# Reusable tracing method and reader checklist

This is an application of established source-tracing practice, informed by the [University of Cincinnati's tracing guide](https://guides.libraries.uc.edu/c.php?g=1128926&p=8352356) and its [credit to Mike Caulfield's work](https://guides.libraries.uc.edu/c.php?g=1128926&p=8348634). New Eruditions' original work is the case execution, preservation checks and reusable records.

## Eight-step tracing procedure

1. **Freeze one claim.** Write the exact amount and currency notation, unit, period, geography, category, estimate/forecast status and attributed producer. Capture a short passage and its location before chasing links. Do not use a title as a substitute for the claim sentence.
2. **Inventory the citation.** Record hyperlink destination and source credit separately. Decode a tracking URL's destination without assuming its endpoint resolves. Distinguish an about-page link from a report-specific link. Plain-text credits still count as evidence.
3. **Inspect the next document.** Confirm its body contains the claim, not just a title/search snippet. Record requested/resolved URLs, displayed dates, update date if actually known, retrieval timezone and access result. Separate Browser observation from extractor metadata.
4. **Discover missing links narrowly.** Search a distinctive phrase plus value, period and named publisher. Inspect the best matches. Similar prose and older dates guide discovery; they do not establish acquisition routes. Limit each edge to the statistic under review.
5. **Recover a moved origin.** Try relevant publisher search/migration records, a surviving exact document/PDF and public archives. For an archive, inspect the actual body and verify original URL plus capture timestamp. Log failed access and incorrectly scoped queries honestly. Stop bounded searches when the origin is inspectable or a useful unrecovered boundary is documented.
6. **Compare fields across nodes.** Mark each value, currency, unit, period, geography, scope, estimate/forecast status and qualifier as unchanged, documented change, not present or cannot check. Preserve omission; do not fill it from another account without labeling the source of that addition.
7. **Classify evidence and limits.** Separate verified page content, verified attribution, derived operations, inferred document routes and unresolved inputs. Record why each arrow exists. Original estimate publication and underlying data/model verification are separate outcomes.
8. **Choose the citation and stop.** Cite the inspectable origin/version for its estimate. If unrecovered, cite the account actually checked with explicit secondary attribution, or reject it for the intended assertion. State what new evidence would change the decision. Do not add links simply to make the claim look better supported.

## Reader checklist

- Have I frozen the exact number rather than a nearby quarter or forecast?
- Are the printed currency marker and unit retained?
- Do I know the year/quarter, geography and market category?
- Is this an estimate, reported observation or forecast?
- Does the citation lead to a specific document or merely a publisher home/about page?
- Have I inspected the full relevant passage, rather than a search or AI summary?
- Are requested URL, resolved destination and access result recorded separately?
- For an archive, did I verify original URL, capture time and actual claim body?
- Does every graph arrow have attribution/link evidence and a stated uncertainty?
- Did I check scope and qualifiers as well as the amount?
- Have I distinguished published-estimate recovery from validation of raw inputs/model?
- Does my citation say exactly what I inspected, including any unrecovered boundary?

## Blank templates and usage

Use [blank-lineage-ledger.csv](blank-lineage-ledger.csv) for one row per document/version and [blank-lineage-edges.csv](blank-lineage-edges.csv) for one row per relationship. Duplicate a node when versions differ; do not overwrite its date or claim silently. Use a stable claim ID across the trail. The [populated ledger](lineage-ledger.csv) demonstrates how to retain access limitations and unknowns.

The node template includes core document fields and additional evidence/access fields. Do not fill unknowns with guesses. Use `not established`, `not present in inspected content` and `cannot check` where appropriate; they express different limits. A blank cell is reserved for a not-yet-completed field. For edges, record publisher attribution separately from exact-document confirmation and actual acquisition route.

The directed graph may follow source → downstream, while the reader moves downstream → upstream. Label that convention. Supply a text/table equivalent, and never rely on color alone to convey uncertainty.