# MacroVision v93: transaction-data access and recruitment feasibility
29 October 2026. Public-documentation comparison only. No contact, registration, data application, payment, recruitment or microdata processing performed.

## Decision
Prioritize a bounded access-and-coverage assessment of existing linked data before commissioning a new panel. UASFin is the first documentation candidate for survey/transaction linkage; JPMorganChase Institute is a parallel institutional collaboration candidate; the UK Exact.One source is a separate-country alternative. None is approved for the fixed v91 estimand.

## Evidence and access comparison
| Route | What public evidence establishes | What remains unknown | Practical next gate |
|---|---|---|---|
| USC UASFin | Official participant documentation describes balances and dated transaction amounts/descriptions linked through financial accounts. General UAS access requires a legitimate research purpose and agreements. The panel cohort paper reports approximately 1,000 historical financial-linkage consenters; this is not a current analytic sample. | Transaction-file access tier, current linked sample, dates, exact involuntary-loss questions, food detail, household account completeness, cost and approval timing. General survey access does not establish transaction access. | Obtain a provider-confirmed file inventory and aggregate eligibility counts without importing participant data. |
| JPMorganChase Institute | Institute research examines spending after job loss and liquidity differences. Its academic collaboration model includes research fellows. Privacy protocols restrict research data use and export within secure systems. | Eligibility for collaboration, project acceptance, event definition, household coverage, pure food detail, costs and schedule. No open microdata entitlement was established. | Establish a suitable institutional research partner and confirm approved analysis/output route. |
| Exact.One / ClearScore, UK | IFS's January 2026 primary report uses dated transactions from this source and a final sample of 10,760 users. The report distinguishes users from households and notes incomplete categorization/coverage. | Our licensing, available vintage, balances, exact separation reasons, household coverage, costs and permission for new research. Published study sample is not our eligible N. | Verify research access and whether the fixed event/resource/basket definitions can be implemented. UK study needs a separate transportability assessment. |
| Newly commissioned monthly panel | V91 specifies measurements and a hypothetical precision grid; no recruitment vendor or quote has been verified. | Sampling frame, linked-account consent, event incidence, all-in cost, timeline, retention and design variance. | Seek feasibility counts and a quote only after existing-data routes fail; do not launch collection now. |

This ranking is an inference from published capabilities and barriers, not a provider endorsement or a claim of availability. Prices and approval times are unknown; do not invent cost estimates or promise a free linked dataset.

## Access constraints specific to this workflow
UAS's current data policy forbids supplying microdata to LLMs or AI tools and forbids AI editor indexing of microdata directories. Its FAQ allows code generation without microdata and aggregate analytic outputs without microdata. Any eventual analysis must use an authorized researcher and permitted environment, with data kept outside AI-accessible directories. Review provider terms for the actual transaction product before access. No UAS microdata were accessed here.

JPMorganChase protocols describe approved-purpose analysis, confidentiality obligations and secure systems preventing data export. An accepted collaboration is needed; published findings do not grant access.

## Common measurement stop rules
A dated bank debit is not automatically food consumption; supermarket baskets mix food and non-food, cash and benefit accounts may be missing, transfers and refunds can be misclassified, and partner accounts may be unlinked. Missing payroll or a first unemployment-benefit payment is not proof of an involuntary separation date. A backfilled balance is not a prospectively recorded baseline.

Require a provider-confirmed event reason/date, strictly pre-event resources, the three baseline and three follow-up months, defensible household identity/coverage, and a complete primary basket before claiming the unchanged v91 design. Otherwise record failure or commit a separately named exploratory proxy design before contrasts. No source's existing published coefficients enter MacroVision calibration.

## Recruitment arithmetic and timing
The v91 illustrative case (change-log-spending SD .35; event share .10; low-buffer share .50; DEFF 1.5) requires 25,640 complete households and 34,187 recruits at .75 retention. These are hypothetical assumptions, not an observed event rate or provider capacity. The historical approximately 1,000 UAS financial consenters cannot be substituted for that target; present-day coverage and precision still require confirmation.

A provider's total users, transactions or repeated interviews cannot be counted as independent complete households. With a new prospective cohort, baseline three months plus event month plus follow-up three months requires at least seven indexed monthly observations for each qualifying event; recruitment and waiting for sufficient events can take longer. No completion date can be asserted yet.

## Ready-to-use provider feasibility questions (not sent)
1. Which exact files, vintages, codebooks and authorized access route include transaction dates, food components, account balances and survey employment reasons/dates?
2. At what unit do IDs refer to a person, family, account or household? Can partner accounts and changes in household membership be distinguished?
3. Can you supply aggregate counts only for employed ages 25–60, three complete pre-months, verified involuntary separation and three post-months, partitioned by baseline buffer? Define missingness/retention denominator.
4. Are balances contemporaneously recorded or retrospectively backfilled? Is cash, benefit-funded food and unlinked-account coverage known?
5. Are sample design/weights and variance inputs available for the linked consented cohort, rather than the full panel?
6. What eligibility, agreements, institutional sponsorship, approved environment and output review are required? What are total access costs and expected process times?
7. What published or documented food classification validation, duplicate/refund treatment and linkage-error evidence exists?

## Decision after provider response
Proceed to source-specific blinded membership audit only if access is permitted and field definitions are adequate. Freeze eligible counts, variance assumptions and power simulation before outcome contrasts. If no route meets the precision target, report infeasibility and scope a funded prospective project; do not relax the target to obtain a result.

Immediate independent next action: inspect public UAS monthly employment and financial-linkage metadata for exact separation definitions and overlapping fieldwork periods. Provider contact or submitting a data agreement remains a distinct person-directed action requiring explicit user instruction.

## Meaning for the thesis
Existing transaction-based research offers relevant external evidence about spending, liquidity and benefits, but it is not our replication and does not validate worldwide automation, price declines or fiscal sustainability. Keep v90 inconclusive and causal/global calibration gates closed.

## Primary sources inspected
- UAS financial-consent instrument: https://uasdata.usc.edu/index.php?ssid=UAS192&type=2&vsid=consent
- UASFin participant help: https://uasfin.usc.edu/help/
- UAS access/AI policy: https://uasdata.usc.edu/index.php?r=eNpLtDK0qi62MrFSKkhMT1WyLrYyBLFLk3Iyk_WSS4sSS1JTlKxrAe9cXAyW
- UAS cohort paper (historical linkage count, not current membership): https://doi.org/10.1136/bmjopen-2024-088183 (indexed primary-paper text reviewed; direct DOI fetch unavailable)
- JPMorganChase research: https://www.jpmorganchase.com/institute/all-topics/labor-markets-and-workforce/spending-after-job-loss-from-the-great-recession-through-covid-19
- Institute collaboration: https://www.jpmorganchase.com/institute/news-events/academic-partnership-and-collaboration-for-impact
- Institute privacy protocols: https://www.jpmorganchase.com/about/governance/data-privacy-protocols
- IFS primary report: https://ifs.org.uk/publications/unemployment-benefits-and-household-spending-new-evidence-uk-bank-account-data

Validation: public primary source review and explicit known/unknown separation. No new calculator or empirical software change. Existing site CI must pass before merge; live report and research link must be verified after deployment.