> For the complete documentation index, see [llms.txt](https://docs.cortex.aic.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.cortex.aic.io/sources-and-evidence/source-lineage.md).

# Source Lineage

**Open this area:** In the CORTEX menu, select **Sources & evidence → Evidence memory**.

Trace exact reporting chains, distinguish repeated reporting from independent observations, and deliberately review corrected provenance.

**Who uses this:** Collector, Analyst, Reviewer.\
**Open in CORTEX:** `/memory`.\
**In-app guide:** `/help?topic=source-lineage`.

**Expected result:** Attributable human reasoning and its limitations inform confidence without changing original material or earlier judgements.

## Before you start

* Use the correct current organisation and workspace. Source and Memory reads are required for exact reports and all retained linked origins.
* Source authors propose origins. Independently appointed reviewers need the separate reporting-origin review permission; they cannot review their own proposal.
* Analysis authors assess independence and respond to corrections. Reading or confirming an origin does not itself assess analytical confidence.

## What to do

{% stepper %}
{% step %}

### Open the exact original context

In Memory, choose a retained extraction and open **Review reporting origins**.

Follow its pinned source revision and original hash. Source/material versions are retained separately from origin assertions; no upload or extraction bytes are rewritten.
{% endstep %}

{% step %}

### Propose an evidenced origin

Choose unknown, directly observed, reported by, derived from or quotes.

Unknown/direct observation has no parent report; other relationships use 1–8 exact upstream report versions selected through the shared search.

Record:

* Evidence
* Certainty
* Limitations, including deception/bias
* Why this proposal or correction is needed
  {% endstep %}

{% step %}

### Review independently

The reviewer checks the current proposal and exact evidence, then confirms or rejects with rationale.

Only confirmed assertions affect the reviewed view. Rejection leaves the prior confirmed assertion effective; every proposal and decision remains attributable in history.
{% endstep %}

{% step %}

### Assess reporting independence

From analytical detail, open **reporting independence and corrections**.

Compare 2–8 distinct exact report versions already on the evidence board. Inspect their bounded chains, select supporting and contradictory evidence, and explain:

* The human conclusion
* Unknown provenance
* Possible deception
* Source bias

Separate document identifiers are not independent corroboration.
{% endstep %}

{% step %}

### Explain the confidence effect

Retain the assessment, then amend the judgement's confidence basis to cite up to four retained assessments and explain the effect on confidence.

The exact cited evidence must remain on the board.

Saving a new assessment or correction response requires fresh peer review; its author cannot independently review their own work.
{% endstep %}

{% step %}

### Respond to reviewed corrections

Analytical detail shows an in-app notice for currently visible outstanding corrections.

The correction inbox identifies the exact analytical revision, affected evidence and reviewed origin revisions. Record the effect and deliberate next action, then retain a new independence assessment and amend confidence reasoning where needed.

Fresh sign-off rejects changed origin snapshots. Earlier judgements and released bytes stay historical.
{% endstep %}
{% endstepper %}

## Check the result

* Do apparently separate reports share an actual reporting source?
* Is missing provenance still visible as a limitation?
* Have contrary evidence, deception and source bias affected the human reasoning?
* Does each confidence citation retain its exact original evidence?
* Have reviewed corrections received an explicit analytical response?

## If you get stuck

<details>

<summary>Why can I not claim independent corroboration?</summary>

Unknown provenance, shared reporting sources or circular chains prevent an independent-observation conclusion. Explain the actual uncertainty, shared origin or possible circularity and assess its confidence effect. Different document IDs alone do not establish independence.

</details>

<details>

<summary>Why did an origin proposal fail?</summary>

Check exact retained versions, current source/Memory access, required evidence and the permitted number of parents. Self-links and candidate cycles are rejected. A stale version requires deliberate reload and reconciliation; draft text is retained after a failure.

</details>

<details>

<summary>Why does a confidence citation fail after editing evidence?</summary>

A retained assessment is tied to exact evidence in its input revision. Restore the same evidence or remove the citation, save the new analytical inputs and assess them again. Old assessment evidence cannot be relabelled onto changed inputs.

</details>

<details>

<summary>What happens when history reaches a limit?</summary>

Origins retain up to 128 revisions and 4 MiB; an analysis retains up to 16 independence assessments and 64 correction decisions within its 8 MiB history limit. Capacity failure preserves earlier history. Follow-on analytical work must keep explicit references; nothing is silently deleted.

</details>

<details>

<summary>Does reviewing an impact revise the product?</summary>

No. It retains an accountable response and requests fresh analytical review. Amend the affected judgement, repeat assurance where its basis changed, and follow normal product/dissemination controls. Existing exports and released packages are never silently regenerated.

</details>

## Responsibility and limits

* A bounded view contains up to eight selected reports, 32 visible nodes and eight upstream levels. Missing, inaccessible or out-of-bound chains share a generic unknown presentation; hidden nodes, topology, identifiers and counts are not exposed.
* Current source, material, organisation, workspace and handling access is checked again for retained snapshots. A previously visible assessment can become unavailable if its sources become protected.
* Correction impacts are derived from durable reviewed origin revisions with stable keys for the exact analytical revision and evidence. They are shown on opening or refreshing analytical detail/inbox; this workflow sends no email or background push notification.

## Reference and scope

The reviewed JDP, PHIA and intelligence-assessment requirements inform this workflow. Reviewed origins describe evidence-backed assertions; they do not certify truth, grant investigative authority or automatically establish corroboration.

## Related workflows

* [Find evidence and select exact retained passages](broken://pages/e6eab51e146c4db1e002b720557a5faea932e41c)
* [Plan collection and record observed results](file:///7397340/collection/collection.md)
* [Develop and independently review an assessment](broken://pages/79db57a6c6245feef3fb58e4fe6d4a507b2a74a0)
* [Explain confidence, changed judgements and dissent](broken://pages/b5f23381c9372db8261edb88339424566cabc7f2)
* [Plan analytical assurance and obtain separate sign-off](broken://pages/1ff25928b8eecd677eef29b9b4c97fb68c33a32a)
* [Author an intelligence Product and inspect its exports](file:///7397340/product/products.md)