Chain-of-Custody or Lineage Check¶

Provenance protocol — instantiates Use-Time Source Attribution Calibration

Reconstructs an item's unbroken trail back to its origin — every handoff and transformation logged beside the content — so its source class is established rather than assumed when it is used.

A confident label on an item — "observed," "verified" — is only as good as the trail behind it, and trails break silently. Chain-of-Custody or Lineage Check reconstructs and maintains that trail: it walks an item link by link from the point of use back to its origin, recording every handoff and transformation, and keeps that custody record as a separate artefact travelling beside the content. Its defining commitment is to the integrity of the whole path — a single undocumented gap downgrades the item, no matter how authoritative its current tag looks. It establishes what class of source an item is so it can be weighted correctly at use; it is not a judgement of whether that source was any good, and not an audit of the downstream risk the item might carry.

Example¶

A security analyst is about to push a firewall block based on an indicator sitting in the shared threat store — an IP address tagged "observed malicious, high confidence." Before acting on it, the lineage check walks the trail: which feed carried it, which analyst enriched it, and — the load-bearing question — was it seen in our own telemetry or copied from a vendor list or inferred by a scoring model? The walk finds the "observed" tag actually traces to an unverified third-party import that a later enrichment step relabelled. The chain has a break. The indicator isn't discarded, but it is reweighted from first-hand observation down to hearsay, and the block is held for corroboration. The confident label survived only until someone reconstructed the path it rode in on.^[1]

How it works¶

The check is a link-by-link reconstruction from present use back to first origin, with each node labelled by how the item entered there — observed, inferred, copied, or generated. What distinguishes it from a single provenance lookup is that it verifies the continuity of the whole sequence, not just the existence of one source record: an intact chain of five weak links is treated differently from an authoritative origin reached through a broken one. The custody log is kept structurally apart from the content it describes, so an item's origin can be evaluated independently of what the item happens to say.

Tuning parameters¶

Trace depth — how far back to walk, from the immediate parent to the full origin. Deeper traces are more trustworthy but cost real reconstruction effort at each hop.
Break tolerance — whether one missing link voids the chain outright or merely downgrades confidence a notch. Strict voiding protects high-stakes uses; graded downgrade keeps usable items usable.
Record separation — how strictly the custody log is held apart from and linked to the content: a fully separate ledger versus inline provenance tags that travel with the text.
Re-trace cadence — a one-time reconstruction versus re-verification each time the item is reused, since links rot and get relabelled over time.

When it helps, and when it misleads¶

Its strength is exposing the item whose confident source class is unsupported — the "observed" that is really "copied," the "verified" whose verifier can't be named. That is precisely the failure the archetype targets, and only a walk of the whole path surfaces it.

Its limits are real. The reconstruction is expensive, and it is only ever as good as the logging it walks; unrecorded handoffs are invisible. And a chain can be perfectly intact yet still terminate at a bad origin — integrity is not truth. The classic misuse is to run it backwards: assembling a clean-looking custody record after the fact to make a shaky item appear well-sourced. The discipline that guards against this is to treat a complete chain as necessary but not sufficient, and to let a broken link downgrade the item honestly rather than papering over the gap.

How it implements the components¶

provenance_backlink_or_trace — it builds the actual link-by-link backlink from the point of use to the item's origin, node by node.
source_content_separation — it maintains the custody trail as a separate record travelling beside the content, so provenance can be read and weighted apart from the item's assertions.

It does not credit ideas borrowed unconsciously into your own store — that's Borrowed Idea Attribution Scan — nor score the confidence of the attribution (Source Attribution Confidence Rubric), nor supply the source-class taxonomy it labels links with (Source Attribution Training Set).

Instantiates: Use-Time Source Attribution Calibration — it establishes the provenance an item's use-time weight depends on.
Sibling mechanisms: Borrowed Idea Attribution Scan · Provenance Lookup Before Publication · Generated Content Disclosure Gate · Hallucination Intrusion Triage · Memory Source Probe · Observation Recheck or Replication · Reality Monitoring Checklist · Source Attribution Confidence Rubric · Source Attribution Training Set · Source Confusion Matrix Review · Source-Label Preserving Summary Template

Notes¶

A trustworthy chain tells you where an item came from and how to weight it — not whether the origin itself was sound. Establishing source class is a separate job from vetting the source's quality or the liability it carries downstream; conflating the two lets a well-documented trail launder a bad origin.

References¶

[1] Chain of custody — the forensic and legal practice of documenting every transfer of an item of evidence so its integrity from origin to use can be established. A single undocumented gap can render otherwise-genuine evidence inadmissible, which is the same logic applied here to a shared information store. ↩