Evaluation and media provenance

Benchmarks, task datasets, attribution, metadata, interoperability and quality review

• articleIn Sargent's Ch. XXIII-XXIV (1916), which passages are his own claims, which are his worked examples, which are people he quotes, which are market assertions, and what OCR and provenance limits apply?
  1916 US practice, source-layer map. Layer A (Sargent's rules): the routes to an idea, the two methods, the diagnoses in the Ben and Mary revisions, the sub-plot rule, the advice to alternate and to let a story cool. Layer B (his worked examples): Dave Vroom (XXIII 6), the dynamo girl (9-10), the press-clip stories (17-18), the circus drummer (21-22), the wheat and castaway stories (26), the locomotive facts (27), the Edison harem story (34), and all of Ben and Mary (XXIV 15-24, 30-32). Layer C (people or sources he reports, unverified): Dana (XXIII 25), a Boston newspaper's Idea Machine (29), Phil Lang (XXIV 38), Geoffrey of Monmouth and Shakespeare (XXIV 3-4) and Bannister Merwin's quoted letter (XXIV 26), with which Sargent explicitly disagrees (27). Layer D (market assertions): editors take none of a flood of stories, no editor will bother with a one-hit author, the coal dealer analogy, prices 'a few dollars' against 'a hundred'. The OCR has about 6 'tl' for 'th' errors, 10 stray...
• articlePROPOSAL (1916 practice as the basis): what check could an agent run on a finished draft before submission using only what Sargent names in Ch. LXIII, and how could it be tested?
  PROPOSAL, not a source finding; 1916 practice supplies each step and the agent adaptation is mine. Six steps: (1) review in a separate context from the one that wrote the draft (the closest analogue of 'rest for weeks and regard it as another's work'; unvalidated); (2) run the ten questions with a reason from the draft for each, marking Q1 and Q3 as outside-text; (3) record verdicts as yes/mixed/no without a score; (4) run the scene pass for repeated devices, overlong scenes and doubled inserts; (5) repeat the pass later and after any external comments; (6) treat each external comment by comparing it with the draft. Nothing here has been run. Sargent gives no test that any step works and one worked case of the ten-question audit.
• articleHow does Sargent's 1916 pre-submission self-criticism compare, by candidate ID, with Aristotle's answers to objections (C3-B269-S82), Baker on testing against an audience (C3-B271-S83) and a rule audit of a specimen script (C3-B287-S95), and where does it differ?
  Comparison, synthesis throughout; every claim about the other batches rests on their candidate summaries and findings only (their evidence was not re-read). Against C3-B271-S83 (Baker 1919): both rely on a detached author review, but Sargent explicitly rejects audience proxies (theatre staff) and names no audience trial (0 hits for audience, public, aloud, reader in Ch. LXIII), while Baker's chapter adds rehearsal and performance reactions as signals and likewise no reading procedure (per C3-B271-S83-03). Against C3-B269-S82 (Aristotle XXV): Aristotle's text is a defence-side classification of objections; Sargent's only triage is to test each editorial comment against the script, but his Q10 passage weighs a broken convention against effect, loosely like S82's end-of-art test. Against C3-B287-S95: Sargent's ten-question audit is a human-run version of a fixed-rule audit, with reasons but no quoted script lines, and Q1-Q3 resemble the 'not testable from the text' class that C3-B287-S...
• articleHow does Sargent (1916, Ch. LXIII) apply his ten questions to one of his own scripts, what does he find wrong, and which of his statements are general claims rather than facts about that script?
  1916 practice (Ch. LXIII paras 12-23). Sargent runs the ten questions on Example A, an unproduced two-reel script printed in the Appendix (the Appendix describes its form as 'used by this writer', so it is probably his own script, an inference). Reading his answers, the ten questions give roughly four affirmatives (cost, motivation, settings, plot adherence), four mixed answers (newness, censors, suspense versus struggle, climax) and two negatives (struggle not enough; characters well drawn but not well chosen because the protagonist's aim is improper). He concludes the story would have little chance of sale and that good sense would urge withholding it; he says it was sent out once. General claims in the passage: suspense needs struggle in proportion; drawing is not choosing; technique cannot sell a story. The affirmative/mixed/negative tally is this worker's reading; Sargent does not tally.
• articleWhat are the ten test questions Sargent (1916, Ch. LXIII) gives for reviewing a finished script, what does each ask for, and what does passing or failing mean?
  1916 practice (Ch. LXIII para 11). Sargent reproduces a ten-question test, which he says was earlier offered in the Moving Picture World: new or novel; practicable; will it pass the censorships; struggle; suspense; motivation; characters well chosen; settings correct; plot adhered to; climax correct. Each is followed by a one- or two-sentence rule. His verdict rule: a story that passes 'should sell' (not always), one that does not stand the test has little chance. The ten questions mix market tests (cost, censorship, novelty), plot-craft tests (struggle, suspense, motive, adherence), and fit-and-values tests (characters, settings, righteous ending). Grouping and the 'needs outside knowledge' split are synthesis. Earlier Sargent candidates cover the craft behind Q2 and Q4-Q7 (see related).
• articleWho can I trust to criticise my finished script before I send it out, and whom should I not rely on, according to Sargent (1916, Ch. LXIII)?
  1916 practice (Sargent, third edition, Ch. LXIII paras 1-8, 27). Sargent says criticism is useful only from someone who both knows the subject and can state reasons, that such people are rare and busy, and that three groups a writer might consult are poor judges: the local theatre's staff (flattering a patron and judging finished films, not the visualizing of action), a high-school English professor (knows literature, not pictures) and actors (see only their own side). He names three sources: a knowledgeable helper, editorial comment (with checked rejection slips largely ignorable) and the writer's own critical study; exchanging scripts with a fellow author has a theft-suspicion cost. He ends by saying the writer is about the only person whose judgment can be relied on. These are his assertions from a trade manual, without data.
• articleWhat does George Pierce Baker (1919, Chapter I) mean by dramatic 'technique', and does he say it can be learned or taught?
  Baker defines technique as a dramatist's 'ways, methods, and devices for getting his desired ends'. He says nobody has it as a gift at birth or gets it merely by writing plays; the route he describes is reading and seeing many plays, imitating models at first, and (in a line from Pinero that he quotes) 'hard study and generally ... long practice'. Chapter I never uses 'teach', 'taught' or 'rule'; that technique can be studied rests on his statement that the universal part is what a beginner 'must study and may copy' and on the book's own purpose. This is 1919 stage practice, argued from Baker's survey of plays, with no test of whether study works.
• articlePROPOSAL: given the concordance, what decision table could an agent use to cite, check or apply classic craft rules on conflict, character change, length and outline order without overstating what the authors say?
  PROPOSAL (code_derived_inference; not an author's claim; untested). An eight-row decision table turns candidates 01-06 into agent instructions: pick the strength of the conflict claim (DT1), check four things on a character change (DT2), name the medium before using a length figure (DT3), outline first but revisable and define 'scenario' (DT4), look up the coverage cell before attributing a rule (DT5), quote within 25 words against pinned bytes (DT6), say 'not found' when no row answers (DT7), label any transfer to a new medium as an extension (DT8). Each row lists the evidence it rests on and a caution. No row has been run on a draft.
• articleWhich arithmetic and internal tensions are already flagged in the classic-text batches on conflict, character change, length and outline order, which new ones appear when authors are compared, and are any resolved?
  Register (code_derived_inference) of 19 tensions: 12 already flagged by the published batches and 7 new in this batch (T06, T07, T08, T09, T14, T16, T18). Kinds: arithmetic inside one author (Phillips x2, Baker), arithmetic across authors (Lutz vs Phillips reel time; Munsterberg vs Phillips scene rate and stage-to-screen compression), rules in tension inside or across authors (Archer Ch. XXII vs XIII/XIX; Phillips vs Archer on fixing the outcome; Aristotle XXVI vs VII; Baker vs Phillips on conflict), verdicts on the same play (Nora), a terminology collision ('scenario') and dating of the Esenwein and Leeds text. None is resolved by the sources; each row lists both sides with evidence ids and says who flagged it. Hand-checkable divisions are in work/recount_results.json.
• articleWhich of the eight classic craft authors can I cite for each of four basic questions (is conflict essential, how to prepare a character change, what length figures, outline first or write first), and where is there no record?
  Coverage matrix (code_derived_inference): 8 authors x 4 questions = 32 cells. 19 cells carry a recorded position or figure, 4 are scoped silences, 1 is relayed only (Archer on conflict) and 8 have no record. Baker and Phillips have a recorded position or figure on all four questions; Archer on three (one relayed); Forster only a scoped silence on one. A 'no_record' cell means none of the 27 assigned batches looked, not that the author is silent; a 'silent_scoped' cell names a line range and a word. Each cell with a position has an anchor evidence row copied from a published batch and re-checked against the pinned text.
• articleShould a story be outlined before it is written, or written first? What do Aristotle, Baker, Archer, Phillips, Esenwein and Leeds and Lutz each say (with evidence ids)?
  Concordance (code_derived_inference). Five of the eight authors record an outline-first order in some form: Aristotle (sketch the general outline, then names, then episodes), Baker (scenario before dialogue, kept flexible; some dramatists never outline), Phillips (choose the climax first, then cause and effect), Esenwein and Leeds (rough synopsis, then continuity, final synopsis last) and Lutz (a skeleton scenario first, details not fixed). Archer is the cautious voice: action, character and dialogue are interdependent, so a fixed scenario is an unfavourable sign and room should be left for unexpected developments; he reports playwrights who differ. No assigned source recommends writing without any plan. The word 'scenario' means a play outline in Baker but only the scene-by-scene continuity in Esenwein and Leeds. Munsterberg has no record; Forster is a scoped silence on the word outline.
• articleWhat length, act and scene-count figures do Baker, Esenwein and Leeds, Phillips, Lutz, Munsterberg and Aristotle give, in which medium, and do they agree?
  Concordance (code_derived_inference). Figures belong to different media and years and mostly cannot be pooled. Stage (Baker 1919): 3, 4 or 5 acts, a 2 to 2.75 hour evening, 5-15 minute intervals, no rule for act length. Silent film: Esenwein and Leeds (1919 ed.) and Phillips (1914) agree on a 1,000-foot reel with 25 to 50 scenes (Phillips elsewhere 25 to 60, and 75 to 200 scenes an hour, which his own 20-minute reel makes 75 to 150). Lutz (1920) gives 16 pictures per foot and one foot per second (a 16.7-minute reel) for animation timing and no film length in Ch. X. Munsterberg's 170 scene changes in ten minutes is 5 to 14 times Phillips's rate and unchecked. Aristotle gives no numeral for length (tragedies per poem, 'embraced by the memory'). Archer and Forster have no record. Checked arithmetic and internal tensions are listed.
• articleHow must a change in a character's will, heart or conduct be prepared, according to Baker, Archer and Aristotle, and who is silent or only indirect (with evidence ids)?
  Concordance (code_derived_inference). Baker (1919) and Archer (1912) agree that a change must be shown, not asserted, and both fault changes that fall between acts or off stage (both fault Pinero's Iris). They give different remedies: Baker plants earlier development, an emphasised motive and other characters' acceptance, and repeats illustrative action; Archer requires an on-stage 'psychological' scene, a new fact or motive for a change of resolve (not the clock) and, at a crucial point, external certification of a change of heart. Their verdicts on Nora differ (Baker: thoroughly motivated; Archer: compressed into a week). Aristotle XIII-XV has no arc rule (scoped negative) but demands consistency and 'necessary or probable' conduct. Phillips is indirect (motive and cause chain); Esenwein and Leeds, Munsterberg, Lutz and Forster have no record. No source gives a count of scenes or minutes of preparation.
• articleDo the classic craft texts agree that conflict is essential to drama, and which author says what (with candidate id, evidence id and pinned offset)?
  Concordance (code_derived_inference, not an author's claim) over eight texts. Phillips (1914) is the clearest yes: 'In all drama there is struggle with an obstacle.' Baker (1919) says by Brunetiere's definition conflict is central but 'covers a large part of drama but not all of it', naming an inert-figure exception that rests only on audience emotion. Munsterberg (1916, Ch. VIII) and Lutz (1920, cartoon plots) state it in passing or inside a narrower scope. Archer's view reaches this batch only second-hand (not indispensable, but 'no obstacle, no drama' as practice). Aristotle (Poetics XIII-XV) and Esenwein and Leeds (Ch. XIV) have scoped negatives; Forster has no record. The authors do not mean the same thing by 'conflict': an opponent (Baker's quoted list), an obstacle of any kind (Phillips) or a clash between two figures (Lutz).
• articleHow does Baker (1919, Chapter X) say a play is tested against an audience before and after production (readings, rehearsal, tryouts, revision after performance)?
  1919 stage practice. Chapter X contains no procedure for readings, tryouts or preview audiences (0 hits for 'tryout', 'try-out', 'preview', 'out of town' in the chapter; 0 for tryout terms in the whole text; 'readings' by others are not described). What it does give is a sequence and four signal sources. Sequence: a first draft for clearness; a re-writing for characterization and dialogue; a later re-writing shaped for the intended public; a detached consideration 'before it is tried on the stage'. Signal sources: rehearsal (Kismet: actor and author judged that sympathy for Hajji would be lost), performance reaction (Kismet torture scene too horrible 'when acted'; Dumas fils's first-night 'dissenting cries'; the German Belinda's silence), revivals of old successes (taste drift), and the record of trade predictions overturned. His warning: changes left to after production may come too late.
• articleHow should an agent review a draft against a list of objections and decide which are real faults, using the XXV method, with the classical theory (in translation) kept apart from my proposal?
  A ten-step review workflow assembled from XXV: tag each objection with the object imitated and the fault class, apply the type-specific answers the text gives (end of art, commonly said, dialectical test, necessity), run the context test, read apparent contradictions from the poet's own words and the critic's premise, and return a verdict per objection with quoted text. Each step is labelled classical theory (in translation) or proposal. The text gives no scores or thresholds; the pairings for "morally hurtful" and "contrary to artistic correctness" are inferred; the workflow is untested and an L10 experiment is sketched, not run.
• articleA reviewer says "that is not how things really are": which replies does Poetics XXV (Butcher) give a poet, and how should an agent decide which reply the work can actually claim?
  To "not true to fact", the text lists three replies the poet may make: the objects are as they ought to be (Sophocles said he drew men as they ought to be, Euripides as they are); if the representation is of neither kind, this is how men say the thing is (tales about the gods; the text allows they may be neither higher than fact nor true to fact); or the description is no better than the fact, but it was the fact (spears upright on their butt-ends, the custom then and among the Illyrians). These map onto the three objects named at the start. The replies are defences the poet may offer, not automatic wins; deciding whether the work really claims a given object is a separate check that the text does not spell out.
• articleHow does Poetics XXV (Butcher) decide whether a flagged flaw is a real fault, an excusable one or none: what is the end-of-art test, the essential/accidental ranking and the by-whom/to-whom/when/means/end test?
  For matters of the poet's own art, XXV allows an impossibility to be justified when the end of the art is thereby attained (the pursuit of Hector is the case in point), but not when the same end could have been reached as well or better without violating the rules of the art, since every kind of error should if possible be avoided. Faults are also ranked: ignorance that a hind has no horns is less serious than painting it inartistically. Separately, an act or saying is not to be judged in isolation as poetically good or bad: ask by whom, to whom, when, by what means and for what end, for instance whether it secures a greater good or averts a greater evil. The text offers no measure for "more striking" or "less serious".
• articleBefore I call something in a story a contradiction, what reading procedure does Poetics XXV (Butcher) prescribe, and what mistake in critics does it warn against?
  XXV says to treat sounding contradictions by the rules of dialectical refutation: ask whether the same thing is meant, in the same relation, and in the same sense, and settle it from what the poet says himself or from what an intelligent person tacitly assumes. For a word that seems inconsistent, count the senses it can bear in that passage. Language solutions listed with Greek examples include metaphor ("all" for "many"), accent, punctuation, ambiguity and usage. It warns, quoting Glaucon, against critics who adopt a groundless conclusion, reason from it and fault the poet for not matching their own assumption; the Icarius case is the worked example. Examples are tied to Greek verse; their use on modern scripts is a proposal.
• articleFor each of the five objection types in Poetics XXV (impossible, irrational, morally hurtful, contradictory, contrary to artistic correctness), which answer does Butcher's text give, and which pairings does the text make explicit versus which are my inference?
  XXV closes by listing five grounds of censure and saying the answers lie under "the twelve heads above mentioned", without numbering the heads. Reading the paragraphs, the text itself labels the answer for three types: impossible (justify by artistic requirements, the higher reality, or received opinion; or by attaining the end), irrational (appeal to what is commonly said; censure it when no inner necessity requires it) and contradictory (the dialectical-refutation test). Depravity of character shares the "no inner necessity" censure with the irrational. For "morally hurtful" and "contrary to artistic correctness" no paragraph is labelled; pairing the "by whom, to whom, when, by what means, for what end" test and the language-based solutions with them is an inference. The probable-impossibility and chance material is already in C3-B014-S4-02 and C3-B163-S34-03.
• articleWhen someone objects to a poem or story, what does Poetics XXV (Butcher) say the poet may have been imitating, and which kinds of fault count against the work itself?
  In Butcher's rendering, XXV opens with two preliminary distinctions that every later answer rests on. First, the poet, like a painter, imitates one of three objects: things as they were or are, as they are said or thought to be, or as they ought to be. Second, faults inside the art are of two kinds: those that touch its essence (imitating incorrectly through want of capacity) and accidental ones (a wrong choice, such as a horse's legs or medical details from another art). The text calls these the points of view for answering critics; it states no thresholds, and the passage reads as a first triage before the specific objection types.
• articleHow can an agent locate the turning point between complication and unravelling in an outline, using Poetics XVIII (Butcher) as classical theory in translation, and what check follows?
  Classical theory (in Butcher's translation): the Complication runs to the part that marks the turning-point to good or bad fortune, the Unravelling from the beginning of the change to the end, and presupposed incidents can belong to the Complication. Proposal (mine, untested): an agent lists beats with fortune direction, tags presupposed incidents, finds the first beat where fortune turns, records the boundary and the convention for the turn beat, and flags multiple candidate turns. The text gives no threshold or test, so every step below the definitions is a proposal.
• articleIs this story outline a complete whole of suitable magnitude: what agent check follows from Poetics VII, with the classical theory (in translation) kept apart from my proposal?
  Combining VII's definitions of whole, beginning and end, its rule against haphazard edges, its limits on length (embraced by memory, perspicuous) and its rough change-of-fortune bound gives a nine-step outline check. Each step is labelled as classical theory in Butcher's translation or as my proposal. The text gives no thresholds, and the check is untested; an L10 experiment with seeded defects is sketched, not run.
• articlePROPOSAL: when writing for young audiences, how can an agent separate claimed audience effects from evidence, using Münsterberg's Chapter XI as a worked case?
  PROPOSAL (this worker's, not a source claim; untested). A six-step claim-grading check: extract effect claims, class them, record the support attached, refuse to convert hedges into facts, keep rules and editorial choices in separate lists, and label the result. Piloted by the worker on Chapter XI as a single grader: of 30 claims, 25 have no support attached, 3 rest on unnamed attribution, 1 is an unnamed quoted figure the author calls exaggerated, 1 is an explicit admission that the effect is unmeasured, and 0 name a source. Of the 7 claims about the young, 5 have none. The check gives no legal or safety conclusions; where a venue's written rules are supplied they go in a separate list.
• articleCompared by candidate ID only, how does Münsterberg's 1916 Chapter XI argument differ from the 1914 and 1919 trade accounts of content limits (C3-B256-S71, C3-B242-S65), and what does it not repeat from C3-B187-S41?
  Comparison by candidate ID only, built from the candidate summaries read on 2026-10-22 (the Esenwein and Phillips texts were not re-read here; C3-B256-S71 was an unpublished draft when read, so its IDs are provisional). 1916 argument, unmeasured. Münsterberg offers a theory-level treatment test and reports industry attitudes to censorship, where the 1919 trade advice (C3-B242-S65-03, -04, -06) records a second-hand board list and conditional rules, and the 1914 advice (C3-B256-S71-01 to -04, draft) states author opinions and one reported board effect. All three agree in direction that crime must not be offered as spectacle or glamour; they differ in test (atmosphere of the whole play versus scene-level conditions), in whether endings matter, and in whether any rule source is named. The 1914-1916-1919 order is not read as a change over time.
• articleWhich risks does Münsterberg name in Chapter XI, what mechanism does he give for each, and how certain is his wording?
  1916 argument, unmeasured. The chapter names thirteen risks, which this worker groups in four: perceptual (hallucinations), moral (imitation, loss of resistance and moral balance, harm to the young mind, 'psychical infection', unmeasurable weakening of general righteousness, glamorised vice with an appended punishment, suppression of contrary ideas), intellectual (trivialising, simplification and stereotyping caused by wordlessness, wasted lessons) and aesthetic (the commonplace triumphing, derision of the medium as an aesthetic vehicle). Wording ranges from 'may' (19 uses in lines 3746-4071) and 'cannot be overlooked' to 'must be fraught with dangers'; the only statement that a risk cannot be measured is the 'no psychologist can determine' sentence. No risk is tied to a measured case.
• articleWhat aesthetic function does Münsterberg assign to the photoplay in Chapter XI, and what does he say it takes to achieve it?
  1916 argument, unmeasured. He calls aesthetic cultivation the photoplay's 'greatest mission' because no art reaches a larger daily audience in a more receptive frame of mind. He argues aesthetic training cannot follow the audience's current liking: the commonplace triumphs over the significant unless systematic effort is made, as with symphony and opera, and a public that first prefers Sousa to Beethoven has to be led by slow steps. He defends the medium against the 'canned drama' objection by analogy with printed poetry and sculpture, and declines to predict whether a great art will emerge. The assertions about taste are unsourced.
• articleWhat educational function does Münsterberg give the photoplay in Chapter XI, and how does he think teaching should reach audiences?
  1916 argument, unmeasured. He reports a plan for a 'Universal Culture Lyceum' to circulate educational pictures to schools, churches and colleges, but says the far larger educational contribution must come from ordinary picture houses, where teaching must be absorbed by people seeking entertainment, not forced on them. Intellectual uplift is 'the easiest': photoplays can lead young and old to new knowledge; the danger he names is trivialising, not dangerous knowledge. He doubts strictly educational films teach as much as producers hope, for untrained (and young) viewers, unless human interest connects the events. None of this is tested in the chapter.
• articleDoes Münsterberg's Chapter XI take a position on censorship, and what does it say about how crime and vice may be shown?
  1916 argument, unmeasured. The word 'censor' occurs twice in the whole pinned book, both in one Chapter XI paragraph, and both report what producers think: some companies 'welcome censorship' if it is high-minded and liberal, and most seem doubtful about Federal censorship. Münsterberg states no vote of his own in those sentences. His own stated position is about treatment, not a rule: a villain punished in the last reel does not undo glamorous vice, a play should not be a dramatized sermon, and if a 'moral atmosphere' pervades the whole screen world then no crime is barred from frank telling. No board, statute, list or age rule is named.
• articleWhat does Münsterberg's Chapter XI (1916) say about children and young people as viewers, and what support does he give for it?
  1916 argument, unmeasured. He says the 'young mind' may suffer from rapid contrasts between farce and passionate romance, that boys and girls shown glamorous crime feel sure they would escape, that youth (and the untrained) follow instructive films poorly (a child 'grasps very little' of a factory), and that French comedies once 'poisoned youth'. The only number about the young is a prominent criminologist's claim that 85 per cent of investigated juvenile crime was traced to films, which he calls 'probably a strong exaggeration' without giving another figure or a source. {nyoung} of the 30 graded claims concern the young; none names a source.
• articleIn Chapter XI of The Photoplay (1916), what influence on audiences does Münsterberg claim for the movies, and does he cite any evidence for it?
  1916 argument, unmeasured. Münsterberg opens with two attendance figures (ten million a day from 'enthusiasts', two or three million from 'sceptics'), picks neither, and calls the movies one of the strongest social energies of the time. The effects he asserts on audiences are pleasure and heightened vitality (from the fast rhythm of wordless action), reported sensory hallucinations in neurasthenic persons, rural applause as a 'symptom', and high suggestibility, and he argues the intensity 'cannot remain without strong social effects'. In lines 3746-4159 no study, statistic, experiment or named author is cited for any of these; the sourcing wording is 'reported', 'enthusiasts', 'sceptics' and an unnamed 'sociological observer'.
• articleHas Adobe changed or clarified the Illustrator FAQ sentence about not adding Content Credentials, and do Illustrator release notes explain the 29.4 behaviour?
  The sentence is byte-for-byte the same in three captures: a Wayback capture dated 2025-10-06 (page date line 'Sep 11, 2025'), the cycle-2 fetch of 2026-10-21 and the live page of 2026-10-22 (date line 'Jun 1, 2026'). Between the 2025 and 2026 captures only the surrounding labels and small phrasing changed ('With version 29.4:' became 'With Illustrator versions 29.4 and later:'). The Illustrator desktop release-notes page, covering 29.8 (Aug 2025) through 30.8 (Aug 2026), contains no Content Credentials wording; 29.4 is outside its range. Two captures of that sentence do not show what happened between them or before September 2025.
• articleIs there a primary source (a named previs supervisor, studio or published study) giving a proportion of previs that is used versus discarded?
  No proportion was found in the sources actually read. The one pinned primary statement is a qualitative one from a named Third Floor previs editor (Ian Differ, ProVideo Coalition interview, 11 August 2018): sometimes previs work ends up in the film and sometimes it does not. A '99 percent' figure in the same interview is about where handheld camera shake comes from, not retention, and should not be cited as a discard rate. The negative is scoped to the pages read plus web-search leads that were not fetched.
• articlePROPOSAL: before planning a story for a particular platform, what check can an agent run to keep that platform's written hard content rules separate from genre, market or cost preference, using the 1919 manual's layers as a pattern only?
  PROPOSAL (this worker's, not a source claim): a six-step gate-classification check. Name the venue and its current governing text; classify every risky beat as HARD (a written rule quoted with locator and date), CONDITIONAL (rule with a testable condition), OTHER-VENUE (a rule that applies elsewhere), PREFERENCE (genre, market, taste, cost) or UNKNOWN; record each rule's source grade; handle HARD beats by rewrite and PREFERENCE beats by story need; and re-run when the venue changes. The pattern comes from five lessons in the 1919 trade text; the text supplies no current rule and the check needs the venue's own policy as input. No test of the check has been run.
• articleWhat do open-access empirical studies report about eyeline matches and editing-linked gaze that bear on shot/reverse-shot dialogue, beyond the 180-degree study already in cycle 2?
  Thin result. The one on-topic open item found is a conference abstract (Ildirar Kirbas and Smith, ICIS 2018) on whether 12-month-olds and adults follow a head turn across an eyeline-match cut: adults followed gaze in edited and unedited versions; infant data were preliminary and suggested less success across edits. Bruckert et al. (2021) report editing/scanpath links in movie clips (abstract only). Neither tests two-person dialogue coverage. The 180-degree experiments are cycle-2 C2-B01-A-06 and are not repeated.
• 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.
• articleWhat does Spotify require, accept or detect about AI music: AI credits, AI Persona badges, impersonation and spam rules, and what is announced but not yet observed?
  Spotify's pages at 2026-10-22 describe four separate mechanisms. (1) AI credits: optional, supplied through a distributor, at contributor-plus-role level, shown only on mobile Song Credits and Now Playing during a beta, audio only. (2) AI Persona badge: about the artist's public identity, not how music was made; artists may self-disclose in Spotify for Artists and Spotify also reviews profiles and may apply a "Likely AI Persona" badge, which the artist can confirm or appeal; badged music is excluded from editorial and personalised recommendations by default. (3) Impersonation: Spotify removes unauthorised voice replicas, but only when a claim is received from the artist or their representative. (4) Spam/slop filtering using many signals including third-party AI detection tools, which Spotify says has limits. A listener reporting tool for AI Personas was announced in August 2026 for "the coming months" and was not observed. Spotify's figures (for example tracks 10 times more likely t...
• articleHow far can ToMi-style story scores be trusted as evidence that a model tracks who knows what, given what the benchmarks' critics found?
  The ToMi authors themselves say earlier benchmarks could be solved without theory of mind because of dataset biases. SymbolicToM's authors report ambiguity (in a 20-worker MTurk check, human accuracy on one template fell from 80 to 20 percent when the nouns changed), mislabelled second-order items, and large drops for supervised models on altered structure (TTT -54 points on paraphrase). SimToM's authors note the original ToMi lacks answer choices, which lowers baselines. Ullman and Shapira et al. (abstracts only) report that small alterations or adversarial items reverse apparent successes; FANToM and OpenToM add multi-format consistency and shortcut control. Consequence: scores are comparable only within the same corrected version and answer format.
• articleDoes tracking where a character thinks an object is carry over to what the character feels about an event they partly saw, and does a perspective-filter prompt help?
  OpenToM (arXiv 2402.06044v3) has 696 narratives (596 GPT-3.5-Turbo, 100 longer GPT-4-Turbo) with characters given personality traits and intention-driven actions, and 23 first- and second-order questions each: location (coarse and fine), multi-hop (container fullness, accessibility) and attitude of an observer to the mover's action. GPT-4-Turbo led most columns but scored 0.544 macro-F1 on attitude against 0.862 for humans, and chain-of-thought barely helped attitude. SimToM (perspective filter first, then answer) was mixed on OpenToM: gains on coarse location for the GPT models, mixed elsewhere. Models of 2023-2024; zero-shot only.
• articleHow does the needle-mirror describe how it is produced, and which parts of a mirror-based claim can be checked without downloading the package archive?
  The mirror organisation says it holds "versions of packages available in UPM, without any changes", is "Not affiliated with Unity Technologies", and "updates automatically"; the mechanism is not described in what I read. The pinned tag and commit name "Unity Technologies" as author/tagger, which is a label set in the mirror, not evidence of Unity authorship. The tree also holds a .signature (JSON, timestamp 2026-07-29T17:56:58Z) and a 10,920-byte .attestation.p7m, signing artifacts that could support archive-level verification, but I did not verify them and no inspected source says what they cover. Checkable without the archive: registry metadata (candidate -01) and docs text (candidates -02 to -04). Not checkable: the .cs files.
• articleItem by item, which of Baker's 1919 scenario requirements and failure modes carry over to current Screen Australia guidance, which do not, and what checklist can an agent use to produce and self-check a treatment, outline or series pitch document?
  This is a synthesis across Baker 1919 (cited only through C3-B194-S47-01..-03, not re-read) and Screen Australia's guide and FAQ. Shared ground: the short document is a problem-finding tool before drafting; events and actions rather than named effects; selection over completeness. Differences: Baker prescribes a persons-list-then-settings header and sets clearness before brevity for stage plays; the agency prescribes page ranges by document type, a series engine, and a funder's self-analysis of weaknesses. Not found in the agency texts read: Baker's withheld-key-fact failure by name, any scored test, any rule for referencing other works (the bible instead asks for antecedents or references). The checklist below is a PROPOSAL, untested.
• 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.
• articleHow should these Netflix partner help-centre pages be cited and dated, given that the HTML carries no date and the Zendesk API gives three timestamps?
  None of the four assigned pages prints a revision date. The Zendesk article API gives created_at, updated_at and edited_at: Live Action updated 2023-12-31T18:24:45Z (edited 2020-12-07T06:28:31Z), Animation updated 2023-11-30T20:02:40Z (edited 2020-06-25T00:10:03Z), 5.1 Mix Room updated 2024-04-06T11:27:25Z (edited 2020-12-07T06:30:17Z), Secondary Audio updated 2024-03-18T14:25:34Z (edited 2022-10-03T14:11:23Z). In every case updated_at is later than edited_at, and the API does not say which tracks a body change, so these are metadata, not proven revision dates. The practical citation is the HTML sha256 with the fetch time, the API values labelled as metadata, and, for a page that moved, an archive memento with its timestamp.
• articleI searched for the Netflix "Secondary Audio Delivery Specification": does that page contain the requirements, where do the file format, channel order, loudness and peak numbers actually live, and can I still read them?
  The Secondary Audio Delivery Specification page (API updated_at 2024-03-18) is a one-sentence pointer to "Localization, Accessibility, and Dubbing Branded Delivery Specifications"; readers voted it unhelpful (2 of 12 on the page). The live target now redirects (301) to knowledge.netflix.net, which redirects to a login, and the help-centre API returns 404 for it. An Internet Archive memento of the target (page version 5.1, captured 2024-08-10) is readable and holds the numbers: 48 kHz, 24 or 16 bit WAVE Extensible, 5.1 required and 2.0 optional, listed 8-channel orders, mono duplicated on channels 1 and 2, -27 LKFS +/- 2 LU dialogue-gated, -2 dBFS true peak, downmix trims. It is a past version of a spec whose current text was not readable.
• articleWhat public head-movement and saliency datasets exist for 360-degree content, and what do they show about where viewers look (and what they cannot show about story)?
  Two open-access datasets papers were read: Sitzmann et al. (arXiv v2; 22 computer-generated static stereo panoramas, 169 users, 1980 gaze-and-head trajectories, three viewing conditions) report an equator bias and strong between-user agreement; Nasrabadi et al. (MMSys 2019; 28 videos built on a moving-object by camera-motion taxonomy, 60 participants, no audio; MIT-licensed GitHub repo with traces, questionnaires and scripts) report preliminary signs that viewers tend to look at moving targets, with exceptions, and more dispersed viewing in higher-altitude shots. Nasrabadi et al. also list other datasets with participant counts (e.g. 59 participants/5 videos; 48 subjects/20 videos; 57 subjects with eye data) and note viewing is more scattered under free viewing than when questioned afterward. These datasets describe viewing of short, mostly narrative-free content and do not test story comprehension or missed events.
• articleIs there a consistent, quantified effect of guidance cues on viewer experience in cinematic VR across studies, or do the results conflict?
  A 2024 arXiv preprint (Zhang et al.; ACM template with placeholders, peer-review status not shown on the page) screened 3444 papers, reviewed 45 and meta-analysed 13 on CVR viewing modalities versus a 'swivel-chair' control, and reports inconsistent effect sizes even within the same modality, small samples, interchangeable use of presence/immersion/engagement, and mostly negative results for limited or forced rotation. It supports the batch's central caution: there is no universal rule, and individual cue studies (candidates 03 to 05) are not additive. The outcomes are questionnaire experience measures, not missed-event or gaze-target accuracy.
• articleWhat should a design/text handoff record so a different renderer can reproduce the text correctly or at least detect that it cannot?
  SYNTHESIS (a proposal, not a source finding). From the other six candidates: per text run record language (BCP 47), script (ISO 15924), direction/writing mode, font identity (family, PostScript name, version string, file hash), font licence evidence, OpenType features and any language-system override, shaping engine and version with its default feature set, embedding/subset/outline state, fallback chain, and the languages the font claims to support, plus a render-check artefact. Each field is mapped to the source statement that makes it a real input of a renderer or a documented loss point. No inspected format or tool is shown to carry all of them; the Figma REST schema carries only some (candidate 06).

More