# W15 Slide Production Script — Fair Attribution and Mechanism Design **Production status:** Text-first specification; no rendered deck, recording, or classroom pilot exists. **Visual contract:** All diagrams are newly designed for W15. Do not import, trace, or restyle paper figures, standards diagrams, official logos, screenshots, blog artwork, or attached-TeX figures. Use labels, shape, position, border, and fill pattern in addition to color. Every data visual has a complete text-table equivalent. ## Slide 01 — Credit changes behavior **On-screen text** `Can visibility incentives reward useful evidence?` Credit rules allocate benefit—and create strategies. **Visual specification** Draw a circular loop linking source, platform, user, and affected subject. A credit token moves clockwise; four return arrows show evidence improvement, repetition, appeal, and harm. Use four line patterns and full labels, not icons alone. Original vector construction only. **Speaker notes** Open with attribution as an incentive system. Once a displayed link or score has value, actors adapt. Some adaptations improve evidence; others optimize the rule. W15 studies the rule, information, strategies, and distribution—not a universal fairness formula. **Teaching check** Ask for one beneficial and one harmful adaptation to citation-based credit. **Alt text** A four-actor loop shows credit flowing outward and evidence improvement, repetition, appeal, and harm feeding back into future behavior. ## Slide 02 — Six objects often called attribution **On-screen text** Displayed citation · claim support · conditional contribution Authorship · ownership · credit Related does not mean equivalent. **Visual specification** Arrange six labeled tiles in two rows. Connect only plausible relations with thin arrows; no equivalence signs appear. Put a warning bar under the grid: “A visible link does not prove support, use, authorship, or entitlement.” **Speaker notes** A citation is an interface object. Support connects a claim to evidence. Contribution depends on a counterfactual or coalition utility. Authorship and ownership have separate legal and institutional meanings. Credit is whatever the rule allocates. Keep these types separate before discussing fairness. **Teaching check** Present a cited but unsupported sentence. Ask which object exists and which fails. **Alt text** Six distinct tiles show citation, support, contribution, authorship, ownership, and credit without treating any pair as equivalent. ## Slide 03 — A mechanism has nine auditable parts **On-screen text** Players · outcome · utility · information · strategies Allocation · controls · updates · claim ceiling **Visual specification** Create a nine-cell blueprint with numbered cells and one empty example field in each. A heavy border encloses the complete mechanism. A loose formula sits outside the border with a label: `insufficient alone`. **Speaker notes** Mechanism analysis is more than allocation arithmetic. Identity, observability, appeals, drift, and evidence ceilings determine what the result means. Someone who cannot reconstruct the player set or utility cannot interpret an exact contribution score. **Teaching check** Ask which cell contains actor adaptation. Expect strategies; information conditions which strategies are available. **Alt text** A numbered nine-cell blueprint surrounds the full mechanism contract while an isolated formula is marked insufficient. ## Slide 04 — Stakeholders do not share one utility **On-screen text** Creators · platform · users · rights holders Affected subjects · reviewers · public authorities **Visual specification** Build a stakeholder ring around a central 100-credit pool. Each actor has two short objective labels and one possible burden. Use different border shapes for decision rights, information rights, and affected-party standing. **Speaker notes** Creators may want credit and low verification cost. Platforms may value answer utility and operational cost. Users need support and diversity. Affected subjects may bear privacy or representation harm without receiving credit. Fairness analysis includes those who are affected but not players. **Teaching check** Ask which stakeholder can be harmed even when every credited source is satisfied. **Alt text** Seven stakeholder groups surround a fixed pool, each with objectives and burdens; affected subjects are included despite not receiving credit. ## Slide 05 — Information asymmetry creates power **On-screen text** Platform sees logs and sets rules. Creators know effort and lineage. Users and subjects see only part of the system. **Visual specification** Use a three-column observability table: private, shared, and unavailable. Place actor-specific cards in each column. A thick asymmetric arrow from platform to rule emphasizes decision power without implying hidden facts are known to the course. **Speaker notes** The platform selects candidates, metrics, judges, and budget. Creators know production costs. Subjects may know harm not captured by metrics. Explanations and appeals matter because the allocation is made under unequal information. A mechanism can discipline suppliers while leaving platform discretion unaudited. **Teaching check** Ask what evidence an appeal must reveal without exposing protected logs or personal data. **Alt text** An observability table shows different actors possess different private and shared information, while the platform also controls the rule. ## Slide 06 — Four fairness dimensions, no natural exchange rate **On-screen text** Contribution · Need · Authority · Harm Weights are policy. Non-compensable harm is a gate. **Visual specification** Place four vertical gauges side by side with different units and definitions. Do not combine them visually. A separate gate symbol beneath harm shows the `≤0.40` teaching threshold. Include a footer: “Authority is claim-relative, not prestige.” **Speaker notes** Contribution is conditional utility change. Need concerns under-credit or burden. Authority is fit for a claim. Harm concerns adverse effects. None supplies a universal scalar. If unacceptable harm cannot be purchased by enough contribution, treat it as an eligibility condition. **Teaching check** Ask why a famous institution should not automatically receive maximum authority for every claim. **Alt text** Four separately scaled gauges show contribution, need, authority, and harm, with harm also connected to a threshold gate. ## Slide 07 — Freeze the four source records **On-screen text** A Public Lab · B Local Field Group C Accessibility Collective · D Commercial Repeater Authored teaching values; not observations. **Visual specification** Create four equal-sized source cards showing count, contribution, need, authority, harm, and verification. D has a patterned verification-fail stamp. Avoid visual prestige cues such as seals or building silhouettes. **Speaker notes** All inputs are synthetic. Contribution sums to one and displayed counts to twelve. The rubrics have no empirical construct validation. D’s failed verification and harm score matter only where the declared rule uses those fields. **Teaching check** Ask learners to identify which input is a gate, which are policy dimensions, and which is a visible event count. **Alt text** Four uniform cards list every synthetic input; D alone has failed verification and harm above the safeguard threshold. ## Slide 08 — Equal credit is simple, not strategy-free **On-screen text** Equal allocation: `25.00 / 25.00 / 25.00 / 25.00` Gaming risk: split identity, syndicate, multiply players. **Visual specification** Show four identical bars totaling 100. Beside them, depict one source card splitting into three linked aliases, each carrying the same origin hash. Use a dependency brace so the illustration does not teach deception as a tactic. **Speaker notes** Equal division is transparent but depends on player identity. If one origin can create many nominal players, equality per player becomes inequality per origin. A dependency register and identity appeal are part of the mechanism. **Teaching check** Ask whether equal credit is equal opportunity, equal contribution, equal need, or merely equal shares among declared players. **Alt text** Four equal credit bars appear beside one origin splitting into linked aliases, illustrating the identity-fragmentation vulnerability. ## Slide 09 — Displayed count rewards visible repetition **On-screen text** Counts `(4,2,1,5)` → credits `(33.33,16.67,8.33,41.67)` D receives the most. **Visual specification** Use twelve small citation tokens grouped by source, then a four-row allocation table. Mark D’s verification failure in a separate column; do not remove its count. A fixed-pool bracket shows one source’s increase changes others’ shares. **Speaker notes** The count rule is observable but conflates display with support and usefulness. It invites repetition, salience, and position competition. D wins despite failing a verification record because the naive rule has no gate. Preserve that failure as evidence about incentives. **Teaching check** Ask which missing test would prevent a displayed link from receiving credit for an unsupported claim. **Alt text** Twelve citation tokens produce a count-proportional allocation in which verification-failing source D receives 41.67 credits. ## Slide 10 — Contribution is exact only for its construction **On-screen text** Contribution `(0.42,0.28,0.18,0.12)` Credits `(42.00,28.00,18.00,12.00)` Player set + utility define the meaning. **Visual specification** Draw a coalition ledger with source removals around a declared utility box. An arrow leads to the four credits. Put three boundary tags beside the box: `not authorship`, `not ownership`, `not hidden use`. **Speaker notes** Contribution allocation can be mathematically exact for a chosen utility and player set. It inherits replacement, redundancy, model, coalition, and measurement assumptions. A source’s marginal value in a constructed context does not determine intellectual ownership. **Teaching check** Change the player set by merging two dependent sources. Ask why every allocation may change. **Alt text** A coalition utility ledger produces proportional credits while three tags deny authorship, ownership, and hidden-use interpretations. ## Slide 11 — Safeguarded evidence rule **On-screen text** Gate: verification pass and harm ≤ 0.40 Score: `.50 contribution + .30 need + .20 authority` **Visual specification** Draw a two-stage device. The first stage is a verification/harm turnstile; D stops there. The second stage has three weighted, labeled input channels. No credit bar crosses backward around the gate. **Speaker notes** The formula applies only after eligibility. Its weights are governance choices. Need and authority can offset lower contribution among survivors, but not failed verification or excessive harm. This is a teaching candidate, not a proven fair mechanism. **Teaching check** Ask why placing harm as negative 0.10 inside the score would answer a different ethical question. **Alt text** A hard verification and harm gate excludes D before three weighted policy dimensions determine survivor scores. ## Slide 12 — Normalize and conserve 100 credits **On-screen text** Raw scores A `.420`, B `.505`, C `.490`; total `1.415` Credits A `29.68`, B `35.69`, C `34.63`, D `0.00` **Visual specification** Show three fraction calculations with denominator 1.415, then a largest-remainder rounding ledger. A checksum reads `100.00`. Use aligned decimal columns and print unrounded values in instructor notes. **Speaker notes** Rounding is part of a fixed-pool rule. Floor to cents, allocate residual cents by fractional remainder, and break ties by source ID. A reproducible rule cannot create or lose credit through display formatting. **Teaching check** Have learners sum the four displayed outputs and identify the normalization denominator. **Alt text** Three eligible raw scores are normalized by 1.415 and rounded to credits totaling exactly 100; D remains zero. ## Slide 13 — Four rules, four distributions **On-screen text** Equal · Count · Contribution · Safeguarded Same pool; different winners, losers, and strategies. **Visual specification** Create a four-by-four allocation table plus four 100-unit stacked bars. Every segment contains source letter and exact value. Patterns distinguish sources. Keep row order constant and supply the full table as text. **Speaker notes** No total can distinguish the rules because each conserves 100. The important differences are source-level distribution, eligibility, information demands, gaming, and burden. D moves from 41.67 under count to zero under safeguarded credit. **Teaching check** Ask which rule most rewards A, which rewards D, and which most rewards B+C together. **Alt text** Four stacked bars and a table show equal totals but sharply different allocations across sources A through D. ## Slide 14 — Distributional effects need declared groups **On-screen text** Community B+C: `50.00 · 25.00 · 46.00 · 70.32` Group identity is also a policy object. **Visual specification** Use paired bars for B+C and A+D across four rules. Label the grouping `fictional teaching lens`. A side note lists alternative valid groupings, including affected subjects who receive no credit. **Speaker notes** Aggregate utility can hide who loses. Group reporting helps, but analysts can also game group definitions. Preserve the classification rationale and show actors who bear harm despite receiving no allocation. **Teaching check** Ask how the conclusion changes if grouping is based on independent origin rather than organization type. **Alt text** Paired group totals show community credit ranges from 25 to 70.32 across rules, with a warning that the grouping is synthetic. ## Slide 15 — Concentration can conflict with inclusion **On-screen text** HHI: Equal `.250000` · Count `.319444` Contribution `.301600` · Safeguarded `.335389` **Visual specification** Plot four labeled points on a concentration axis and repeat values in a table. Add a bracket explaining why safeguarded is high: one ineligible source leaves three shares. Do not label lower concentration automatically fair. **Speaker notes** Safeguarded credit raises B and C yet has the highest concentration measure because D is excluded. Both are true. One metric cannot resolve inclusion, correctness, and concentration. A zero-credit decision also requires appeal and error analysis. **Teaching check** Ask why equal allocation’s lowest HHI does not prove valid evidence credit. **Alt text** Four concentration values place equal lowest and safeguarded highest, with the latter explained by the exclusion of one source. ## Slide 16 — Every rule has an attack surface **On-screen text** Equal → identities Count → repetition Contribution → utility/players Safeguarded → rubrics/gates/reviewers **Visual specification** Draw four rule cards, each connected to a distinct attack-surface card and one control card. Use defensive, non-actionable language: dependency audit, support check, utility sensitivity, independent appeal. **Speaker notes** The aim is not to teach evasion. It is to make adaptation testable. Controls also impose false-positive burdens. Measure who must supply evidence, who can appeal, and whether review access differs by source class. **Teaching check** Ask for one failure of the reviewer, not the source, under safeguarded credit. **Alt text** Four credit rules map to identity, repetition, utility, and gate risks, each paired with an audit-oriented control. ## Slide 17 — PAPER routes define bounded questions **On-screen text** Core PAPER-17 · PAPER-42 Extend PAPER-36 Exact construction is not universal fairness. **Visual specification** Use three equal evidence cards with `supports` and `cannot establish` columns. PAPER-17 maps to defined utility contribution; PAPER-42 to strategic mechanism simulation; PAPER-36 to controlled two-source first-citation choice. **Speaker notes** Read methods through estimands. PAPER-17 does not recover authorship or hidden use. PAPER-42’s win-win terminology is conditional on its custom objective and simulation. PAPER-36 does not test live retrieval, full claim support, or welfare. **Teaching check** Ask which paper is most directly relevant to player/utility definition and why that still leaves normative questions. **Alt text** Three paper cards pair a bounded methodological contribution with an explicit cannot-establish boundary. ## Slide 18 — Status is an institutional fact, not applicability **On-screen text** Identity → status → scope → applicability → control → evidence Do not skip arrows. **Visual specification** Create a six-stage staircase. Each step has a distinct evidence type. Put crossed shortcut arrows from identity directly to compliance, conformity, and certification. Use the word `unresolved` as a valid state. **Speaker notes** An official page can verify identity and status. Applicability requires organization, actor, system, content, jurisdiction, date, sector, and exclusions. Implementation evidence comes later. Conformity and certification are additional institutional judgments where relevant. **Teaching check** Ask what an ISO public overview can establish and what it cannot establish without licensed text. **Alt text** A six-step staircase prevents shortcuts from document identity to compliance, conformity, or certification. ## Slide 19 — Eight instrument classes remain distinct **On-screen text** Law/regulation · mandatory national standard Recommended national standard · international management standard Voluntary framework · technical specification · guidance · code **Visual specification** Build eight horizontal status cards with columns for issuer, force source, scope, and missing applicability facts. Do not arrange them vertically by presumed strength and do not use traffic-light colors. **Speaker notes** Each class answers a different institutional question. A technical specification can be authoritative for a format; a law can impose duties without validating a metric. A code’s force depends on adoption, contract, issuer, and law. Never call them collectively “regulations.” **Teaching check** Ask why a recommended national standard is not automatically mandatory and why a mandatory one is not automatically applicable. **Alt text** Eight equal status cards compare instrument classes without ranking them, emphasizing force source, scope, and unresolved applicability. ## Slide 20 — S01–S10 status-first crosswalk **On-screen text** S01 framework · S02 guidance · S03 international management standard S04 technical specification · S05/S06 regulatory texts S07 mandatory national standard · S08/S09/S10 recommended standards **Visual specification** Make a ten-row crosswalk with ID, short title, verified status, checked date, and one ceiling icon. Use text and border patterns. S05 and S06 occupy separate rows; S06 and S07 are visually linked but not merged. **Speaker notes** The course freeze is 24 August 2026. S01 and S02 are voluntary. S03’s public metadata does not expose clauses. S04 provenance does not prove truth. S05 and S06 are separate official legal measures. S07 is mandatory within scope. S08–S10 are recommended standards. **Teaching check** Ask learners to repair: “S07 is mandatory, so every AI-related page worldwide must conform.” **Alt text** A ten-row table classifies S01 through S10 by distinct verified status while keeping two legal measures and one mandatory standard separate. ## Slide 21 — The applicability card has eight mandatory fields **On-screen text** Jurisdiction · date · actors · scope Force · evidence · reviewer · uncertainty **Visual specification** Draw an eight-field card with required markers. Add optional identity, clause/function, control, exception, and refresh fields below. A release stamp remains disabled until all required fields are populated and reviewed. **Speaker notes** An applicability card preserves unresolved facts. Evidence means the artifact supporting status, applicability reasoning, or control—not a bare hyperlink. Reviewer role must match the judgment. Uncertainty is not a drafting defect; it can block release. **Teaching check** Give a card with status and URL only. Learners list the seven remaining categories of work. **Alt text** An eight-field applicability card requires jurisdiction, date, actors, scope, force, evidence, reviewer, and uncertainty before review. ## Slide 22 — Never infer paywalled clauses or certification **On-screen text** Public metadata verifies identity and overview scope. It does not supply licensed clauses, conformity, or certification. **Visual specification** Show an official metadata card on the left and a locked full-text volume on the right. A dotted line reaches only title, edition, and overview. Separate blocked arrows point toward clause claim, conformity, and certificate. **Speaker notes** This is especially important for S03. A course may record the published international standard and high-level scope from official metadata. It must not fabricate requirements from title or overview. Qualified review of licensed text and actual organizational evidence remains necessary. **Teaching check** Ask learners to rewrite “certified compliant with S03” when only the public identity page and an internal checklist exist. **Alt text** An official metadata card connects only to identity facts; arrows to clause content, conformity, and certification are blocked. ## Slide 23 — L07 passes the audit and blocks release **On-screen text** `L07 PASS: 6 cases, 1 escape(s), release BLOCKED` AB-05 accessibility/attribution escape · residual score 9 **Visual specification** Use two separate status panels: audit execution `PASS` and release `BLOCKED`. A six-row case table shows three blocked, two detected, one escaped; AB-05 is patterned and linked to the block rule. Never use a green overall badge. **Speaker notes** The audit succeeds by preserving the adverse result. Automated structure passed, while manual review found a lost claim-source relation for assistive technology. Detection evidence does not make the release safe. The claim ceiling excludes safety guarantee, certification, compliance, and legal advice. **Teaching check** Ask why changing the audit status to FAIL would misdescribe the fixture. **Alt text** Separate panels show a successful six-case audit and a blocked release caused by one high residual accessibility-and-attribution escape. ## Slide 24 — Bounded decision: review the rule, do not deploy it **On-screen text** Safeguarded credits: `29.68 · 35.69 · 34.63 · 0.00` Advance to independent review only. Status ≠ applicability ≠ conformity ≠ certification. **Visual specification** Create a signed decision card with allocation, winners/losers, gaming risks, appeal, standards boundary, L07 block, and next reviewer. A prominent `no deployment` band crosses the bottom. No trophy, platform logo, or upward-only arrow. **Speaker notes** The synthetic rule can reward verified evidence under declared contribution, need, authority, and harm choices. It is not proven fair and creates its own incentives. S01–S10 retain distinct statuses; applicability remains unresolved. No paywalled clauses, legal advice, conformity, certification, safety, or production effect is claimed. **Teaching check** Exit prompt: name one rewarded actor, one burdened actor, one gaming path, one unresolved applicability fact, and one claim the package blocks. **Alt text** A decision card advances the safeguarded allocation only to review, records risks and the L07 block, and forbids deployment or certification claims.