# Verified methods layer — second-pass field audit v0.2

Date: 2026-09-26  
Scope: 19 full-text-verified core records; observability counts use 14 primary-study papers and exclude five reviews

## Audit method

The first pass coded one paper at a time. The second pass reviewed each observability field horizontally across all records. This is an internal consistency check, not independent double coding and not an inter-rater reliability exercise.

For every `yes`, the audit asked:

1. Does the evidence location identify analyzed material rather than background discussion?
2. Does the value satisfy the codebook's positive threshold?
3. Is a neighboring construct being mistaken for the target field?
4. Would `no`, `unclear`, or `not_applicable` more accurately preserve the evidence boundary?

## Correction made

S14 changed from `responsibility_disagreement_observed=yes` to `no`.

The paper distinguishes primary and secondary schedulers and documents unequal participation. That supports responsibility allocation, but the verified evidence does not establish conflicting perceptions of responsibility or an observed dispute over who is responsible. Allocation must not be substituted for disagreement.

No other value changed during this pass.

## Audited distributions

These counts describe papers in the purposive verified core, not households and not real-world prevalence.

| Field | Yes | No | Unclear | Interpretation boundary |
|---|---:|---:|---:|---|
| Multiple household perspectives | 3 | 10 | 1 | A single participant describing others is still `no`. |
| Direct behavior observed | 2 | 12 | 0 | Interviews, static artifacts, and participant diaries are not direct behavioral observation. |
| Natural artifact examined | 10 | 4 | 0 | Study questionnaires and research diaries do not count as household artifacts. |
| System trace examined | 0 | 14 | 0 | Static screenshots and calendar copies are not event logs. |
| Self-report only | 3 | 11 | 0 | Artifact or task-performance evidence moves a paper out of this category. |
| Cross-medium observed | 9 | 4 | 1 | Multiple digital applications constitute one medium class. |
| Handoff evidenced | 10 | 3 | 1 | Reported transfer may be evidenced without being directly observed behavior. |
| Access asymmetry evidenced | 7 | 4 | 3 | General individual differences do not count as access differences. |
| Version conflict evidenced | 1 | 6 | 7 | Multiple storage locations alone do not establish conflicting versions. |
| Responsibility disagreement evidenced | 0 | 6 | 8 | Unequal allocation alone does not establish disagreement. |

## Findings that are safe to carry into drafting

Subject to the purposive-corpus limitation, the verified layer supports the following methods findings:

- Artifact inspection is common in the verified core, but direct observation of behavior is uncommon.
- None of the 14 primary-study papers analyzed system event logs or comparable traces.
- Most primary studies do not retain multiple household-member perspectives, which constrains claims about agreement, negotiation, and shared responsibility.
- Handoffs and access differences are often represented through interviews and artifact walkthroughs, but version conflict is rarely operationalized directly.
- No primary-study paper in this verified core meets the strict threshold for measuring disagreement about responsibility, even when unequal responsibility allocation is visible.

These are statements about what the selected papers measure. They must not be rewritten as claims that version conflicts or responsibility disagreements are rare in households.

## Claims that remain prohibited

- Percentages of households experiencing any fragmentation mechanism.
- Claims that one method is universally superior.
- Claims that a phenomenon is absent because a study did not measure it.
- Causal claims from qualitative, survey, or correlational records.
- Generalization from the verified core to all household-information research without the 60-publication evidence-map layer.
- Product, feature, integration, or competitive implications.

## Targeted verification completed

S10's participation structure was resolved through the exact author-hosted paper. The study deliberately interviewed one self-identified primary information manager from each of ten families. Guided tours could include a spouse's archives, but other family members did not contribute study data. The record is therefore `single_member`, `multi_perspective_data=no`, with high confidence.