# W10 Worked Case — The Northbridge Source Ecosystem

## Case status

This is a fictional, frozen, offline teaching case. `Northbridge Data Cooperative Ltd.`, its jurisdiction, registry, documents, audit, publisher branches, and correction events are authored for W10. They do not represent a real entity or platform. The case demonstrates entity resolution, claim-relative authority, provenance clustering, conflict preservation, authorized correction, deterministic refresh, and rollback. This case does not evaluate a live platform. It does not evaluate retrieval, answer use, citation, ranking, traffic, reputation change, or business outcome.

The human-facing reading boundary remains fixed. PAPER-14 is Notes catalog-only in the current absorption state; the latest local export lacks its URL, abstract, and access date. It is used only as a metadata-limited context prompt, not as substantively absorbed evidence. PAPER-35 is a preprint route for evidence-ecosystem questions, with its architecture and findings kept study-specific. PAPER-01 is quarantined and contributes only a time-provenance audit lesson. PLAT-13 describes vocabulary semantics only. LAB-L05 is a deterministic local intervention, not an external system experiment.

## 1. Frozen question and decision rule

Question: within the supplied records, is the claim that the standard service retains audit exports for 90 days corroborated beyond one owned page as of 20 July 2026?

The instructional rule requires: a resolved entity; one atomic, temporally scoped claim; applicable primary support; at least one materially independent and applicable origin; preserved and classified conflicts; and a dated conclusion. This rule is not a universal threshold. Higher-risk claims can require additional independent origins, direct verification, regulatory evidence, or specialist review.

The output must report pages, material origin clusters, and applicable independent corroborating origins separately. It must not use a majority vote.

## 2. Entity resolution

The supplied candidate cards contain three organizations abbreviated `NDC`. Name similarity therefore fails. The accepted record is:

| Field | Frozen value |
|---|---|
| Stable project ID | `ENT-NBDC-001` |
| Canonical name | Northbridge Data Cooperative Ltd. |
| Entity type | Organization |
| Jurisdiction | `EX-NORTH` |
| Registry key | `EXN-88421` |
| Accepted current alias | Northbridge Data |
| Accepted historical alias | Northbridge Cooperative, through 30 June 2026 |
| Rejected stand-alone resolver | `NDC`, ambiguous |
| Resolution status | Resolved inside supplied fixture |
| Reviewed | 20 July 2026 |

The official registry card `SRC-REG-01` connects the canonical name, jurisdiction, and registry key. It is primary for those registry facts. It says nothing about export retention, so it cannot support the retention predicate. This distinction prevents identity authority from leaking into unrelated claims.

## 3. Atomic claim record

The loose statement “Northbridge stores exports securely for 90 days” is split. Only retention enters this case:

| Field | Frozen value |
|---|---|
| Claim ID | `CLM-RETENTION-001` |
| Entity ID | `ENT-NBDC-001` |
| Predicate | Retention duration for audit exports |
| Value | 90 days |
| Product/plan | Standard service |
| Jurisdiction | `EX-NORTH` |
| Effective from | 15 July 2026 |
| Effective to | Open at freeze date |
| Status | Current under review |
| Not evaluated | Security, compliance, deletion guarantees, availability SLA |

Publication, observation, retrieval, effective, review, and superseded dates remain distinct. A new access date cannot make a legacy claim current.

## 4. Source-card classification

| Source ID | Supplied content | Role for this claim | Origin cluster | Disposition |
|---|---|---|---|---|
| `SRC-REG-01` | Official legal-name and jurisdiction record | Primary for identity; inapplicable to retention | `ORG-REG-01` | Retain as entity evidence |
| `SRC-OWN-01` | Current standard-plan documentation, reviewed 15 July 2026, states 90 days | Primary and owned; directly applicable | `ORG-OWN-01` | Current support |
| `SRC-OWN-02` | Owned PDF dated 1 November 2025, states 60 days | Primary and owned for an earlier state | `ORG-OWN-01` | Superseded, never deleted |
| `SRC-IND-01` | Procurement audit dated 20 July 2026, tests availability through day 90 | Independent direct observation; applicable | `ORG-INDEPENDENT-01` | Independent corroboration |
| `SRC-SYN-01` | News page repeats current press-release wording | Externally published but materially dependent | `ORG-OWN-01` | Distribution only |
| `SRC-AGG-01` | Aggregator extracts the syndicated statement | Dependent transformation | `ORG-OWN-01` | Distribution only |
| `SRC-CON-01` | Independent comparison dated February 2025, says 30 days for legacy plan | Independent but not comparable to current claim | `ORG-LEGACY-01` | Scope/time mismatch retained |

This table shows why owned is not automatically weak and external is not automatically independent. `SRC-OWN-01` is the proper primary record for the entity’s declared current specification. It cannot establish a comparative reputation claim. `SRC-SYN-01` has a different domain, yet its fact source is the owned release. `SRC-REG-01` is independent and official, but predicate applicability fails.

## 5. Provenance graph reproduction

The authoritative text-equivalent edge table is:

| From | Edge | To | Scope/status |
|---|---|---|---|
| `SRC-REG-01` | identifies | `ENT-NBDC-001` | legal name, jurisdiction, registry key |
| `SRC-OWN-01` | asserts | `CLM-RETENTION-001` | current standard plan |
| `SRC-IND-01` | observes-support | `CLM-RETENTION-001` | tests through day 90 |
| `SRC-OWN-02` | asserts-prior | `CLM-RETENTION-001` | prior 60-day state, superseded |
| `SRC-SYN-01` | syndicates | `SRC-OWN-01` | same material origin |
| `SRC-AGG-01` | transforms | `SRC-SYN-01` | same material origin |
| `SRC-CON-01` | qualifies-conflict | `CLM-RETENTION-001` | legacy plan, February 2025 |
| `COR-001` | supersedes | `SRC-OWN-02` | authorized current documentation |

Sort by target, then origin cluster, then source ID. Validate that every edge endpoint exists and every edge type belongs to the declared vocabulary. The current support graph contains two applicable retention origins: `ORG-OWN-01` and `ORG-INDEPENDENT-01`. Only the latter is materially independent of the entity’s claim origin. The registry and legacy comparison origins remain in the package but fail retention applicability.

## 6. Page count versus origin count

Four supplied publication artifacts display a 90-day current value: current owned documentation, the independent audit, the syndicated page, and the aggregator. That is a displaying-page count of four.

After following material reliance, those four artifacts reduce to two relevant origin clusters: the owned origin, which includes the syndicated and aggregator branches, and the independent audit origin. The number of applicable independent corroborating origins is one.

The correct expression is not “four independent sources agree.” It is: “One current owned primary record and one applicable independent audit origin converge; two additional pages depend on the owned origin.” Distribution reach and independent corroboration are both recorded, but they are never substituted.

## 7. Conflict resolution without deletion

`SRC-OWN-02` states 60 days. Its date and plan history place it before the current effective interval. It is marked `superseded` with a pointer to `SRC-OWN-01`, not removed.

`SRC-CON-01` states 30 days. It describes the legacy plan in February 2025, so it is a time and plan mismatch. The independent status of its publisher does not make it applicable to the July 2026 standard-plan claim. Preserve it because a downstream user may encounter it and because its existence should trigger a clarification of plan and effective date.

If a credible source observed 30 days on the current standard plan after 15 July 2026, the state would be `unresolved exact-scope contradiction`. The public formulation should narrow or freeze until investigation closes. Source prestige and page count would not settle the conflict.

## 8. Authorized correction and refresh loop

Assume the owned current page omitted its effective date, causing the syndicated page to look inconsistent with the old PDF. The authorized workflow is:

1. Preserve hashes and snapshots of current and old records.
2. Verify `ENT-NBDC-001`, the standard-plan scope, and the change approval.
3. Assign the product-documentation owner and a separate evidence reviewer.
4. Add the effective date and link a visible change note on the owned page.
5. Record old and new representations, reason, approver, date, and rollback target in `COR-001`.
6. Contact the syndicated publisher through its real correction channel with the exact affected sentence and current primary record.
7. Mark the publisher `notified`; change to `corrected` only after confirmation or observed revision.
8. Mark the aggregator separately. A scheduled refresh, nonresponse, or decline remains visible.
9. Rebuild the frozen graph and require reviewer approval.

The team must not invent a third-party article, conceal sponsorship, pay for coverage represented as independent, or create mass copies. Those actions manufacture apparent independence and stop the exercise.

## 9. Update triggers and deterministic reproduction

The dossier reopens on a legal-name, jurisdiction, registry, ownership, or alias-collision event. The claim reopens on a plan revision, effective-date change, primary-source correction, credible new contradiction, expired review, or missed deadline. A broken source link opens an accessibility/provenance task but does not automatically invalidate the historical record.

A deterministic reproduction bundle contains the dossier, claim table, source cards, edge table, cluster assignments, conflict ledger, correction log, and SHA-256 hashes. The build rejects missing endpoints, duplicate stable IDs, undeclared edge types, undated current claims, and an independent-origin count inconsistent with cluster assignments. Given identical inputs and rules, the bounded conclusion must match byte for byte.

## 10. LAB-L05 controlled intervention reproduction

From the LAB-L05 directory, run:

```bash
python3 scripts/audit_intervention.py \
  --control data/control.html \
  --treatment data/treatment.html \
  --claims data/locked_claims.json \
  --card data/intervention_card.json \
  --checklist data/factual_equivalence.csv \
  --output /tmp/geo-l05
```

Expected terminal result: `L05 PASS: 3 locked claims, 1 changed factor, 0 error(s)`.

Inspect `intervention_audit.json`, `normalized_claims.json`, `before_after.diff`, `equivalence_review.csv`, `rollback_manifest.json`, and `run_manifest.json`. The changed factor must be `evidence_layout_proximity`; the environment is a course-owned synthetic local fixture; three claim IDs and three source identities remain fixed; the six checklist rows pass; and the synthetic reviewer differs from the experimenter.

The pass establishes structural equality and one-factor scope only. It does not establish improved comprehension, crawler or retriever response, answer use, citation, ranking, traffic, business outcome, production behavior, or cross-engine generalization.

## 11. Rollback test

The rollback manifest maps the treatment HTML/CSS pair to the control HTML/CSS pair. Before release, verify every hash, restore into a fresh directory, rerun the audit, and compare the normalized claims and source identities. Stop on factual, attribution, accessibility, or hash difference. A rollback promise without an exact artifact target and verification test is incomplete.

For the source ecosystem, rollback restores the prior authorized owned representation while preserving the correction event and reason. It must not erase the fact that a correction was attempted. The review team then decides whether public wording must remain frozen.

## 12. Bounded conclusion and self-check

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. Syndicated and aggregator pages add distribution but no independent corroboration. The 60-day PDF is superseded, and the 30-day comparison is a preserved legacy-plan mismatch.

This conclusion is not about a real entity, a hidden source-selection mechanism, or a platform effect. Refresh it on a plan, entity, source-correction, contradiction, or review trigger.

Self-check:

- Can another reviewer resolve the entity without name familiarity?
- Is the claim atomic and scoped by time, plan, and jurisdiction?
- Is authority evaluated per claim edge?
- Are dependent publications assigned to one origin cluster?
- Are stale and conflicting records preserved with reasons?
- Is correction authorized, dated, reviewable, and reversible?
- Are PAPER-14, PAPER-35, PAPER-01, PLAT-13, and LAB-L05 held to their stated ceilings?
- Does the conclusion separate displaying pages, origins, and independent corroborators?

If any answer is no, do not publish the corroboration statement.