Skip to content

Summary-Claim Traceability Matrix

Artifact — instantiates Summary-Substance Alignment Audit

Decomposes the summary into atomic claims and gives each one a trace-link to the exact place in the body that supports it — or marks it unsupported.

A Summary-Claim Traceability Matrix decomposes the summary into its atomic claims and gives each one a trace-link to the exact location in the body that supports it — or marks it unsupported. It is the provenance grid for a summary: one row per summary claim, a pointer to the evidence in the substance that backs it, and an explicit "no anchor found" verdict for any claim that floats free. Its defining move is that every summary claim must point somewhere; a claim that cannot be traced is treated as a defect to be resolved, not a rounding error to be waved through.

Example

A private-equity deal team receives a one-page investment summary distilled from a 200-document data room. Before acting on it, an analyst builds a Summary-Claim Traceability Matrix: each summary bullet — "recurring revenue is 80% of total," "top customer is 12% of revenue," "no material litigation" — gets a row and a trace to the source document and page that backs it. Two bullets trace cleanly. "No material litigation" traces only to a management assertion, with no independent document behind it — flagged unsupported. "80% recurring" traces to a figure that, at source, was ≈74% and defined differently — flagged partial / definition mismatch. The tidy one-pager becomes a map of exactly which claims are load-bearing and which are unbacked, so diligence effort flows to the rows where the traces are weakest.

How it works

The matrix is exhaustive and provenance-first, in three moves: (1) atomize the summary into discrete, individually-checkable claims; (2) for each, locate and link the supporting span in the body; (3) record a support verdict — traced, partial, or unsupported. It does not judge amplification tone, author incentive, or reader inference; it answers one question per claim — does this have a home in the substance? — and makes the answer a checkable link. The unsupported and partial rows are the artifact's whole yield.

Tuning parameters

  • Claim granularity — how finely the summary is atomized. Finer decomposition catches compound claims that hide an unsupported half, but multiplies rows.
  • Trace strictness — what counts as "supported": a matching sentence, or matching data at the source. Loose strictness lets paraphrase pass; strict strictness flags more.
  • Coverage target — trace every claim, or only the load-bearing ones. Full coverage is thorough but costly; selective coverage risks skipping a quiet claim.
  • Link durability — whether traces are anchored so they survive the body being re-paginated or edited, or break on the next revision.
  • Refresh policy — one-shot, or rebuilt whenever the body changes so the traces cannot silently rot.

When it helps, and when it misleads

Its strength is making provenance explicit: it turns the vague question "is this summary backed?" into a grid where the unsupported rows are visible and countable. Its failure mode is subtle — a claim can trace cleanly to a body statement that is itself unsupported, so the matrix proves the summary is faithful to the body, not that the body is true. The classic misuse is building it to look rigorous while setting trace-strictness so loose that everything "traces." The discipline is to fix what counts as support before tracing, and to treat "traces to an unbacked body claim" as its own flagged verdict rather than a pass — the artifact is a traceability record, akin to a requirements traceability matrix, not a truth certificate.[1]

How it implements the components

Summary-Claim Traceability Matrix fills the decompose-and-trace side of the archetype:

  • summary_surface_claim_set — the atomized list of the summary's individual claims; the matrix's rows.
  • evidence_anchor_and_trace_link — the per-claim link to its supporting location in the body, or the explicit absence of one; the matrix's core.

It grids one summary against its own body; it does not weigh the author's incentive to amplify (Press-Release Claim Review), scan for dropped qualifiers (Qualifier-Drop Scan), or measure what real readers infer (Summary-Only Reader Test).

Notes

A clean trace certifies traceability, not correctness. The matrix's signature trap is the claim that traces perfectly to a body statement which is itself unbacked — it will pass a naïve read of the grid while remaining false. Carrying a distinct "traces to an unsupported body claim" verdict, rather than collapsing it into "supported," is what keeps the artifact honest and hands the truth question downstream where it belongs.

References

[1] A traceability matrix is the requirements-engineering artifact that maps each requirement to the design, code, or test that satisfies it; here the same structure maps each summary claim to the body evidence that supports it. It documents provenance — what links to what — not whether the underlying item is correct. ↩