# W10 Slide Script — Entities, Authority, and Source Ecosystems Production note: all visuals below are original instructional specifications. They reproduce no external figure, screenshot, logo, or paper artwork. The text and table alternatives are authoritative. ## Slide 01 — From a page count to a corroboration claim **On-screen text** Entities, Authority, and Source Ecosystems. Essential question: How does a fact become corroborated beyond one owned page? Identity → claim → provenance → independent origins → correction. **Visual specification** Create an original horizontal chain with five labeled stations. Use a rounded entity node, hexagonal claim node, three document nodes grouped into origin clusters, a conflict ledger, and a correction loop. Use shapes, labels, and line styles in addition to color. **Speaker notes** Open by rejecting the shortcut that more URLs automatically mean more evidence. A fact can appear on many pages because one press release was syndicated, summarized, and aggregated. Before counting, we must know which entity the pages describe, which atomic claim is at issue, how each page obtained the information, and whether any source performed materially independent verification. W10 ends with an authorized correction and refresh path, not a campaign to manufacture mentions. The package uses PAPER-14 and PAPER-35 as bounded core routes, PAPER-01 only as a quarantined time-provenance audit, PLAT-13 only for vocabulary semantics, and LAB-L05 as a reversible local intervention checkpoint. **Teaching check** Ask: “If five domains repeat one press release, how many independent claim origins are visible?” Expected answer: one, unless another origin is documented. **Alt text** A five-stage chain moves from resolved entity to scoped claim, provenance clusters, preserved conflict, and authorized correction, emphasizing that page count differs from origin count. ## Slide 02 — Outcomes and evidence ceiling **On-screen text** Resolve ambiguous entities. Judge authority per claim. Separate primary, owned, and independent roles. Collapse dependent copies. Preserve conflicts. Design correction, refresh, and rollback. **Visual specification** Design a six-row checklist with an empty evidence-ceiling rail on the right. Each row ends in one of three labeled states: “represented,” “reproduced,” or “not evaluated.” No icons from external libraries are required. **Speaker notes** Explain that learners will produce an entity dossier, a source-ecosystem graph or adjacency table, and a correction workflow. The work is deterministic and offline. The Northbridge records are fictional. LAB-L05 can establish exact local fixture properties, not external response. A source ecosystem can show support, dependence, conflict, and correction state; it cannot by itself prove retrieval, answer use, citation, ranking, traffic, reputation change, or business effect. State the no-recording and no-pilot boundary: all time allocations are authoring budgets, and no classroom outcome has been observed. **Teaching check** Learners name one property the graph can establish and one system outcome it cannot establish. **Alt text** Six learning outcomes align with a boundary rail distinguishing documented representation and deterministic reproduction from unmeasured external outcomes. ## Slide 03 — A name is not an identity **On-screen text** Names can collide, translate, abbreviate, age, or transfer. Resolve with stable ID, canonical name, entity type, jurisdiction, registry key, aliases, dates, and ambiguity status. **Visual specification** Show three candidate cards labeled “NDC.” Connect only one card to `ENT-NBDC-001` through matching jurisdiction and registry-key gates. Cross the abbreviation-only path with a labeled “insufficient” bar, not a red-only signal. **Speaker notes** Introduce the fictional Northbridge case. `Northbridge Data` and historical `Northbridge Cooperative` are accepted aliases under dated records. `NDC` is ambiguous across supplied candidates. A famous-looking source cannot solve that ambiguity without a resolvable identifier. The project stable ID is a governed key connected to external identifiers; it is not a claim of universal identity. Record rejected candidates and reasons because reproducibility includes alternatives, not only the chosen match. **Teaching check** What extra field makes `NDC` usable here? Learners should request jurisdiction plus the matching registry key, not more name similarity. **Alt text** Three entities share the abbreviation NDC; jurisdiction and registry-key gates resolve only Northbridge Data Cooperative to the stable project identifier. ## Slide 04 — The entity dossier **On-screen text** `ENT-NBDC-001` | Northbridge Data Cooperative Ltd. | organization | `EX-NORTH` | `EXN-88421` | accepted and rejected aliases | reviewed 20 July 2026. **Visual specification** Create an original dossier sheet with a fixed identity header, an alias timeline, an external-identifier table, and an ambiguity log. Use numbered reading order and a plain-text table underneath for screen-reader equivalence. **Speaker notes** Walk through the minimum dossier. The stable project ID remains constant when a display name changes. Aliases have type and effective dates: legal, trading, translated, historical, abbreviation, or rejected collision. Jurisdiction and entity type constrain matches. External identifiers identify their issuing authority and namespace. Resolution status can be resolved, provisional, ambiguous, split, merged, or retired. The reviewer and date expose when the decision was made. None of these fields establishes the truth of a product claim; they establish the subject to which later claim edges attach. **Teaching check** Ask learners to identify the field that should change after a rebrand and the field that should normally remain stable. **Alt text** An entity dossier combines stable identity fields, a dated alias timeline, external identifiers, and an explicit ambiguity log. ## Slide 05 — Jurisdiction changes meaning **On-screen text** The same label can refer to different legal entities, operators, products, or policies by jurisdiction. Jurisdiction is a claim field, not background decoration. **Visual specification** Draw two parallel jurisdiction lanes, `EX-NORTH` and `EX-SOUTH`, each with a same-named product but different operator and policy node. A gate blocks cross-lane evidence unless an explicit applicability edge exists. **Speaker notes** Explain that registry identifiers are often unique only inside a namespace. Brand licensing can produce different operators by country. A policy may apply in one locale and not another. Even an independent audit is inapplicable if it tests the wrong jurisdiction. Record jurisdiction both in the entity dossier and the scoped claim. If a source does not state geographic applicability, classify it as unknown rather than assuming global scope. **Teaching check** Can an `EX-SOUTH` policy corroborate an `EX-NORTH` retention claim? Only with evidence that the policies and operators share the relevant scope. **Alt text** Two jurisdiction lanes show same-named products with different operators and policies; evidence cannot cross lanes without a documented applicability relation. ## Slide 06 — Atomize and scope the claim **On-screen text** Entity + predicate + value/unit + plan + effective interval + jurisdiction/locale + conditions + status. One predicate per governed claim. **Visual specification** Turn the loose sentence “Northbridge stores exports securely for 90 days and complies everywhere” into three stacked claim cards. Highlight retention as the only card evaluated in the fixture; label security and compliance “not evaluated.” **Speaker notes** A compound sentence can hide several evidence burdens. The frozen claim is `CLM-RETENTION-001`: standard-service audit-export retention equals 90 days in `EX-NORTH`, effective 15 July 2026, current under review. Security and legal compliance are different predicates and cannot borrow support from the retention evidence. Dates also have roles: publication, observation, retrieval, effective, reviewed, and superseded. Keep them separate so a new access date does not make an old policy current. **Teaching check** Learners rewrite one promotional sentence as atomic claim records and mark which records lack evidence. **Alt text** A bundled statement splits into retention, security, and compliance claims; only the precisely scoped retention card enters the case. ## Slide 07 — Authority is an edge, not a crown **On-screen text** Ask “authoritative for which claim, scope, and date?” Never assign a permanent global authority rank to a publisher. **Visual specification** Create a matrix with claims as rows and sources as columns. Cells contain labeled edge states: strong applicable, partial, inapplicable, stale, or unknown. Do not use stars, crowns, or a single domain score. **Speaker notes** The fictional registry is strong for legal name and jurisdiction but silent on retention. Current owned documentation is primary for the declared product specification. The independent audit is direct and applicable to the tested plan. A comparison page may be independent yet stale and scoped to a legacy plan. This matrix prevents reputation halo: a source strong for one row does not become strong for every row. Assess directness, temporal fit, jurisdictional fit, method, and dependence for each edge. **Teaching check** Why can the registry resolve identity but not corroborate 90-day retention? It never addresses that predicate. **Alt text** A claim-by-source matrix assigns different applicability states to each relationship instead of giving publishers one universal authority score. ## Slide 08 — Primary, owned, and independent are not a ladder **On-screen text** Primary = direct record of the fact. Owned = controlled by the implicated entity. Independent = materially separate from the claim origin. Properties can overlap. **Visual specification** Use an original three-circle relationship diagram without implying vertical rank. Place current vendor documentation in primary-plus-owned, registry in primary-for-identity, and independent audit in independent-plus-direct-observation. Add text labels outside every region. **Speaker notes** An owned source is not automatically weak. It may be the appropriate primary record for a declared price, setting, or policy. It remains insufficient for independent comparative or reputation claims. An external page is not automatically independent: financial sponsorship, shared data feeds, quotation, or press-release dependence may matter. Record relation fields instead of using “first party” and “third party” as verdicts. The purpose is to match evidence role to claim. **Teaching check** Name a source that can be both primary and owned. Then name a claim for which that same source would be insufficient. **Alt text** Three overlapping source properties show that directness, ownership, and independence are separate axes rather than a prestige ranking. ## Slide 09 — Build a typed provenance graph **On-screen text** Nodes: entity, claim, artifact, origin organization, correction. Edges: asserts, observes, quotes, syndicates, transforms, contradicts, supersedes, corrects. **Visual specification** Draw a graph legend with five original node shapes and eight labeled line types. Provide an adjacent table with `from`, `to`, `edge_type`, `origin_cluster`, `date`, and `review_status` columns. **Speaker notes** A bibliography tells us which pages exist; a provenance graph tells us how they relate. Typed edges record material reliance, direct observation, transformation, and correction. Each edge carries scope and date. A `quotes` or `syndicates` edge is not weaker by moral definition; it simply does not add an independent origin. A correction edge does not erase the earlier node. The adjacency table is authoritative and makes the diagram reproducible for screen-reader and low-graphics routes. **Teaching check** Which edge would connect an aggregator sentence to the news story it extracted? `transforms` or `quotes`, with a dependence note. **Alt text** A provenance legend maps five node roles and eight typed relations, paired with a table that fully represents the graph. ## Slide 10 — Syndication is distribution, not corroboration **On-screen text** Four pages can equal one origin: owned release → wire copy → edited story → aggregator. Report page count and independent-origin count separately. **Visual specification** Show four document nodes nested inside one dashed enclosure labeled `ORG-OWN-01`. A separate independent-audit node sits in `ORG-INDEPENDENT-01`. Display counters: five pages, two relevant origin clusters, one independent corroborating origin. **Speaker notes** Dependence follows material information flow, not domain names. If every downstream page derives the 90-day value from the same release, those pages add distribution and correction burden but not independent verification. Preserve them because reach and staleness matter. Simply do not convert them into multiple votes. Origin clustering requires evidence: explicit attribution, near-identical text, shared feeds, or documented syndication. If dependence is unknown, mark unknown instead of assuming independence. **Teaching check** Ask learners for three counts: pages, material origins, and applicable independent corroborators. Require different labels for each. **Alt text** Four dependent publication pages sit inside one origin cluster, while a separate audit supplies the only independent corroborating origin. ## Slide 11 — Apparent abundance, actual support **On-screen text** Count only applicable edges. Registry: identity yes, retention no. Current docs: declared specification yes. Audit: observed match yes. Syndication: no new origin. **Visual specification** Create a funnel from six supplied source artifacts to four applicable retention edges, then to two material origins, then to one independent corroborating origin. Label every exclusion reason beside the funnel. **Speaker notes** This slide turns source counting into an auditable sequence. First filter by entity match. Then filter by predicate, plan, jurisdiction, and time. Next collapse material dependence into origin clusters. Finally distinguish owned primary support from independent corroboration. Do not use majority voting. The most numerous branch may be copied; the most prestigious page may be inapplicable. The final result must preserve excluded and conflicting sources with reasons. **Teaching check** Why is the registry excluded from retention corroboration even though it is independent and official? Claim applicability fails. **Alt text** A labeled funnel reduces six artifacts to applicable edges, material origins, and one independent corroborating origin without discarding excluded records. ## Slide 12 — Corroboration is a bounded rule **On-screen text** Instructional rule: applicable primary support + applicable materially independent origin + preserved conflicts + explicit date and scope. Risk can require more. **Visual specification** Design a four-lock gate. Each lock is labeled primary applicability, independent applicability, conflict accounting, and temporal scope. Beyond the gate place a badge reading “corroborated within frozen case,” never simply “true.” **Speaker notes** W10 uses a transparent instructional threshold, not a universal law. High-risk health, legal, financial, or safety claims can require multiple independent origins, domain review, regulatory evidence, and direct testing. Even here, corroboration is bounded to the supplied records and date. It does not establish downstream visibility. A clear threshold is valuable because another reviewer can reproduce the decision and challenge a specific lock. **Teaching check** Which phrase keeps the conclusion bounded? “Within the supplied synthetic records, as of 20 July 2026.” **Alt text** Four labeled locks must close before the claim receives a bounded frozen-case corroboration status rather than an unrestricted truth label. ## Slide 13 — Conflict states preserve history **On-screen text** Exact contradiction. Temporal supersession. Plan mismatch. Jurisdiction mismatch. Definition mismatch. Correction. Unverifiable attribution. Unresolved. **Visual specification** Create eight labeled drawers in a conflict ledger. Place the 60-day PDF in temporal supersession and the 30-day comparison in plan/time mismatch. Keep an empty unresolved drawer visibly available. **Speaker notes** Do not delete an old page because it complicates the story. The 60-day owned PDF predates the current effective date and is marked superseded. The 30-day independent comparison describes the legacy plan in February 2025, so it is not comparable to the July 2026 standard-plan claim. Both remain in the graph because downstream copies can persist. If a credible current source contradicts the exact scoped claim, status becomes unresolved until investigation closes. **Teaching check** Is a superseded record “false”? Not necessarily; it may accurately describe an earlier state. **Alt text** Eight conflict drawers classify old and mismatched records while preserving an explicit place for genuinely unresolved contradictions. ## Slide 14 — Freshness is predicate-specific **On-screen text** Publication date ≠ observation date ≠ effective date ≠ retrieval date ≠ review date ≠ superseded date. **Visual specification** Draw six parallel date tracks for one source card. Add two examples: an old registry record still current and a recent aggregator already stale. Use labels and numeric dates, not relative color intensity. **Speaker notes** An access date shows when we saw a page, not when its claim became true. Publication and observation can differ. Effective dates determine which policy applies. Review dates record governance. Superseded dates preserve transitions. Staleness depends on claim volatility: an incorporation fact may persist for years, while a price can change quickly. The course freeze also matters. PAPER-01 is quarantined because the local full text contains collection or API dates after 24 August 2026. No substantive finding is admitted from that record. **Teaching check** Which date would you use to decide whether the 90-day policy applies on 18 July 2026? The effective interval, not retrieval date. **Alt text** Six separate date tracks demonstrate why recent retrieval cannot substitute for publication, observation, effective, review, or supersession dates. ## Slide 15 — The deterministic Northbridge fixture **On-screen text** Fictional entity `ENT-NBDC-001`; claim `CLM-RETENTION-001`; standard plan; 90 days; `EX-NORTH`; effective 15 July 2026; frozen 20 July 2026. **Visual specification** Create a case header card with a diagonal “synthetic fixture” watermark rendered as text. Place entity keys on the left, scoped claim fields on the right, and a frozen-date seal below. Avoid resemblance to a real brand. **Speaker notes** State that every name and record is authored for instruction. The fixture avoids claims about a real organization. Stable IDs allow exact joins. The freeze date bounds the graph. Learners receive source cards and do not crawl the web. Their task is to reproduce resolution, classification, clustering, conflict disposition, and correction design. A different decision must cite a changed rule or input, not intuition. **Teaching check** Ask learners to read the claim aloud with plan, jurisdiction, and effective date. Reject any version that drops scope. **Alt text** A clearly labeled synthetic case card displays the stable entity key, atomic retention claim, jurisdiction, effective date, and frozen review date. ## Slide 16 — Source-card audit **On-screen text** Registry; current owned docs; superseded owned PDF; independent audit; syndicated story; aggregator; legacy comparison. Role and applicability differ by claim. **Visual specification** Make a seven-row source table with columns for role, claim applicability, date, plan, origin cluster, independence, and status. Use words such as applicable, inapplicable, dependent, stale, and superseded in every relevant cell. **Speaker notes** Walk row by row. `SRC-REG-01` resolves legal identity only. `SRC-OWN-01` supplies current declared retention. `SRC-OWN-02` is superseded at 60 days. `SRC-IND-01` independently observes availability through day 90. `SRC-SYN-01` and `SRC-AGG-01` trace to the owned origin. `SRC-CON-01` reports 30 days for the legacy plan and earlier period. The table makes the graph’s decision process inspectable. **Teaching check** Which source is independent but inapplicable to the retention predicate? The official registry. **Alt text** Seven source cards are compared by role, scope, date, origin, independence, and conflict status rather than domain reputation. ## Slide 17 — Reproduce the bounded conclusion **On-screen text** As of 20 July 2026: current owned primary support + one matching independent audit origin. Dependent copies add reach, not corroboration. Conflicts remain recorded. **Visual specification** Show a calculation panel with three explicit values: displaying pages, relevant origin clusters, and independent corroborating origins. Under it, print the full bounded conclusion and an “external outcomes not evaluated” strip. **Speaker notes** The result is neither “seven sources agree” nor “the fact is universally true.” It is a reproducible judgment under a frozen rule. The current owned document and independent audit converge on the exact scoped value. The registry contributes identity, not retention support. Syndication stays inside one origin. The old and legacy conflicts are classified and preserved. The conclusion has a refresh trigger and no claim about retrieval, answer use, citation, ranking, traffic, or business effect. **Teaching check** Have learners state the result in two sentences: one support sentence and one evidence-ceiling sentence. **Alt text** A count panel and bounded paragraph distinguish publication reach, material origins, independent corroboration, conflicts, and unevaluated external outcomes. ## Slide 18 — Authorized correction workflow **On-screen text** Detect → preserve → verify entity/scope → assign owner/reviewer → correct authorized source → record change → notify legitimate channels → refresh graph. **Visual specification** Design an eight-stage circular workflow with a side branch for “unresolved: freeze formulation.” Every stage writes to an append-only change ledger. A rollback arrow returns from validation failure to the preserved prior version. **Speaker notes** Correction begins with evidence preservation. The team verifies the entity and atomic claim, then assigns an authorized owner and a separate reviewer. Changes occur only on surfaces the owner controls or through the publisher’s legitimate correction route. The record stores old value, new value, reason, approver, effective date, and rollback target. If the conflict cannot be resolved, the public formulation remains narrowed or frozen. This workflow is governance, not a visibility guarantee. **Teaching check** Why preserve the prior source before correction? It supports audit, conflict history, propagation analysis, and rollback. **Alt text** An eight-stage correction loop preserves evidence, verifies scope, records authorized changes, tracks legitimate downstream requests, and supports rollback. ## Slide 19 — Downstream status is not presumed **On-screen text** Notified. Acknowledged. Corrected. Declined. Unresponsive. No authorized contact. A correction request is not proof of downstream change. **Visual specification** Create five downstream document branches from the corrected primary record. Each branch ends in a written status badge and date. Use dotted request arrows and solid confirmed-correction arrows; include a table equivalent. **Speaker notes** Publishers control their own records. Use errata forms, editorial contacts, database feedback, or contractually authorized feeds. Cite the exact corrected statement and source. Do not demand favorable coverage or disguise placement. Some publishers will add a correction note; others may leave historically accurate text unchanged. Aggregators can refresh later. Keep nonresponsive and declined states visible instead of claiming the ecosystem is clean. **Teaching check** What state follows sending an email? `notified`, not `corrected`, unless confirmation or observed revision exists. **Alt text** Downstream branches carry explicit notification and correction states, showing that requests and confirmed changes are different events. ## Slide 20 — Update triggers and review ownership **On-screen text** Triggers: legal identity, ownership, alias collision, plan revision, policy date, source correction, new contradiction, broken link, expired review, missed deadline. **Visual specification** Draw a trigger board feeding a review queue. Each card includes trigger ID, entity or claim scope, owner, reviewer, opened date, due date, disposition, and next review. Use a deterministic priority rule printed below. **Speaker notes** Calendar review alone misses important events. Predicate-specific triggers open review tasks. The owner investigates; a separate reviewer confirms disposition when risk warrants. A legal-name change may update identity and aliases. A plan revision may supersede a claim. A source correction may reopen dependent branches. A broken URL affects accessibility but does not automatically negate historical evidence. Every trigger ends with an explicit state and next action. **Teaching check** Ask for one trigger that changes identity and one that changes only a claim version. **Alt text** Ten event triggers feed an owned review queue with explicit scope, dates, disposition, and next-review fields. ## Slide 21 — LAB-L05: one changed factor **On-screen text** Three locked claims. Three fixed source identities. One factor: evidence-layout proximity. Local fixture. Hypothesized human-comprehension stage. No external endpoint. **Visual specification** Show control and treatment page wireframes side by side. Keep all claim and source blocks identical; change only the visible distance between each claim and its source. Add a checksum strip and a one-factor lock icon rendered from basic shapes. **Speaker notes** LAB-L05 is a deterministic local intervention. The HTML pair contains the same three claims and source references. Stylesheets change evidence-layout proximity only. The intervention card authorizes a course-owned synthetic fixture with no external endpoint or user traffic. Its expected later test concerns comprehension, but this audit does not measure comprehension. It validates factual equivalence, declared scope, source identity, and rollback preparation. **Teaching check** If the treatment changes a claim word, can the one-factor test proceed? No; factual equivalence fails and the studio stops. **Alt text** Two local page wireframes contain identical claims and sources while only the distance between evidence and claim changes. ## Slide 22 — LAB-L05 pass and rollback boundary **On-screen text** Expected: `L05 PASS: 3 locked claims, 1 changed factor, 0 error(s)`. PASS means local structural equality and scope, not system response. **Visual specification** Create a rollback manifest diagram mapping treatment HTML and CSS hashes to control HTML and CSS hashes. Beneath it, list outputs: audit, normalized claims, diff, equivalence review, rollback manifest, and run manifest. **Speaker notes** Reproduce LAB-L05 into a fresh temporary directory. Inspect outputs rather than trusting the terminal line. The rollback manifest must identify the exact treatment-to-control pairs. Stop on claim, attribution, accessibility, or hash mismatch. A pass does not establish crawler behavior, retrieval, ranking, answer use, citation, traffic, business effect, or cross-engine generalization. It also does not prove classroom learning or accessibility conformance. **Teaching check** Ask learners to complete: “LAB-L05 proves ___ and does not prove ___.” Require local fixture equality in the first blank. **Alt text** A hash-linked rollback diagram and six-output checklist define what the deterministic LAB-L05 pass does and does not establish. ## Slide 23 — Reading and vocabulary boundaries **On-screen text** PAPER-14: metadata-limited catalog-only context. PAPER-35: ecosystem route, preprint. PAPER-01: quarantined time audit only. PLAT-13: vocabulary semantics only. LAB-L05: local equivalence only. **Visual specification** Build a five-row evidence card table with columns for public ID, admitted role, status, and prohibited inference. Make the PAPER-01 row crosshatched and label it “audit-only quarantine,” with no substantive claims displayed. **Speaker notes** Source boundaries travel with the source. The latest local CSV audit classifies PAPER-14 as Notes catalog-only and reports missing URL, abstract, and access-date metadata in the export; it is a context prompt, not substantively absorbed evidence. PAPER-35 supports evidence-ecosystem thinking within its architecture and study conditions. PAPER-01 is not mined for findings because its local date provenance crosses the course freeze. PLAT-13 describes a shared vocabulary but cannot validate truth, authority, independence, visibility, or ranking. LAB-L05 is authored local data and code, not a platform experiment. **Teaching check** Which substantive conclusion may W10 take from PAPER-01? None. Only the provenance-failure lesson is admitted. **Alt text** Five evidence cards pair each public identifier with an admitted instructional role and a clearly stated inference boundary. ## Slide 24 — Exit test: can another reviewer reproduce it? **On-screen text** Resolved entity? Atomic scoped claim? Claim-relative authority? Origin clusters? Preserved conflicts? Authorized correction? Trigger and rollback? Bounded conclusion? **Visual specification** Create an eight-gate exit path ending at a versioned evidence package. Beside it, show the required files: dossier, claim table, source cards, edges, clusters, conflicts, corrections, hashes, and boundary statement. **Speaker notes** Close by asking learners to hand the package to a skeptical second reviewer. The reviewer should reconstruct entity resolution, source applicability, dependence clusters, conflict decisions, correction states, and the final conclusion without network access. If a decision depends on familiarity, domain prestige, hidden deletion, or an undocumented judgment, the package is incomplete. Prohibit fabricated third-party publication, undisclosed sponsorship, mass duplication, and circular citation structures. The desired result is accountable evidence, not manufactured independence. Remind learners that no W10 recording exists and no timed pilot has occurred. **Teaching check** Final prompt: “How many pages display the claim, how many origins exist, how many are independent and applicable, and as of when?” **Alt text** Eight reproducibility gates lead to a versioned evidence package containing identity, claims, sources, graph edges, conflicts, corrections, hashes, and boundaries.