by the transport, useful when a newer movement scene can replace an older one.

SHADOWFRAG places prediction, shot references, deadlines and application receipts above that transport. QUIC delivery cannot by itself prove that the game decoded a scene or accepted an action. A transport name is not a measured latency result. See the QUIC standard and datagram extension.

Quiche is a QUIC implementation used through SHADOWFRAG’s native transport adapter. It carries the same game-level movement, references, identities and receipt rules as the other backend.

Quiche/CUBIC remains selectable while PicoQUIC/CUBIC is the current default. A measured change in one backend’s adapter cannot be claimed as a LAN speedup for the other backend. Comparisons need matched builds, conditions and timing stages. See the upstream Quiche project.

A rail shot uses a hitscan trace rather than an arrow that takes time to reach the target. The server checks the aim, permitted origin, cover, historical target reference and current damage eligibility before confirming the outcome.

Shinobi, Spear, Thunder and Arctic are cosmetic rail models with the same weapon rules. A lingering beam does not keep dealing damage, and changing models does not bypass the cooldown. Railguns are excluded from Shogun’s equipment pool.

SHADOWFRAG’s mouse aim does not add smoothing or a frame-time multiplier. Your configured sensitivity maps movement to rotation, and per-weapon or zoom settings can adjust the aiming setup.

Direct input handling is only part of response. Device delivery, operating-system events, service scheduling and the display path still take time. A steady crosshair or high frame counter therefore does not by itself measure physical mouse-to-display latency.

A network worker handles transport work outside the drawing loop. Native readiness mechanisms notify it when a socket can do useful work; queued output and deadlines can wake it too.

The current implementation replaces an unconditional 100-microsecond idle polling ceiling with useful waits. It still has to serve fresh scene fragments, urgent results, durable state and cosmetics fairly. Sleeping when idle does not allow movement deadlines, authenticated ingress checks or pacing to be skipped.

A reliable application record remains recoverable until its valid acknowledgement arrives. The journal tracks that identity, retry timing and whether an unsent attempt is already owned by the application or native adapter queue.

The current rule permits at most one such queued plaintext attempt per receipt. Acknowledgement can cancel that ownership, but bytes already packetized by QUIC cannot be recalled. This therefore does not promise that the wire carries only one copy. Projectile-lifecycle timing and action or hit-result timing remain separate.

A server update identifies the authoritative movement state and which commands it has already executed. The client restores that state, then replays only the newer commands still awaiting acceptance.

The replay must retain details such as chi readiness, bow state and shadow transitions so the same inputs produce the intended movement. A correction may be visible if the views diverged. This local command replay is narrower than rolling back an entire match, and is separate from playing a saved demo.

A reliable stream presents ordered bytes even when packets are lost or reordered. QUIC can maintain separate streams within a connection; application framing still tells the game which records those bytes contain.

Reliability does not make delivery instant. Missing earlier bytes can delay later bytes on the same stream, while congestion and queues still affect the connection. SHADOWFRAG combines reliable control with datagrams and game-specific receipts so different information can follow its appropriate lifetime. See the QUIC transport standard.

SHADOWFRAG’s rendering consumes immutable scene publications. A successful handoff to the application’s bounded CPU renderer queue records the submitted view used by the gameplay reference contract.

Further rendering, GPU execution, compositor and display work follow. Cancelling or superseding later presentation does not retroactively replace a reference already assigned to input. Keeping these stages distinct lets the game preserve action chronology without pretending to measure actual monitor pixels.

A revision identifies a particular state of an editable document or asset. Map Studio instances pin the asset revision they use, so importing a replacement does not automatically rebuild existing routes around it.

Background compilation also checks the document revision before replacing a preview or launching a playtest. Texture Workshop’s undo refuses to overwrite files changed since its batch. These guards make authoring changes reviewable and reversible; they do not themselves prove that the chosen revision is balanced or correct.

A joining viewer receives a reliable baseline identifying participants and their round-locked appearance. Later score-only updates can omit data the client already knows. An appearance chosen in settings applies next round rather than changing the active wardrobe with every movement update.

The baseline is tied to occupant identity, so a reused slot cannot inherit a former player’s name or look. A baseline elsewhere in networking means the accepted starting state for later updates; it must be known and valid before anything depends on it. Startup traffic is separate from steady-state measurements.

RTT measures a journey to another endpoint and back. A timing probe or transport sample can estimate it without subtracting unrelated clocks on two machines. People often call a ping measurement RTT, although tools may include different processing.

The two directions can have unequal delays. Game feedback also includes scheduling and validation, so it is not simply the path RTT. SHADOWFRAG uses validated timing observations for bounded recovery estimates; the estimate cannot authorize an action beyond its gameplay deadline.

Sanitizers add checks to a development build so invalid memory use or other selected programming errors can become visible during execution. They help inspect complex movement, terrain and replay paths beyond whether a test produced the expected answer.

Different sanitizers detect different problems, and instrumentation changes runtime cost. A sanitizer result is therefore correctness evidence for the exercised paths, not a release-performance benchmark or proof that a build has no possible defect.

Scene age asks how old the delivered world description is, using its simulation timestamp and the client’s estimated server time. It exposes delivery, assembly or scheduling delay that a server tick timer alone cannot reveal.

It is not the time to render the scene, a physical mouse-to-display measurement or necessarily the extra buffering used to draw remote animation. A percentile also describes a distribution rather than a maximum. Read each study’s exact sampling definition and clock assumptions before comparing values.

A seeded generator produces a defined sequence from its starting value. Precision Dojo uses the same setup and seed to reproduce the same uninterrupted movement pattern, helping you change one practice condition at a time.

Capacity workloads use seeds for declared action schedules and offsets. Real scheduling, network timing, deaths and contacts can still differ, and the seed does not replace the server’s independent combat-order key. Reproducing a whole match requires more evidence than one seed value.

The server tick timer measures the work included in the harness’s whole-tick boundary. At a 128 Hz simulation, the step represents 7.8125 ms of game time; the actual computation can take much less or overrun its intended schedule.

A short tick is necessary evidence about server workload, but transport queues, fragmented scene assembly or client scheduling can still fall behind. The capacity study therefore checks tick percentiles alongside position age, feedback and generator timing. A whole-tick timer is not a separately instrumented receive-to-enqueue critical path.

Each settlement has two supporting outposts. An attacking team must own both without a contest before neutralizing or claiming the settlement. An enemy-owned point must first be neutralized, then captured.

A settlement phase takes twelve uncontested seconds in the current rules. More teammates do not speed it up, and opposing presence stops progress. Cutting supply does not immediately erase settlement ownership: an attacker must still take the point. The court is the eligible capture area, not merely its map marker.

Shadow Council brings together several kinds of contribution. The Balance Council playtests rules, Map Makers build and refine arenas, Server Moderators administer matches, and the Content & Clip Team makes original guides and footage. Community Shoguns organize regional activity.

These are community roles, not combat classes or Shogun equipment. Anyone can request an invitation; requesting one does not guarantee access. The team reviews Council introductions and priority personally. See the current roles and invitation process.

A shadow frag requires actual body overlap at the teleport destination. Passing near someone during the throw or drawing a trail through them does not count, and cover still matters to the arrival contact check.

This is an exception to ordinary armor-then-shield damage: an eligible arrival kills through both layers. Teammates and players under spawn protection are safe. Warping ends the attacker’s own spawn protection. The server confirms the kill; a locally drawn smoke effect does not award it.

Press Q with the default bindings to throw a shadow grenade. After its fuse, your ninja warps to the allowed destination. The throw is a movement ability, separate from shrapnel ammunition, and access can be configured by the match rules or bought in Shogun.

The game checks the destination against the full player hull, so a visible gap is not automatically a legal arrival. Momentum continues after the warp, but you do not physically travel through the space between its endpoints. Smoke explains departure and arrival. Only arrival overlap can create a shadow frag; the grenade’s flight and the trail do no damage.

A shield turns an otherwise lethal ordinary hit into a shield break. Under kill-recharge rules, earning a kill restores your shield; merely waiting does not. The server decides the current shield state and confirmed hit result.

Armor is a separate layer above the shield and does not repair it. Spawn protection is also a separate eligibility rule, not another ordinary shield layer. Shadow frags bypass armor and shields but cannot kill a spawn-protected player.

Shogun spreads a match across six settlements and twelve supporting outposts on Shogun Highlands. Own both of a settlement’s outposts, with neither contested, to unlock its attack and maintain forward supply.

Teams begin with 400 tickets. Deaths cost tickets, and holding fewer settlements causes periodic territory drain. Gold survives death; bought equipment lasts for the current life. Railguns are unavailable. The current limit is 64 players including bots; the separate 128-client Highlands FFA study does not qualify full Shogun AI or Internet play. Read the mode guide.

A Shogun entrant starts with 150 gold. Eligible players earn income away from their own supplied buy zones, receive settlement income and can earn capture rewards. Purchases are available during the respawn wait and at supplied friendly camps.

Gold carries between lives and rounds in the match, but a new match resets it. Bought equipment does not survive death. Shadow is purchased as regenerating access for that life, rather than one consumable throw. The server validates prices, available equipment, location, identity and balance; a displayed shop preview cannot grant an item.

Shogun Highlands connects six settlement districts through ground routes, six bridge crossings and exposed skyways. Forests, villages, rivers and ridges divide long views into local fights. Every objective remains reachable on foot.

The landscape can also host other objectives. The 128-client research confirmation used spread-out FFA here, with synthetic protocol clients rather than full rendered Shogun players or squad AI. It is one declared workload, not a larger general server permission. Explore the current map atlas.

Shrapnel grenades travel through the world and burst when they contact the world or an eligible opponent. The blast and fragments make cover and approach angle matter.

Direct contact, blast and fragments share a victim set, so one explosion cannot remove several protection layers from the same player. Ammunition and replenishment depend on the match. A networking study with frequent respawns may permit different throw counts from another encounter, so it cannot establish one fixed bandwidth cost per grenade holder.

Shogun Highlands has four bidirectional skyway connections linking distant outposts. Launch pads start the crossing and high rings support the route. Timed air steps help you stay in the stream; chi can break away.

The last ring releases you above the approach instead of placing you on an arrival platform. You can shoot or throw equipment while descending, but a missed route can leave you over a canyon. Walking routes remain available to every objective without purchasing a mobility ability.

A smoke test exercises a concrete path, such as opening the native menu, entering values, changing a setting or loading a playable scene. It is useful after a broad change because an otherwise successful build may still fail at startup or interaction.

SHADOWFRAG’s native UI checks also inspect layout and interaction across recorded frames and resolutions. Passing those paths is evidence of function, not proof of every possible user action, final balance or Internet capacity. More focused rule tests and human play answer different questions.

A network snapshot records gameplay facts at an identified simulation time. SHADOWFRAG’s combat scenes carry the canonical player state needed by the recipient, while private owner corrections and other records have separate jobs.

The client validates and installs a complete scene, then uses timestamped states to draw remote motion. A large scene may arrive in several fragments, so receiving one packet is not proof that the whole scene decoded. Snapshot publication rate, simulation tick rate and rendered frame rate are distinct.

The client receives discrete snapshots but draws many images between them. Interpolation chooses a point between eligible earlier and later states to reconstruct smooth remote motion. Its buffer deliberately draws a recent past moment rather than guessing unlimited future movement.

Buffering trades some age for continuity and cannot repair arbitrary packet loss or prolonged stalls. Teleports and respawns mark real discontinuities, so the game must select the proper endpoint rather than draw a player sliding through the gap. Statistical interpolation in the cost calculator is a different use of the word.

The server can advance game rules more frequently than it publishes a scene for clients. Current SHADOWFRAG movement and combat simulate at 128 Hz, while scenes are published at 64 Hz.

A publication is not an arrival guarantee. Encoding, service turns, transport, loss and assembly all affect when a client can accept a complete scene. Interpolation can draw smooth remote motion between accepted states. None of these rates dictates how many images your GPU renders each second.

Sound Studio organizes cues and recordings, compares the current and replacement take, and edits levels, trims and fades. WAV or MP3 imports are converted to the game’s mono 48 kHz PCM format while retaining the originals.

A worker validates replacement data outside the audio callback. Playing one-shots retain their samples; subsequent triggers use the new bank and ambience crossfades. Applying a local swap is different from explicitly publishing assets for later embedded builds. Reserved design slots stay inactive until the game has a trigger for them.

Reach five kills in one life and the selected spree aura becomes active. Its shapes and color add a recognizable effect around your ninja.

The aura is presentation. It does not expand a hitbox, protect you or increase weapon power. Clients draw the detailed effect locally from game state rather than receiving every particle position over the network. A dated render benchmark measures only the aura designs and crowd it actually used.

Supply makes territory useful for reinforcement and purchases. A friendly settlement needs its linked outposts secure for its forward camp to remain supplied.

Raiding an outpost can therefore remove a useful spawn or shop and interrupt an attack before the settlement itself changes hands. A settlement may remain owned while unsupplied. Supply is an objective rule, not a bandwidth budget or an ammunition pickup. The tactical map shows friendly supply so you can plan where to reinforce or buy.

Testing only an arrow’s position at each tick could miss a target crossed between those positions. A sweep checks the intervening path. Relative sweeps also account for a player moving during the interval, helping catch fast contacts consistently.

The game applies appropriate shapes and cover checks rather than treating every visible effect as collision. A shadow warp is a discontinuity: the path before departure must not be joined through the world to the arrival point. Its arrival overlap check is separate.

A protocol-client generator creates controlled workloads over actual native connections. Its scripts submit movements, weapons and abilities while recording state age, feedback and the generator’s own timing.

It is different from a human running the complete client and from a server-controlled bot with tactical AI. If the generator misses its own input schedule, the run cannot establish a server ceiling. The confirmed 128-client Highlands workload is therefore conditional networking evidence, not a qualified 128-player Shogun match.

Most actions may finish promptly while a smaller group waits much longer. Those slow observations form the latency tail. A lower average can coexist with a worse tail, which matters in a shooter when an occasional action feels late.

p99 and p99.9 select points toward this slow end; neither is the maximum. A short run may have too few independent observations to say much about rare delays. Retained SHADOWFRAG comparisons keep client-specific tails and failed runs rather than using one improved median as a promise for every player.

A fixture fixes the setup, such as map, camera, actor poses, transport conditions or a clock. Repeating it helps determine whether a change affected the intended operation instead of an unrelated encounter.

The retained completed-frame render fixture waits for GPU completion but excludes live input, HUD, audio and networking. A fixed-clock adapter regression can expose a scheduling bug without measuring a LAN speedup. Always keep the fixture’s scope attached to its result.

Texture Workshop lets an author compare surfaces in an actual map, change role assignments such as stone or timber, undo a choice and save a reusable preset. It preserves geometry and collision.

Publication stages source documents, atlases, compiled packages and catalog digests before replacing installed files. Failure and undo rules protect intervening edits. Textures are embedded in the resulting map package; players do not need the workshop library or a new runtime texture-pack setting to load it.

A texture supplies surface image data; a material describes how a surface is represented. An atlas packs several texture regions into one image, and UV coordinates say where mesh points sample that image. Baked shading stores appearance work prepared ahead of runtime rather than recalculating all of it during play.

Texture Workshop assigns choices by semantic role, such as roof, timber or ground. Its surface changes preserve geometry and collision. The compiled map embeds the required textures, so players do not need the author’s complete library or an extra texture-pack setting.

The Crimson Path introduces SHADOWFRAG by giving its systems a use in a mission: recover a scroll, cross a sky route, use a bow and grenades, then learn why smaller captures can open a larger objective.

Master Hejian is the tutorial guide. Stage-specific narration and captions explain the current task. This story experience has its own training space and is separate from the multiplayer map roster. It complements Precision Dojo, which repeats a chosen aiming or movement problem without the same mission sequence.

A simulation tick advances movement, collision and other rules by a defined time step. At 128 ticks per second, each step represents 1 ÷ 128 seconds, or 7.8125 ms. Client prediction and server movement share those rules.

That interval is a budget of simulation time, not a claim that the server spends 7.8125 ms computing every step. Scenes are currently published at 64 Hz, rendering runs at its own cadence, and some objective work runs less frequently. A higher tick rate alone cannot remove delivery queues or connection delay.

Both Shogun teams start with 400 tickets. A death costs one. Every five seconds, the team holding fewer settlements loses two tickets for each settlement of difference: four against two drains four from the trailing team.

Equal settlement control causes no territory drain. Outposts affect access and supply but do not directly count for this calculation. At zero tickets a team loses the round; simultaneous zero is a draw. Tickets are shared team state, separate from each player’s gold.

A well-timed fresh Jump press around landing can carry momentum into the next leap. This lets a practiced player join movement actions into a route rather than stopping after every arc.

SHADOWFRAG also allows held Jump to chain ordinary leaps, so learning timed hopping is different from merely keeping the key down. Landing buffers make a small timing window usable; they do not grant unlimited speed or change the server’s movement rules.

The transport handles connection and packet delivery mechanisms. SHADOWFRAG’s native adapter connects a QUIC library to socket I/O, pacing, readiness and bounded queues. The game above it still defines commands, scenes, identities, hit rules and application receipts.

A transport accepting a record is only one stage: it does not prove the bytes left the machine, a complete scene decoded or a hit was accepted. Comparing backend names without matched measurements cannot establish which gives better player feedback.

Delivery needs information that is not itself a movement command or hit result. IP and UDP add addressing envelopes; QUIC adds its own framing, protection and control traffic. Recovery and acknowledgement also consume traffic.

The explorer separates known game records and IP/UDP headers, then reports the remaining received bytes together. Sampling-boundary buffering can affect that residual. It would be incorrect to label the entire remainder encryption, or to assign precise percentages to acknowledgements without finer counters.

Traversal aids make the map part of a movement chain. A launch feature can send you toward another level, while rings can sustain a fast crossing. Their shape and placement tell you what route they support.

They do not automatically carry you safely through every connection. Use an air step, chi or a deliberate exit to choose the next move. On Highlands, ring routes form exposed skyways, while every objective also has a walking route. Different aid types have their own behavior; a ring is not a shadow teleport.

The viewmodel shows your weapon handling close to the camera. An observer model shows the weapon held by a ninja in the world. Their poses and camera needs differ, especially for a drawn bow or an ultrawide screen.

SHADOWFRAG’s bow presentation pass measured coverage of a defined central aiming region across ready, drawn and zoomed poses. Reducing that coverage made the target area clearer without changing the gameplay fingerprint. These models explain the weapon; they do not independently change its collision, damage or draw rules.

An application programming interface, or API, is a defined way for software components to communicate. Vulkan is the graphics API used by SHADOWFRAG’s renderer and Map Studio. The native engine is written in C++20 and uses SDL3 for platform services.

The renderer draws game geometry, materials, weapons and local effects. Its performance depends on the actual workload, implementation, hardware and driver. A fixed-scene optimization measured under Vulkan is useful evidence for that fixture, rather than a universal frame-rate promise or a network latency result.

A shadow warp arriving beside a wall refreshes chi and opens a short wall-jump opportunity. Press Jump to kick upward and away, or use the refreshed chi launch instead. These consume the same refreshed launch opportunity; they are not two free independent launches.

The air step is available afterward, so a wall arrival can become the middle of a route rather than its end. A wall somewhere in view is insufficient: the movement controller checks the arrival and continued proximity.

A benchmark can run for a declared warm-up period before collecting its main sample. For example, the archived small-match traffic study used 15 seconds of warm-up followed by 60 seconds of measurement.

This boundary does not make startup free. Joining, asset agreement and baseline delivery may happen outside the sample. Actions begun during warm-up can also produce feedback inside an arrival-based measurement window, so those counters are not necessarily a closed set of action lifecycles. A gameplay warmup phase is a related pre-match state rather than the same measurement procedure.

The harness creates an attack request, then times the returned disposition: whether the action was accepted or rejected and what result is available. Projectile launch acceptance is separate from a later contact, so an arrow’s intentional travel must not be counted as network confirmation delay.

The capacity tables report the largest individual client’s p99, not a pooled percentile or maximum. Earlier studies use labels such as weapon-disposition; always check their stated timing endpoints. These software observations do not establish USB input, monitor presentation or total human-perceived response.

The word wire is only precise when its boundary is stated. In the traffic explorer, received IP bytes are split into application records, IP/UDP envelopes and a transport remainder. The archived IPv4 fixture assigns 28 header bytes to each UDP datagram.

This is not a measurement of every physical-link byte: Ethernet and Wi-Fi framing sit outside that accounting. Sent and received counters, sample boundaries and buffering can also differ. A record is not a packet; several game records may share a packet, and a scene may require several messages.

A wire protocol specifies how bytes represent commands, identities, scenes and results. A version change allows builds to reject incompatible layouts instead of interpreting the same bytes differently.

The current native game protocol is 37. The website also retains dated wire-24 and wire-27 byte studies. Their layouts explain those captures, but their sizes and defaults should not be silently presented as current. The QUIC transport version is a different layer from SHADOWFRAG’s application protocol.

A map described as 3,000 × 2,520 world units is measured in the simulation’s coordinates. A leap distance in world units uses that same scale, letting developers compare a route with the movement needed to cross it.

World distance is separate from pixels, which depend on camera angle, field of view and resolution. A landscape’s dimensions alone say little about the number of useful fights, the time to reinforce or the supported player count.