# MacroVision v92: CE documentation feasibility screen
29 October 2026. Documentation only. No microdata downloaded, sample counts or outcome estimates produced. Follow-up to v91.

## Decision
Do not approve the inspected CE source for the fixed v91 monthly food-expenditure/job-loss test. Required event and measurement gates remain unproven. This is a scoped negative feasibility decision, not proof that all CE vintages or restricted files lack useful information.

## Verified documentation
BLS April 2024–March 2025 Interview questionnaire:
- p374: FDBAWYWX measures usual weekly restaurant/away food and drinks, including alcohol and delivery. GREXPWX measures usual weekly groceries including non-food products and SNAP/WIC purchases.
- pp670–671: ANYWORK asks about wages since the reference month; INCWEEKQ asks weeks worked in the preceding year. These are not dated involuntary separations.
- p695: LIQUDYRX refers to checking/savings/money-market/CD balances one year before interview; it is retrospective.
These are questionnaire identifiers, not a verified public-file crosswalk.

BLS microdata guide:
- MTBI has monthly expenditure records, but file periodicity does not certify the original timing of every expense category.
- Interview panels cover up to four quarters at an address; household replacement and attrition complicate linkage.
- Interviews are separately weighted. Those weights are not automatically valid for a selected event-window cohort.
- Diary observation lasts two weeks, insufficient alone for the six-month pre/post design.

## Gate assessment
| Requirement | Scoped finding | Decision |
|---|---|---|
| Exact involuntary separation reason and month | Not established in inspected work questions | Closed; annual work totals cannot substitute |
| Dated monthly primary food basket | Inspected usual-week measures do not establish monthly realized spending | Closed; no monthly interpolation |
| Pure food measure | Grocery question includes non-food; away category includes alcohol | Require component mapping before reuse |
| Assets measured at end of month -1 | One-year-ago retrospective reference is a different date and reporting process | Closed for unchanged v91 |
| Household continuity | Address panel can change residents | Unverified |
| Valid longitudinal variance | Separate interview weighting does not settle event-sample variance | Unverified |
| Power / precision | No eligible sample or empirical variance audited | Not assessed |

A failed keyword search is not evidence of absence. No exact job-loss field was verified; the claim here is failure to establish the required combination. Older CE waves, other questionnaires and restricted products were not exhaustively reviewed.

Do not merge Diary and Interview as if they followed the same households without a documented linkage product. Do not divide usual-week or annual values into months and claim observed event-time changes. Do not reclassify less than 52 weeks worked as involuntary job loss.

## Consequence for the thesis
PSID and CE screens have not supplied the fixed monthly test. Keep V90 inconclusive and macro calibration closed. The next priority is practical feasibility of a prospective monthly household panel or an existing consented provider dataset with the equivalent fields. Historical associations may be useful separately, with a new specification committed before outcomes; they do not become independent causal validation.

## Prospective/provider feasibility brief — proposed requirements
This is a design brief, not a claim that a provider or participants are available:
1. Enroll employed households before separation and retain three complete baseline months. Event-triggered recruitment after job loss cannot reconstruct prospectively recorded assets.
2. Follow monthly for at least three months after each eligible event. Record date, reason, temporary-layoff status, index-person identity, partner employment and household changes.
3. Measure food at home/delivery/away from dated transaction records or monthly itemized diaries. Resolve mixed shopping baskets, refunds, cash purchases, missing accounts and benefit-funded food. Bank debits alone are not total household consumption.
4. Record baseline cash and accessible account balances and monthly total spending; keep asset reference date separate from collection date. Identify debt service separately.
5. Define sampling frame, event incidence, retention, provider coverage and variance design. Use the v91 planning grid only as a hypothetical starting range; household counts are not known.
6. Obtain documented pricing/access terms and field coverage before selecting a provider or launching recruitment. No spending, outreach, account creation or collection is performed by this step.
7. Lock source, decoder, sample allocation and survey-based power simulation before outcome comparisons. Stop if coverage or recruitment cannot support the target; report infeasibility without tuning thresholds.

Next deliverable: a provider/data-access and recruitment feasibility comparison with explicit unknowns, before commissioning collection. An observational panel still does not identify the causal effect of cash transfers, automation or worldwide unemployment.

## Official references inspected
- https://www.bls.gov/cex/research_papers/pdf/2024-2025-ceqquestionnaire.pdf (printed pages 374, 670–672, 695)
- https://www.bls.gov/cex/pumd-getting-started-guide.htm (file structure, longitudinal limitations and weights)
- https://www.bls.gov/opub/hom/cex/ (Interview/Diary collection scope)

Documentation reviewed live; exact file hashes/public-variable mapping remain prerequisites to empirical implementation. Only public aggregate/documentation content is included.