# W10 No-Video Equivalent — Entities, Authority, and Source Ecosystems

## Status and use

No W10 recording exists. This document is a planned, readable no-video equivalent and a future narration script. It is not a transcript of delivered instruction. The chapter durations below are instructional planning budgets, not observed media timestamps or classroom data. Any later recording requires verified duration, captions, speaker labels, graph description, keyboard-accessible navigation, and human accessibility review.

The route is designed around the essential question: How does a fact become corroborated beyond one owned page? It uses a fictional source ecosystem and frozen local files. It makes no claim about a real entity, commercial system, hidden retrieval process, or production outcome.

## Planned chapter budget

| Chapter | Focus | Planned minutes |
|---:|---|---:|
| 1 | Identity before authority | 4 |
| 2 | Scoped claims and dates | 4 |
| 3 | Claim-relative authority | 4 |
| 4 | Primary, owned, and independent roles | 4 |
| 5 | Provenance, syndication, and origin clusters | 5 |
| 6 | Northbridge source-card reproduction | 5 |
| 7 | Conflict, staleness, and quarantine | 4 |
| 8 | Authorized correction and refresh | 4 |
| 9 | LAB-L05 intervention and rollback | 4 |
| 10 | Deterministic conclusion and exit test | 4 |

**Total planned route: 42 minutes.** The acceptable no-video design range is 35–45 minutes, pending actual rehearsal.

## Chapter 1

Let us begin with the tempting but wrong question: how many pages say the same thing? A search can return an owned page, a news story, an industry directory, and an aggregator, all repeating the same number. Four domains look like four confirmations. Yet the news story may quote a press release, the directory may ingest the vendor feed, and the aggregator may extract the news story. We have four publication artifacts but perhaps only one information origin.

Before we can inspect that dependence, we must know what the statement is about. Names are not stable identities. An organization can have a legal name, a trading name, a translated label, an old name, an abbreviation, and a product name that resembles the parent company. The same abbreviation can refer to several organizations. A brand can be operated by different legal entities in different jurisdictions.

Our fictional case uses Northbridge Data Cooperative Ltd. Its course-controlled stable ID is `ENT-NBDC-001`. Its fictional jurisdiction is `EX-NORTH`, and its supplied registry key is `EXN-88421`. `Northbridge Data` is an accepted current alias. `Northbridge Cooperative` is accepted as a historical alias through a stated date. The abbreviation `NDC` is not accepted by itself because the supplied candidate set contains collisions.

Notice what the stable ID does and does not do. It gives this governed project a durable join key. It does not claim that every external database will use the same ID or that identity is settled forever. We record canonical name, entity type, jurisdiction, external identifiers and their namespaces, accepted and rejected aliases, effective dates, resolution status, reviewer, and review date. A second reviewer can inspect the same candidate cards and reproduce the choice.

Jurisdiction is not a footnote. A registry number can be unique only in the issuing system. A product with the same brand can have a different operator or policy elsewhere. If a source does not identify the relevant jurisdiction, we mark applicability unknown rather than assuming global scope.

Here is the first teaching check. Suppose a highly regarded publication calls an organization “NDC,” but provides no registry key, jurisdiction, address, or parent relation. Has the publisher’s reputation resolved the entity? No. Authority cannot repair an unresolved subject. We need identity evidence before evaluating claim evidence.

## Chapter 2

Once the subject is resolved, we still do not have a claim that can be corroborated. “Northbridge stores exports securely for 90 days” bundles at least two predicates: retention duration and security. If it adds “and complies with all regulations,” there is a third predicate with a much larger jurisdictional burden. One source may support retention while saying nothing about security or compliance. We therefore split the sentence into atomic claim records.

The frozen claim in this lesson is `CLM-RETENTION-001`. Its entity is `ENT-NBDC-001`. Its predicate is retention duration for audit exports. Its value is 90 days. Its product scope is the standard service. Its jurisdiction is `EX-NORTH`. Its effective date begins on 15 July 2026. At the freeze date, its status is current under review. Security, legal compliance, deletion guarantee, and availability service level remain outside the claim.

This record forces us to distinguish dates. A source may have a publication date, observation date, retrieval date, effective date, reviewed date, and superseded date. A page retrieved today can contain last year’s policy. A news story published today can repeat an old release. An old registry record can remain current. A recent price page can already be obsolete. Freshness belongs to the predicate and its change process, not simply the age of the page.

Status also matters. `proposed`, `current`, `disputed`, `superseded`, `corrected`, and `unresolved` should not become one binary flag. A 60-day policy can have been correct for a prior plan. If a new policy supersedes it, we preserve the old record and connect it to the new record. We do not delete the history and then pretend that every source always agreed.

The test for this chapter is to read the claim aloud without dropping fields: Northbridge Data Cooperative Ltd., standard service, audit-export retention, 90 days, `EX-NORTH`, effective from 15 July 2026, current under review. If that feels wordier than a slogan, that is the point. Corroboration requires a target precise enough for sources to converge or conflict in a meaningful way.

## Chapter 3

Now we can discuss authority. The most important sentence in W10 is this: authority is claim-relative. A publisher does not wear a permanent crown that transfers to every fact on its domain.

Consider the supplied registry. It is the appropriate primary record for a legal name, registration key, and jurisdiction. It does not mention audit-export retention. Its institutional status cannot make it evidence for that predicate. Conversely, current product documentation controlled by the entity can be the proper primary record for its declared setting. The fact that it is owned does not make it useless. It tells us what the organization currently represents as its specification. It is not enough, however, to establish that the product is better than competitors, trusted by users, or independently verified.

Build an authority matrix with atomic claims in rows and sources in columns. Each cell asks: does this source address this predicate? Does the entity match? Does the plan match? Does the jurisdiction match? Does the effective time match? Is the source direct or transformed? What method produced the fact? Is the source independent of the claim origin? Is there a conflict or correction?

This prevents reputation halo. A regulator can be authoritative for its own order while an appellate record controls a later legal status. A researcher is primary for the data that researcher collected, but that does not make every interpretation uncontested. A product owner is primary for its declared plan name while an external audit may be needed for observed performance.

PAPER-14 appears in the W10 core route, but its current handling must be unusually explicit. The latest local CSV audit says it is catalog-only in the Notes absorption state and lacks URL, abstract, and access-date metadata in that export. The local paper catalog marks its status as preprint/venue not confirmed. W10 therefore uses PAPER-14 only as a metadata-limited context prompt about reputation and discovery, not as substantively absorbed evidence or validation of this case.

PAPER-35 is a preprint route for thinking about trajectory-aware evidence ecosystems. Its architecture, data, and results retain their study-specific limits. It does not reveal the hidden architecture of a named production system, and it does not validate Northbridge.

Our reporting rule is: state what a source is authoritative for, under what scope and date, rather than declaring the source authoritative in general.

## Chapter 4

Primary, owned, and independent are separate properties, not rungs on a ladder.

A primary source directly records or produces the fact at issue. An official registry is primary for its registration record. A signed contract is primary for its terms. Current authorized product documentation can be primary for the organization’s declared specification. A researcher’s dataset is primary for the observations that researcher collected.

An owned source is controlled by the entity whose interests are implicated. Ownership raises governance and incentive questions, but it does not erase directness. An independent source is organizationally and materially separate from the claim origin. Independence is valuable only when the source is applicable, competent, methodologically adequate, and sufficiently current.

These properties overlap. The current Northbridge document is both primary and owned for the declared 90-day setting. The fictional procurement audit is independent and directly observes access through day 90 under the matching standard plan. The registry is independent of the product-documentation channel and primary for legal identity, yet it is inapplicable to retention. A news article can be institutionally external but materially dependent when it paraphrases a press release without verification. An ostensibly independent directory can depend on the same vendor feed.

For each source card, record origin organization, responsible author or unit, method, direct observation or quotation, funding and sponsorship where known, transformed sources, dates, correction route, and claim applicability. “Third party” is too vague. A commissioned audit might include useful direct observations while retaining a financial relation that must be disclosed. Neither disclosure nor independence alone tells us whether the method supports the claim.

PLAT-13 enters only as a candidate vocabulary route. It documents shared structured-data semantics that can help represent an entity, alias, identifier, and page relation consistently. That can improve machine-readable clarity. It does not verify the values, establish publisher independence, prove that a crawler consumed the record, or demonstrate a visibility or ranking effect. Syntax is not truth. Representation is not corroboration.

The chapter check is a pair of questions. Can an owned page be primary? Yes, for a suitably scoped first-party fact. Can an external domain still be dependent? Yes, when its material information comes from the same origin.

## Chapter 5

A list of references shows which artifacts exist. A provenance graph shows how information moved among them. W10 uses nodes for entities, atomic claims, source artifacts, origin organizations, and correction events. It uses typed edges such as `identifies`, `asserts`, `observes-support`, `quotes`, `syndicates`, `transforms`, `contradicts`, `supersedes`, and `corrects`.

Every edge needs scope, date, and review state. `SRC-OWN-01 asserts CLM-RETENTION-001` is incomplete unless we know it refers to the standard plan and the effective interval. `SRC-SYN-01 syndicates SRC-OWN-01` records dependence. `SRC-IND-01 observes-support CLM-RETENTION-001` records direct testing through day 90. A source may have several edges for different claims.

Material dependence is more important than separate domains. Imagine an owned release, a wire copy, an edited news page, and an aggregator extraction. The wording can change while the fact still comes from one origin. Those four pages form one origin cluster until evidence shows independent verification. We should preserve all four artifacts because they matter for distribution and correction burden. We should not count them as four corroborators.

The graph must have a text equivalent. Use an adjacency table with `from_id`, `to_id`, `edge_type`, `origin_cluster`, `effective_date`, `review_status`, and `note`. Sort by claim and origin cluster. Validate that every endpoint exists and every edge type is declared. This table supports screen readers and also makes deterministic reproduction easier.

For the case, four publication artifacts display the current 90-day value: the current owned document, independent audit, syndicated story, and aggregator. After dependence clustering, they reduce to two relevant origins: `ORG-OWN-01` and `ORG-INDEPENDENT-01`. Only the second is materially independent of the claim origin. So we report four displaying pages, two applicable material origins, and one applicable independent corroborating origin.

Corroboration here is a bounded instructional decision. The rule requires applicable primary support, at least one applicable materially independent origin, preserved conflicts, and explicit temporal scope. This is not a universal minimum. A medical, legal, safety, or financial claim could require several independent studies, qualified review, and different evidence standards.

The key sentence is: syndication can extend distribution without adding an independent origin. Reach is a useful property. It is simply not the same property as corroboration.

In real source work, dependence may not be fully observable. Do not treat “no syndication label found” as proof of independence. Give the relation a state. Confirmed dependence might come from explicit attribution, a publisher’s feed disclosure, or a traceable quotation chain. Probable dependence might come from distinctive shared errors or near-identical phrasing, but it still needs reviewer justification. Confirmed independence requires positive evidence about a separate method or observation, not just a different hostname. Unknown remains a valid state.

One useful sensitivity exercise reports the decision under two rules. The narrow rule collapses only confirmed dependence. The conservative rule also collapses probable dependence. If the corroboration conclusion changes, say so and make provenance verification the next action. Our frozen Northbridge fixture does not have this ambiguity because the source cards explicitly declare the syndication and aggregation paths. The exercise nevertheless prevents learners from carrying a deterministic classroom assumption into open-web work without qualification.

## Chapter 6

Let us reproduce the Northbridge source decision row by row.

`SRC-REG-01` is the official fictional registry card. It connects the canonical name, `EX-NORTH`, and `EXN-88421`. It resolves the subject. Because it does not address product retention, it contributes zero applicable retention edges.

`SRC-OWN-01` is the current owned standard-plan documentation, reviewed 15 July 2026. It states 90 days. It is primary and owned for the declared specification, directly applicable, and assigned to `ORG-OWN-01`.

`SRC-OWN-02` is an owned PDF dated 1 November 2025 that states 60 days. It is evidence for a prior state, not a current contradiction. We mark it superseded and retain it in the same owned origin cluster.

`SRC-IND-01` is a fictional independent procurement audit dated 20 July 2026. It tests audit-export availability through day 90 under the standard plan. It is a materially independent applicable origin, `ORG-INDEPENDENT-01`. Its role is direct observation within the supplied case, not a guarantee beyond its method.

`SRC-SYN-01` is a news page that repeats the owned release. `SRC-AGG-01` extracts that news statement. Both are separate pages but trace to `ORG-OWN-01`. They add two publication nodes and no new independent origin.

`SRC-CON-01` is an independent comparison page dated February 2025 that says 30 days. It describes a legacy plan. That makes it a scope and time mismatch, not a refutation of the July 2026 standard-plan claim. We preserve it under `ORG-LEGACY-01` because people may still encounter it and because it should trigger clearer plan/date labeling.

Now apply the decision rule. The entity is resolved. The claim is atomic. `SRC-OWN-01` supplies applicable primary support. `SRC-IND-01` supplies one applicable independent origin. The dependent branches are collapsed. The 60-day record is superseded. The 30-day comparison is a preserved legacy mismatch. Therefore, within the supplied synthetic records and as of 20 July 2026, the 90-day standard-plan claim is corroborated beyond one owned page.

That conclusion does not say “seven sources agree.” It does not claim universal truth. It does not describe a real entity. It does not infer retrieval, answer use, citation, ranking, traffic, or business outcome. It is a frozen evidence judgment with a refresh condition.

## Chapter 7

Conflicts are governed records, not clutter. Removing old or contradictory sources destroys the history needed for audit and correction.

Classify differences before resolving them. Useful states include exact contradiction under matching scope, temporal supersession, plan mismatch, jurisdiction mismatch, definition mismatch, transcription error, correction or retraction, unverifiable attribution, and unresolved conflict. Each conflict entry connects claim versions and sources, records relevant dates and scopes, assigns a reviewer, and states the next action.

The Northbridge 60-day PDF is temporally superseded. It may have been accurate for its period. The 30-day comparison is independent but not comparable because it describes another plan and date. If a credible audit after 15 July 2026 found only 30 days on the standard plan in `EX-NORTH`, that would be an exact-scope contradiction. The public formulation should narrow or freeze while the team investigates.

Staleness is predicate-specific. Legal identity may change slowly. Product settings can change quickly. Define both review intervals and event triggers. A recent aggregator can be stale if it copied an old source. An old registry record can still control identity if unchanged.

PAPER-01 provides the course’s audit-only caution. The local record has a DOI and title, but the full text contains collection or API dates after the 24 August 2026 course freeze. W10 does not speculate about how to reconcile that discrepancy. It preserves the quarantine status and uses PAPER-01 only to teach that metadata plausibility cannot override incoherent time provenance. No substantive empirical finding enters the lesson.

That boundary is an example of good uncertainty handling. The right response to unresolved provenance is not to invent a date, silently ignore the conflict, or average incompatible records. It is to quarantine, document why, and define what evidence would be required to reopen the record.

## Chapter 8

A correction loop begins with a disputed atomic claim, not with a request for more mentions.

First, preserve the current evidence. Record hashes or snapshots and the time of observation. Second, verify the entity, predicate, plan, jurisdiction, and effective interval. Third, assign an owner authorized to change the primary surface and a separate reviewer. Fourth, decide whether the issue is error, staleness, ambiguity, unsupported wording, or unresolved contradiction.

If correction is warranted, update only a surface the owner is authorized to maintain. Record the prior representation, new representation, reason, approver, effective date, and rollback target. In the fixture, the owner can add a missing effective date and a visible change note to `SRC-OWN-01`. That improves the governed primary record. It does not guarantee an external response.

For downstream artifacts, use legitimate publisher correction channels: an errata form, editorial contact, database feedback route, or contractually authorized feed. Identify the exact affected sentence and link the corrected primary evidence. Then record state. `notified` means a request was sent. `acknowledged` means receipt was confirmed. `corrected` requires a confirmed or observed change. Other possible states are `declined`, `unresponsive`, and `not-contacted-no-authority`.

Never recommend fabricated third-party publication, undisclosed sponsorship, mass duplication, paid coverage represented as independent, or circular citation networks. These tactics manufacture apparent corroboration and destroy the independence model. A legitimate intervention improves authorized primary evidence and routes genuine errors through real processes.

Refresh triggers include a legal-name or jurisdiction change, ownership transfer, alias collision, plan revision, policy effective date, source correction, credible new contradiction, expired audit, broken correction route, or missed review deadline. Each trigger opens a dated task with owner, scope, disposition, and next review.

The ecosystem is never declared permanently clean. Nonresponsive branches remain visible. A cached page can persist. A publisher may preserve historical text while adding an editorial note. Correction state is observed per node, not inferred from the request.

The correction log also needs role separation. The content owner can propose a revision, but a reviewer should verify the evidence, scope, and downstream notice before closure when the claim is consequential. Each event receives a stable ID, timestamp, actor role, affected artifact hash, reason code, and result. The log is append-only: a later reversal adds an event instead of rewriting history. A correction can therefore be reproduced as a state transition rather than reconstructed from memory.

Be precise about downstream notification. A legitimate correction request identifies a factual discrepancy and supplies verifiable evidence. It does not instruct an independent publisher to adopt promotional language. If sponsorship or a commercial relation exists, disclose it. If the team lacks authority to contact or update a channel, record that limit. These safeguards keep correction separate from covert distribution.

## Chapter 9

LAB-L05 gives W10 a concrete intervention and rollback checkpoint. It is deliberately local, synthetic, offline, and deterministic.

The fixture contains three locked claims and three fixed source identities. The control and treatment HTML differ only through their stylesheets. The declared factor is `evidence_layout_proximity`: the treatment changes how closely visible evidence sits to the claim. The hypothesized stage is human comprehension. No external endpoint, user traffic, crawler, or model is involved.

The intervention card records authorization, expected later test, nonclaims, confounds, stop conditions, and rollback mapping. The synthetic experimenter differs from the checklist reviewer. The six factual-equivalence checks must pass. If claim wording, source attribution, keyboard behavior, or rollback hashes fail, the intervention stops.

Run the audit with the control page, treatment page, locked claims, intervention card, and checklist, writing outputs to a fresh temporary directory. The expected line is `L05 PASS: 3 locked claims, 1 changed factor, 0 error(s)`. Do not stop at that line. Inspect the intervention audit, normalized claims, before/after diff, equivalence review, rollback manifest, and run manifest.

The rollback manifest maps the treatment HTML and CSS pair to the control pair. A real rollback test verifies exact hashes, restores the control artifacts, and reruns validation. Keeping an old file somewhere is not enough. Rollback needs a target, an authorized operation, and a post-restoration test.

What does the pass establish? It establishes the declared local structural equality and one-factor scope under the frozen fixture. What does it not establish? It does not measure comprehension improvement. It does not establish crawler behavior, retrieval, answer use, citation, ranking, traffic, business outcome, production behavior, or cross-engine generalization. It is not pilot evidence and not accessibility conformance.

The connection to source ecosystems is governance. An intervention should preserve factual identity, record the changed factor, use an authorized surface, declare its intended stage, set stop conditions, and be reversible. If the intended mechanism is easier human verification, we do not relabel it as a ranking intervention.

There is a useful parallel between the lab and the source graph. LAB-L05 freezes claim wording and source identity while changing layout proximity. The ecosystem case freezes identity, scoped claim, source cards, and clustering rules while evaluating support. In both, a declared invariant limits interpretation. If an invariant changes, the comparison no longer answers the same question. This is why a production edit log should identify preserved as well as changed fields.

Accessibility belongs inside that governance. Evidence moved closer visually can still become harder to reach by keyboard or screen reader if focus order or semantic structure changes. The LAB-L05 stop conditions include keyboard regression, while the package notes that automated equality is not a human accessibility audit. A reversible intervention therefore needs factual, attribution, technical, and accessibility checks before release.

## Chapter 10

We will finish by making the whole decision reproducible without network access.

The frozen package needs an entity dossier, atomic claim table, source cards, provenance edge table, origin-cluster assignments, conflict ledger, correction log, file hashes, rule version, and evidence-boundary statement. A deterministic validator should reject missing edge endpoints, duplicate stable IDs, unknown edge types, a current claim without an effective date, and an origin count inconsistent with the dependency graph. Identical inputs and rules should emit the same bounded conclusion.

Here is that conclusion one final time. Within the supplied synthetic records, `ENT-NBDC-001` resolves to Northbridge Data Cooperative Ltd. in `EX-NORTH` through registry key `EXN-88421`; `NDC` alone is insufficient. As of 20 July 2026, `CLM-RETENTION-001` has one current owned primary record and one applicable independent audit origin supporting 90 days for the standard plan. The syndicated and aggregator pages add distribution but no independent corroboration. The 60-day PDF is superseded. The 30-day comparison is a legacy-plan and time mismatch. Refresh is required when an entity, plan, source, conflict, or review trigger fires.

PAPER-14 remains a metadata-limited, catalog-only context prompt rather than substantively absorbed evidence. PAPER-35 remains a study-specific preprint route. PAPER-01 remains quarantined and audit-only. PLAT-13 remains vocabulary semantics only. LAB-L05 remains local equivalence and rollback evidence only. These boundaries are part of the conclusion, not optional footnotes.

Use this exit test. Can a second reviewer resolve the entity from recorded keys? Is the claim atomic, dated, and scoped? Is authority evaluated per claim edge? Are dependent pages clustered? Are conflicts preserved rather than deleted? Is every correction authorized, dated, and assigned? Are nonresponsive downstream nodes visible? Is rollback exact and tested? Does the final prose separate page count, origin count, and independent corroboration count?

If any answer is no, the source ecosystem is not ready for publication. If every answer is yes, the result is still bounded. A well-governed evidence graph does not reveal a hidden retrieval system or guarantee visibility. It makes the evidentiary basis, dependence, uncertainty, correction state, and next review explicit. That is the standard W10 asks us to meet.

Before closing, distinguish three verbs. To **represent** is to encode an entity, claim, or relation, perhaps with PLAT-13 vocabulary semantics. To **support** is to connect a source to a scoped claim through an applicable evidentiary edge. To **corroborate** is to add materially independent applicable support under a declared rule. None of these verbs means that an external system retrieved or cited the page. Keeping those verbs separate is the final defense against overclaim.

The deliverable is ready only when its reader can trace every public sentence to a governed claim and every governed claim to its sources, dependencies, conflicts, dates, and owners. That trace does not make uncertainty disappear. It makes uncertainty reviewable and correction possible.