- articleWhat is the difference between "state consistency" and "behaviour consistency" for a text world model, and which of the two papers (arXiv 2604.13824, 2610.06637) defines which notion?
Both IDs resolve to the described papers. arXiv 2604.13824v2 (Huang et al.) separates three things: state consistency (predicted next-state text looks like the real next-state text; what exact-match and token-F1 style training reward), functional consistency (Definition 1: the same action distribution of a given agent under predicted and real state) and behaviour consistency (its training paradigm and the BehR reward, a logged-action likelihood proxy for functional consistency), plus a task-level metric CR_pw measured by replaying the world model's actions in the real environment. arXiv 2610.06637v1 (Wang et al.) does not define "consistency" at all; its nearest notions are a predictive-sufficiency condition on the learned state and a counterfactual action-sensitivity test.
- articleHow does a developer actually get access to Odyssey Systems models (Odyssey-2 Pro, Odyssey-2 Max, Odyssey-3), and which of them does the self-serve developer API document?
At the captures read (2026-10-22), Odyssey documents a self-serve developer API (JavaScript and Python SDKs, API key from a developer portal) only for Odyssey-2 Pro. Odyssey-2 Max is stated to be in private beta for selected partners, and Odyssey-3 is a research preview in a signed-in app plus a contact route. The "Get API Access" link on the Odyssey-3 launch page leads to the same developer portal, whose documentation index names Odyssey-2 Pro only, so no documented API for Odyssey-3 was found.
- articleWhat baseline (BF16) generation times does the QuantWM paper report for LingBot-World-v2 and HY-World 1.5, how were they measured, and can they be compared with the authors' real-time claims or with WorldRoamBench?
QuantWM v3 (Harbin Institute of Technology Shenzhen / NUS; not the model authors) reproduces released models with their official repositories on NVIDIA A800 80GB GPUs and reports BF16 end-to-end generation latency of 193.56 s for LingBot-World-v2 and 112.17 s for HY-World 1.5 (Table 7), with the main evaluation at 480p and 720p, 93 frames. That is 0.48 and 0.829 frames/s (own arithmetic), far below real time, but the Table 7 text does not give GPU count, resolution, checkpoint (distilled few-step or not) or step count, so it cannot be set against the authors' real-time statements. At WorldRoamBench's single-H20 Hz the same 93 frames would take about 52.0 s (LingBot-World) and 71.5 s (HY-World 1.5); the ratios 3.73 and 1.57 are unreconciled.
- articleWhat do users report when they time HY-WorldPlay, Hunyuan-GameCraft and Matrix-Game-2 themselves (including the offloading, compile and consumer-GPU cases), and how do those logs compare with the maintainers' answers?
User logs show large, flag-dependent gaps. HY-WorldPlay AR-distilled (480p, 125 frames, 4 GPUs): 3-4 minutes reported on 4xH800 and 4xH20 versus about 35 s measured by a maintainer on 4xH20; a user says turning offloading off gives about 35 s (about 3.57 frames/s), and removing repeated transformer CPU-GPU moves cut chunk 1 from 33.59 s to 5.89 s in another report (maintainers merged changes). The Wan2.2-5B variant on 4xA100 shows 90-368 s for the first four chunks and about 3.4 s per chunk afterwards (about 4.71 frames/s if a chunk is 16 frames); a maintainer blames torch.compile start-up. Maintainers say the code is not real time on an RTX 4090. Hunyuan-GameCraft users report about 12 minutes for one action on a 4090 (low-VRAM script) and 40 minutes for under 60 s of video on 8xH20; a maintainer says 8xH20 takes 2-3 times longer than their 8xH800 setting. Matrix-Game-2 users report about a minute of stall before the first response; no user FPS figures were found.
- articleHow do anime-look 3D productions edit vertex normals to get clean face shading, which methods exist, and what do they break?
The Xrd developers state they edited normals on every major feature and hand-crafted face normals, and did not use normal maps. A Japanese ASW lecture deck (published 2025-07-30; glossed here) names three control methods: place vertices so automatic normals come out right, transfer normals from another mesh, and set normals by hand (reserved for faces; whole-body is called impractical), and lists limits: normals rotate with bones, no subdivision surfaces, extra care with hull outlines. Blender documents Normal Edit (with Max Angle to avoid shading artefacts, described as useful for toon-like shading) and Data Transfer (custom normals).
- articleWhat goes wrong with inverted-hull outlines in production (width vs distance and FOV, hard edges, custom normals, floating lines, fabric) and what fixes did a studio document?
The ASW English deck (published 2025-11-28; 79 slides) lists concrete failure modes and fixes for the hull: scale width with camera distance and FOV (tan(FOV/2) factor, FOV from the projection matrix), control width with a vertex-colour channel (default 0.5), push the back-face mesh away from the camera to keep only the outer rim, use a second normal set for the hull when hard edges or customised normals break it (stored in the tangent channel at import), and set the width to 0 in valleys to close gaps that make lines float. Unity Toon Shader exposes parallel controls (width map, Z offset, distance fade, baked normal map for outlines). The deck says it avoids code and is engine-independent; the second-normal-in-tangent approach needs an engine import change.
- articleWhich outline method should I start with (inverted hull, post-process edge detection, Freestyle, Line Art), and what does each trade away?
A practitioner comparison from Arc System Works (speaker Motomura, deck published 2025-11-28) sets post-process outlines (constant line width, lines at mesh intersections, little per-part control) against the inverted hull (per-area control, no intersection lines, thickness upkeep). Xrd used the hull for per-vertex control and in-DCC preview. Unity Toon Shader implements two hull-style modes (extrude along normals, or scale). Blender offers Freestyle (render-time) and Line Art (stroke-based). Epic's post-process documentation shows a CustomDepth/stencil silhouette; it is a general doc with no toon guidance. These are statements by tool makers and one studio; no source compares the methods on equal footing.
- articleHow does toon/cel shading actually work (a stepped ramp on lighting), and what are the first controls to set in Blender, Unity or a custom shader?
Three primary sources describe the same core: lighting is reduced to a hard step. Motomura (Arc System Works, 2015) describes the Xrd shader as a step function of a threshold, a light vector and a normal, with artists painting the threshold per vertex. Blender documents a Color Ramp (Constant interpolation) after Shader To RGB (EEVEE only, with stated PBR costs) as a toon shader. Unity Toon Shader documents base/1st/2nd shade colours with step and feather controls and shade colours independent of light colour. The sources do not give numeric defaults; they treat the look as an art-directed choice.
- articleWhat do the official 5.1.7 Movie Recorder pages state about ProRes, colour and tonemapping, and which part of CA-B309's "ProRes hard-codes Rec. 709" is only in mirror source?
The official Movie Recorder page lists six ProRes codecs, says the ProRes encoder supports Stereo only, does not support variable frame rate, and is unsupported on Linux and Windows Arm64. It contains no colour, tonemapping, Rec. 709, Rec. 2020, PQ or HLG statement (0 hits in the whole 15,316-character page; 0 also in the linear-sRGB page for 709/2020/HLG/PQ and 0 in the changelog apart from dates). CA-B309's "Movie docs contain no colour-tag or tonemapping setting" is corroborated as an absence; "ProRes encoder hard-codes Rec. 709 with Rec. 2020/PQ/HLG commented out" is code-only (ProResEncoder.cs in the mirror) and cannot be byte-checked without the archive.
- articleWhich of CA-B309's Image Recorder colour-space statements (PNG/JPEG sRGB, EXR linear unclamped, tonemapping, dithering) do the official 5.1.7 pages state, reword, omit or contradict?
Official pages state the conditions for the Linear, sRGB (unclamped) option exactly as CA-B309 summarises them: HDRP project, Source not Game View, EXR format. They also tell you to disable Tonemapping and Dithering to get unclamped values, and say the Recorder encodes EXR in 16 bits. Not stated officially: that PNG/JPEG are "forced" to sRGB (only implied by the EXR-only condition), any default for Color Space, the internal "HDR frame capture possible" test, and any statement that the Recorder embeds or omits colour tags. The two official pages disagree on UI paths for the same step (HDRP Default Settings vs HDRP Global Settings; Dithering under General vs Rendering).
- articleWhat does Unity's own documentation say about the Audio Recorder's sample rate and bit depth, and which part of CA-B309's audio claim rests only on mirror source code?
Official Recorder 5.1.7 docs state: WAV is the only Audio Recorder format, only Mono or Stereo is supported (from Project Settings > Audio > Default Speaker Mode), and the audio goes to the Recorder rather than the system output. They state no sample rate, bit depth or sample format anywhere in the five official pages read (0 hits for sample rate, Hz, bit depth, float, 32-bit). So CA-B309's "WAV header rate = AudioSettings.outputSampleRate at Stop(), 32-bit IEEE float" is absent from Unity's documentation and rests on AudioRecorder.cs from the mirror; the "no rate/bit-depth property" and "mono or stereo only" parts are corroborated. The mirror's RecorderAudio.md equals the official page word for word.
- articleWhich automatic voice/persona measures have been checked against human judgement, how well, and what protocol can an agent run to test whether generated characters stay distinguishable and consistent?
Across the sources read, validation of automatic persona/voice measures is thin: PersonaScore had five experts and 1,500 ratings (kappa 0.71; mean Spearman 75.1%); the consistency judge of Abdulhai et al. agreed with humans more (kappa 0.552) than humans agreed with each other (0.303); the persistent-persona judge was checked by one author (kappa 0.65, 250 ratings); RoleBench reports GPT-4 and human win rates without an agreement statistic; the Distinction Score has no human check on generated lines. An evaluation protocol that combines blind attribution, cast-level distance, drift probes and a validated judge is proposed; no source documents it as a whole.
- articleWhat exactly does RoleBench score for "speaking style", how were the human and GPT-4 judgements made, and what can its English data and licences be used for?
RoleLLM (arXiv 2310.00746v3, 18 Jun 2024) builds RoleBench: 95 English and 5 Chinese roles, 168,093 samples. Speaking style is scored as ROUGE-L overlap with GPT-4-generated reference answers (CUS), alongside GPT-4 and human win rates against a GPT-4 prompted baseline (RoleGPT). The human check used three NLP graduate students on a 500-instruction subset, with no agreement statistic in the text read. The setup is single-turn and per-role; the data card says apache-2.0 but the repository has no licence file and the scripts behind the data were not examined.
- articleIs there an automatic measure of whether a model makes different characters sound different from each other, and what did it show?
Speaker Verification in Agent-Generated Conversations (arXiv 2405.10150v2) defines a Distinction Score: embed lines the same model generated for two different roles and take one minus their cosine similarity. In self-play with RoleBench roles, every role-play system scored below the real-utterance value (72.91): from 63.57 (LLaMA2-Chat-7B) to 38.98 (CharacterLLM); the authors conclude that generated lines are similar across role settings and suggest pretraining style persists. The measure rests on verifiers validated on human-written text; the generated-line scores were not compared with human ratings in the passages read.
- articleWhat do colour-blind football fans report about TV graphics (shootouts, cards, substitution boards), and what remedies did they propose?
A 2018 focus group of 8 colour-blind male football fans, run for Colour Blind Awareness and co-funded by the FA and Fans for Diversity, reports TV penalty-shootout graphics as the clearest broadcast graphics problem: green and red balls for scored and missed penalties, with small blocks making it worse. Their proposed fix is a tick or cross. They also report difficulty telling red and yellow cards on TV and not noticing the red number on substitution boards. This is self-reported, small-sample, advocacy-organisation evidence with untested remedies; it is consistent in direction with the BBC Trust decision's accepted remedy of adding labels.
- articleWhat do the football law, UEFA, FA and World Rugby texts say to do about kit clashes for colour-blind viewers, which parts are binding, and what checks do they imply?
Binding text read: IFAB Law 4.3 requires kits that distinguish the teams and officials, without mentioning colour vision. UEFA describes its CVD practice (EURO 2020 kit selections made with teams, e.g. a black away kit for Croatia) as best effort in a news article; the Equipment Regulations article that would carry any rule was not readable. World Rugby's July 2024 document is the one text read that states a policy: for its tournaments from 1 January 2025 a dark primary kit needs a light alternate (and vice versa), with World Rugby directing the kit when a clash is found. Shared checks: greyscale look-alike test, one dark and one light kit, avoid single-colour mid red/orange/green kits, whole-shirt patterns, sock/pitch contrast, number contrast, and approval on physical samples rather than CAD simulations. TV viewing (wide-angle shots, small screens) is named as an aggravating condition, but no inspected text addresses scorebug team-colour chips.
- articleFor a series, game franchise or transmedia project, which relations between installments do the sources define (expansion, modification, transposition; snowball vs planned franchise), and what consistency checks follow?
Ryan's Possible Worlds entry defines transfictionality (migration of characters, plots or settings between texts) and uses counterpart relations and Doležel's three operations (expansion, modification/displacement, transposition) to describe it. Her 'Narration in Various Media' entry distinguishes transmedial 'snowball' growth around a central text from franchises planned across media from the start, and lists four open questions: autonomy of components, same global world vs incompatible worlds, what signals sameness of storyworld, and what stories invite transmedial growth. Toolan's 'Coherence' adds that a name does not guarantee co-reference and that characters should stay self-consistent within narrated limits. A cross-installment canon ledger is a synthesis built on these; no source tests one.
- articleWhat exactly did the WHAM paper measure for persistency, with what values, sample sizes, labelers and limits?
Nature Table 1 (observed 2026-10-22) gives the share of 1.6B-WHAM generations labelled as persisting an edited-in element: with one conditioning frame, Powercell 58%, character 45%, Vertical Jumppad 58%; with five frames, 86%, 85%, 98%. The protocol is small and narrow: 600 videos (480 for Powercell/character, 120 for Jumppad), no-op actions, 10 sampled videos per edited input, each judged only over the first ten generated frames (1 s) by seven of the authors, blinded to input length, with 90% agreement on the 300 double-labelled videos (stricter label kept). The 10-frame condition appears only as an Extended Data figure (values not extracted), and the text says persistency rose from 1 to 5 frames but not from 5 to 10.
- articleWhat failure modes of contextual NPC barks do designers and researchers report, and what (if any) tests do they describe?
First-person writers and designers report four recurring failure modes: over-clever or multi-information barks, pools that cycle too fast so lines become memorable, context-aware bark sets that feel shoehorned (Splinter Cell: Conviction), and tooling built for imagined rather than actual writers. The sources are interviews, a community blog and a short academic paper; none describes a systematic test. The nearest checks are a journalist's manual count of repeated lines and Valve's recommendation to build in-engine debugging tools.
- articleWhat memory and anti-repetition structures do shipped contextual-dialog systems document (stored facts, expiry, chance, deplete-before-repeat), and how do they stop a line from repeating?
In the Valve designer's own account, memory is just facts written back by a matched rule into a per-character (or world) store, with optional expiry; repetition is controlled by writers adding "already said" facts, expiry times, random-chance criteria, and splitting speeches into followups. The SDK 2013 code shows engine-level options (default visit-all-before-repeat in a response group, no-repeat, sequential, odds, speakonce, respeakdelay, matchonce). The speaker calls manual said-variables awkward; a Ubisoft writer separately frames repetition as depending on call frequency relative to pool size. No source here measures how well any of these prevents repetition.
- articleHow does a shipped game (Valve: Left 4 Dead, Portal, Dota) decide which contextual NPC line to play, in terms a newcomer can reproduce?
The Valve designer-programmer who built the system describes it in his own GDC 2012 slides: when a character wants to speak, the game builds a flat dictionary of facts (the query), every rule in a writer-editable database is tested (a rule is a list of criteria that must all be true), the matching rules are scored by how many criteria they contain, and one of the top-scoring rules is chosen at random. General rules act as fall-backs for specific ones. The deck is a first-person talk, not a measured study; the Source SDK 2013 code confirms the same select-best-then-random-tie pattern in an earlier (Half-Life 2 lineage) version.
- articleWhat do the primary papers leave unshown for collaborative previs or multiplayer games with a generated shared world, according to their own stated limitations and evaluation protocols?
Each paper states a limit that matters for collaborative previs or multiplayer play: Solaris and MIRA lose persistent state beyond the context; Gamma-World may accumulate inconsistencies in long rollouts; MetaWorld does not model physical contact; WorldCast does not model hits or network faults; MultiWorld has not been trained at scale. Beyond these, in the eight papers read no paper reports art-direction or director controls, a multi-human study, or output-use rights for generated clips. The last three are absences in the passages read, not demonstrated impossibilities.
- articleWhat scale (players or agents, rollout length, resolution, hardware, speed) do the primary papers actually demonstrate for a shared multi-agent world, and which numbers are measured versus shown only in figures?
Measured consistency is shown for two players in most papers (Solaris, ShareVerse, MetaWorld two views) and up to four players only qualitatively (Gamma-World zero-shot) or with a game-state probe and real-time claim (MIRA, 4 players, 20 FPS on one B200). WorldCast times 2 to 16 players per-client (FPS) but scores consistency in rounds with three to five clients. MASS reports 1,024 simulated players, but these are entities in a grid game with 128 by 128 renders, not 1,024 rendered video players.
- articleWhat from oral-formulaic composition and documented Sunjata performances can transfer to improvised or procedurally generated storytelling by an agent, and what cannot be claimed?
This is a synthesis, not a finding. From Lord's own statements: composition by a trained singer is remembering structured phrases and themes (not a fixed text) and is distinct from free improvisation; formula counts do not by themselves establish orality; comparison should be like with like. From the Sunjata sources read: versions vary around a relatively fixed core and some published texts are rewritten prose. An agent could use a bank of formulas and themes with variable slots to generate, but none of the sources read documents such use or measures its quality, and the Sunjata sources do not show composition-in-performance at text level. Transfer claims are hypotheses.
- articleHow does the Decameron frame (Rigg translation, vol. I) constrain and vary its tales, and what can an agent take for curated anthologies?
In Rigg's English the Decameron frame uses a rotating ruler per day, a stated subject (free on Day 1, "unexpected felicity" on Day 2 per the passages read), advance preparation and one teller with a standing exemption who speaks last. The plague proem is justified as contrast. Observations come from the Day 1 introduction, the Day 1-2 transition and Rigg's introductory notes; the later days and the author's closing remarks were not read.
- articleHow do the Thousand and One Nights hold a listener across nights and nest tales inside tales, and what can an agent copy for serial or nested stories?
In Burton's English the Nights use a fixed cut: the night ends mid-tale, the sister asks for the finish, and the king grants one more night. Tales nest several levels (Burton's contents show four for the Fisherman cycle) and inner tales are told by characters as arguments, sometimes opposed ones. The frame closes by consequence (children, release), not by the stories running out. These are observations from vol.1 and the conclusion in vol.10 only; the mid-volumes were not read.
- articleWhat did BioWare's David Gaider say about keeping the long-term Dragon Age lore in a master lore document, who could see it, and what problem forced it to be written?
In an Oct 2024 Eurogamer interview, David Gaider says the overarching Dragon Age plot lived in his head until Inquisition, when other writers 'dragged' it out into a master lore doc that was hidden from most of the team. The journalist reports about 20 senior people had access. Gaider says the plan has 'more or less held up' in The Veilguard, which he did not work on. The source supports a practice of a restricted master document and a single-person-memory risk; it does not show how contradictions were checked.
- articleWhat do game writers say a lore or story bible contains and how should I keep one usable for a team?
Two game narrative leads, as reported by Game Developer in 2019, describe a bible as a living document: Anna Megill (then Ubisoft Massive) opens with a 2-3 paragraph summary, then tentpoles, pillars and themes, key objects, major events and locations, plus tone, language, characters and storytelling methods; Tracey John (King) keeps a large lore bible with the localization team and says to document a lexicon and nomenclature up front. Leland Chee (Lucas Licensing) states that electronic records replaced printed bibles. These are journalist reports of talks, except Chee's own words.
- articleWhen one model is scored under several action-following protocols (Oasis and Matrix-Game across MineWorld, Matrix-Game, GameCraft and Yume papers), do the scores agree, and are the protocols comparable?
They do not agree. Oasis keyboard accuracy is 0.77 in Matrix-Game 1.0, 0.73 in Matrix-Game 2.0, and F1 0.41 (precision 0.49, recall 0.44) in MineWorld's own table. Matrix-Game 1.0 scores 0.95 keyboard/0.95 mouse in its own paper, yet in GameCraft's tests (150 mixed images, 12 actions) it has the highest translation error and the second-highest rotation error of five methods (RPE trans 0.18, rot 0.35 versus 0.08 and 0.20 for GameCraft) and the lowest user-study action-accuracy rank (1.63 of 5), and in Yume-Bench its human-rated instruction following is 0.271 against Yume's 0.657. The sources differ in test set, domain, estimator and rater, so these are four separate protocols, not four measurements of one quantity. The differences are consistent with domain shift (Yume says Matrix-Game struggles to generalise to real-world scenes) but no source isolates the cause.
- articleWhat do Ryan, Mateas and Wardrip-Fruin identify as the open problem of recognizing stories in simulation output ('story recognition', later 'story sifting'), what failure modes does Ryan catalogue, and what architecture does he propose?
The 2015 paper by Ryan, Mateas and Wardrip-Fruin (paywalled; read here only through the author's quotations in his open 2018 dissertation) names 'story recognition': simulations produce a stew of events, humans can tell which sequences are storylike, but the system cannot. Ryan later renames it story sifting because stories are not already intact in chronicles, catalogues twelve 'pains' of emergent narrative (boringness, causality issues, no curation, poor presentation and others), and proposes a pipeline of chronicler, story sifter (pattern-based or heuristic), narrativizer. Ryan notes that in nearly all existing works the sifter is the human player. The dissertation reports no controlled evaluation of the pain taxonomy in the parts read; his own case studies were not read.
- articleWhat does Dwarf Fortress's creator say about generating stories from a simulation first, what mechanisms and principles does he report, and where does he say it falls short?
Tarn Adams describes Dwarf Fortress as a world simulation whose history is a log of many autonomous agents; his stated method (2008) is to create actors with motivations and let them go, and (2019) to give dwarves autonomy so they are actors in their own stories while the player is the fortress's 'official will'. He is candid about limits: character motivations lag far behind stories, a recorded log needs post-processing to find highlights and can be boring without enough dynamics, numbers make poor stories (hence a body-part damage model), and the memorable stories come from forums and streamers, not from the game narrating them. His Game AI Pro chapter gives four simulation principles (iterate, break down the system, avoid overcomplication, use real-world analogs), but is about world generation, not narrative.
- articleWhy do players see stories in simple simulations, and what does a designer of a systemic game do so that simulated events become stories rather than noise?
RimWorld's designer argues, in a 2013 first-person article, that a game's value lives in the mental model it projects into the player, so simulation detail the player cannot perceive is wasted. Stories come from apophenia (players read intent and emotion into simple events), so the simulation acts as co-author. He gives concrete design moves: borrow archetypes, put human-relevant values and uncertain outcomes at stake, imply primal emotions, model the minimum that supports the stories you want, add 'hair complexity' flavor that feeds nothing back, and track story-richness (the share of interactions that are interesting to the player). The article is a practitioner argument with examples and no measurements.
- articleWhat do RimWorld's designer and developer say about how its storytellers choose events, and what part of the mechanism is only documented by the community?
Ludeon's own pages say RimWorld is a story generator that works by controlling random events, each one 'a card dealt into your story' by the AI storyteller; the designer's 2013 devlog names three storytellers whose policies are a pre-defined tension curve, pure randomness, and 'nothing bad happens'. The numeric mechanism (what inputs, what curve, what thresholds) is not given in any first-person text obtained; only the community wiki describes inputs such as colony wealth and time since the last major event, and a push/breathe cycle for Cassandra. No evaluation of the storytellers by the developer was found.
- articleHow do the Left 4 Dead and Warframe directors choose which events happen where and when (event selection), and what do they do to avoid repetition and wasted events?
Both directors select events with constrained randomness. L4D layers independent population functions (wanderers, mobs on a randomized 90-180 s timer, specials on individual timers, bosses dealt from a shuffled Tank/Witch/Nothing deck with no immediate repeats), picks spawn areas relative to the team's route position (75% of mobs from behind), and lets designers place candidate items that the director then enables or not. Warframe weights enemy types, caps simultaneous rare units, spawns ahead of players and out of sight, and shapes density with a planned two-peak population graph because uniform density is 'dull'. Valve's stated rationale is replayability: static triggers get memorized. The only claim about stories is that combinations of rare randomized events make memorable stories.
- articleWhat could an agent reuse from these text-to-3D-layout papers for previs set dressing or game level blocking, and which parts are documented results versus a proposed use?
Documented: Holodeck generates AI2-THOR scenes for embodied-agent training (validated by ObjectNav), SceneCraft outputs Blender scripts and tests on Sintel scenes with a video-model conditioning experiment, Graph-to-3D edits scenes through a scene graph. Not documented in any inspected paper: use for previs set dressing or game level blocking with production assets and review. Those uses are proposals; this candidate lists a small test that could separate them.
- articleHow could an agent check a frame, UI or chart for CVD problems, and what does a minimal computed check show for common palettes?
No inspected source defines a pass threshold. A defensible procedure assembled from the sources: (1) inventory colour meanings and add non-colour cues; (2) simulate protan, deutan and tritan with a pinned model in linear RGB; (3) compare each pair of meaningful colours before and after simulation; (4) treat the simulation as a prompt for human review, not as what a person sees. The author's own computation (synthesis, not a source result) applied Machado severity-1.0 matrices to six palettes: rainbow_5 collapses to a minimum CIE76 dE of 8.2-8.3 under protan/deutan from 52.9 normal, a mid red/green pair falls to 9.9 (deutan), and the Okabe-Ito set including R's added gray keeps min dE 7.1 under deutan because reddish purple and gray converge.
- articleWhat do the Xbox Accessibility Guidelines and Game Accessibility Guidelines tell a game team to do about colour-vision deficiency: presets, custom colours, filters, testing?
Both guideline sets lead with removing colour-only information. Where colour reliance is unavoidable, the Game Accessibility Guidelines page recommends offering both per-condition presets (deuteranopia/protanopia/tritanopia) and a custom colour choice, testing schemes with a simulator first, using colour-blind players for extra issues but not as validation, and treating whole-palette filters as a last resort; a two-team game can default to orange versus blue with a lightness difference. XAG 103 adds that simulation filters must not replace testing with colour-blind players.
- articleWhat does 'do not use colour alone' require in practice for a chart, UI or game HUD, and what extra cues count?
Three independent primary documents state the same rule: WCAG 2.2 SC 1.4.1 (colour is not the only visual means), Xbox Accessibility Guideline 103 ('colour alone should never be used'), and Okabe and Ito's redundant-coding principle. The extra cue can be shape, pattern, position, line type, text label or icon; the sources do not set a numeric threshold for how different the extra cue must be. XAG 103 also scopes the rule: it applies to content critical to gameplay or narrative.
- articleHow common is colour-vision deficiency (CVD), who is affected, and why do open sources quote different percentages?
Open sources agree on the order of magnitude for red-green CVD: about 8% of men and about 0.4-0.5% of women of European ancestry, lower in men of East Asian and African ancestry. They disagree on the non-European figures (Birch 4-6.5% for Chinese and Japanese men; Machado et al. Table 1 gives 4.2% Asian and 2.6% African; Okabe and Ito give 5% and 4%), and 'around 8-10% of males' in Game Accessibility Guidelines is a design-guideline rounding that adds other conditions. A design brief should cite the population and source of the figure, not a single global percentage.
- articleHow does Runway describe authoring a world for GWM Worlds 2 (WorldPrompt), and what does an agent have to supply?
Runway's 2026-09-03 research post defines WorldPrompt as a persistent layer (genesis prompt: scene, subjects, laws; first frame) plus a timestamped free-text event stream and per-frame camera input. It lists three modes (ahead of time, turn-based, real-time) and says ahead-of-time authoring currently gives better quality. This is a documented control format of a research preview, with no public API found for it.
- articleWhat does each named pattern cost in writing volume and state, what failure modes are attributed to it, and how could an agent recognise the pattern from a story graph?
Ashwell's post gives qualitative cost and failure statements only: a whole-body scan found no cost, hours or word-count terms and one digit string ('80 Days'). So 'writing volume' is stated only in relative terms (gauntlet easiest, sorting hat may mean several games, floating modules need a large amount of content). Graph recognition is not in the source; this candidate offers a labelled-synthesis rule table mapping each pattern's description to measurable graph and state properties, checked against one real graph (Cave of Time: near-tree, 0 cycles).
- articleHow do the primary papers measure whether several players views agree on one world, and can their reported numbers be compared across papers?
Six different consistency measures are in use: a VLM judge asked yes/no questions about screenshots (Solaris), depth reprojection error (ShareVerse, MultiWorld), DINOv2 feature similarity of corresponding regions (MetaWorld), a parser that compares objects across simultaneous views (MASS, zero by construction), detector-and-geometry agreement between clients (WorldCast), and attention maps shown as figures (MIRA). The numbers are not comparable: each paper retrains its own baselines, and the same Solaris model gets different FID values in Solaris and Gamma-World.
- articleHow does a franchise keeper run a continuity database (the Star Wars Holocron): what does it contain, who owns it, how is new material checked, and how are contradictions handled?
In a 2009 Game Informer Q&A, Leland Chee (Continuity Database Administrator, Lucas Licensing) describes the Holocron: an electronic encyclopedia of what has been developed and what is in development, created in 2000 to replace printed bibles, checked against by one licensing-side keeper, with producers given direct access. New material is entered as early as possible so creators can 'stake their claim'. Contradictions are handled by finding a plausible in-world explanation. Films and new TV series were not subject to prior approval by his group.
- articleWhat changes about a fictional world when it is a playable game or simulation (two-level action, point of action, subjunctive mode, chronicle vs story), and what consistency checks follow for game worlds?
Neitzel's LHN entry 'Narrativity of Computer Games' treats games as working in a subjunctive mode, like possible worlds: they pre-form possibilities that can happen in a digital environment. Player actions have a fictional level and a physical level at once, giving a point of action as well as a point of view, metalepsis as the normal case, and simultaneous narration. Simulation or world-building games are said to have the structure of a chronicle rather than a story in the strict sense, unless Ryan's broad definition is used. The entry stresses there is no consensus. For consistency, the sources support separating what the rules allow (possibility space) from what a given play-through realised (record).
- articleExactly how does Left 4 Dead's adaptive pacing work (intensity estimate, build-up/peak/relax states, timings, exemptions), and what did a second shipped game keep or change?
Valve's designer documents a small, reproducible controller: estimate each player's 'emotional intensity' from damage and nearby deaths, take the team maximum, and cycle four states (Build Up with full threats until intensity passes a peak threshold, Sustain Peak for 3-5 s, Peak Fade with minimal threats until a natural lull, Relax for 30-45 s or until the team has travelled far enough). Wanderers, mobs and special infected are throttled; bosses are exempt. Warframe's lead AI programmer independently reports a similar damage/kill-based intensity score and a peak-then-hold-off rule at intensity 100. Valve's slides call the estimator crude and give no decay constants or thresholds, and no ablation, so the pattern is documented but its tuning is not.
- articleHow do the open sources say player agency trades off against authored structure, and what concrete design diagnostics do they give?
Mateas (2000) argues story and agency are not intrinsically opposed: agency arises when material affordances (actions available) are balanced with formal affordances (what the plot makes worth doing); imbalance reduces agency, illustrated with puzzle adventures (too many material options), Quake (balanced) and a hypothetical plot-heavy shooter (plot without new actions). Roberts and Isbell state autonomy and authorial control are in tension; Riedl and Stern describe a coherence-versus-agency trade-off and a story director that adapts the plot; Facade's authors say they traded global for local agency under fixed authoring effort.
- articleWhich data structures do Facade's beat sequencer, the Automated Story Director's narrative-mediation tree and Versu's social practices use to hold story state and authorial control, and what did each paper evaluate?
Three primary papers give three reusable structures. Facade (Mateas and Stern 2005) annotates each beat with preconditions, weights, priorities and tension effects, and picks the unused eligible beat closest to an authored tension arc. The Automated Story Director (Riedl and Stern 2006) holds the story as a partially ordered plan with causal links, precomputes a tree of contingency plans for every player action that threatens a link, and reported a prototype whose tree had over 1000 paths. Versu (Evans and Short 2014) gives characters strong autonomy, uses social practices as suggestion-only joint plans, and makes each episode's story manager a reactive practice. None reports a controlled comparison against alternatives.
- articleWhich metadata fields does the SPN Software Metadata Recommended Format guide define for execution and emulation of software such as games?
The Software Preservation Network's metadata guide (v1.1.0, October 2023) groups its elements by use case. Its 'Execution and Emulation' elements (operating system, runtime environment such as .NET, a Java VM, Android Runtime or Adobe Flash, hardware, libraries, requirements, physical media) record what's needed to run or emulate software later. The guide's identifiers are inconsistent: the cover says 1.1.0 but the PDF metadata title says 1.0.0.
- articleWhat does the Library of Congress Recommended Formats Statement say should be deposited to preserve a software or video game?
For depositing games, the Library of Congress prefers, in order: uncompiled source, the gold master build, a distribution file (app package or executable), release media, then unpublished media. A video of the software in use can substitute when no preservation copy exists or online dependencies can't be recreated. Console or handheld titles need a virtual or emulated platform, deposited files must carry no DRM or encryption that controls access, and source deposits need rights clearance for the Library to run the compiler.
- articleWhat do OBS Studio's libobs docs say about output frame-rate representation and audio/video sync offset, and is a variable-frame-rate recording mode documented?
OBS (libobs at 32.2.2) runs its output at a fixed frame rate expressed as an integer fraction (fps_num/fps_den), set in the UI as a common, integer or fractional value. The UI strings searched have no variable-frame-rate option, so recordings come out constant frame rate, which is convenient for editing.