Earned wage access payroll interface study

Methodology

This brief triangulates the headline measure against official Philippine government, regulatory, development, and labor sources. It translates the evidence into an operating control and separates context from recommendations.

Key takeaways

• Freeze populations and definitions before reviewing outcomes.
• Separate facts, owner decisions, analysis, inference, and uncertainty.
• Use protected references and stop at authority or interpretation questions.

Freeze provider events and versioned inputs

Independently freeze provider, time, and payroll populations.

Preserve event identifier, worker token, event type, timestamps, amount, currency, status, earning-window reference, source-hours version, payroll period, offset reference, reversal link, and last update.

A current dashboard cannot supply historical states.

Include rejected, cancelled, reversed, pending, and duplicate events.

Define how payroll offsets without provider events and provider events after a worker leaves enter the denominator.

Hash extracts and document zones, cutoffs, file or API versions, and retry behavior.

The research layer needs tokens and amounts, not bank destinations, authentication factors, or full timecards.

Trace preliminary time to scheduled payroll

Preserve the source-hours version used for access and the authoritative version supplied to payroll.

Compare population, units, earning dates, approval state, excluded categories, and calculation time.

A later difference does not prove worker error or overpayment.

Bind each provider event to the declared payroll field, cancellation, reversal, or approved treatment.

Matching net pay is insufficient because unrelated items may offset.

Support may find absent references or arithmetic differences, but cannot decide disputed time, wage, deduction, recovery, tax, or consent.

Preserve the owner’s decision and follow the event to a defined final status.

Exercise boundary cases and retries

Test corrected timecards, deleted shifts, retroactive premiums, duplicate messages, partial reversals, multiple events, payroll transfers, leave, termination, final pay, negative net, off-cycle pay, and provider outage.

Predeclare expected system events and decision owners.

Verify that replaying a file or message cannot silently create a second effect.

Verify whether late updates to closed payroll reject, queue, or redirect visibly.

Test administrator and worker access removal.

A case passes only when evidence links provider, time, payroll, owner decision, and final status; aggregate agreement is not enough.

Unsupported scenarios remain declared scope limits rather than production improvisations.

Interpret interface evidence, not product legality

Report events by state, unmatched provider and payroll records, changed time versions, reversals, duplicate messages, unresolved age, and owner-disposition time without netting opposite amounts.

Keep provider assertions, system observations, owner conclusions, and inference separate.

The CFPB opinion supplies current consumer-finance context but does not decide a particular program; DOL and IRS sources are similarly bounded.

Programs differ in funding, fees, jurisdictions, configuration, and populations.

A buyer can use the study to assess whether a controlled support lane is observable.

It cannot establish product classification, wage compliance, correct tax, worker consent, or provider performance.

Trace a changed time record through reversal and payroll

Select an event tied to preliminary hours that later changed.

Preserve the source-hours version used by the provider, event amount and status, update message, reversal or adjustment reference, authoritative time record sent to payroll, scheduled payroll line, and final owner disposition.

Compare timestamps in their original zones and record any replay or retry.

The study should show whether the interface prevents a second effect, whether cancellation reaches the correct payroll period, and how an unresolved item appears to reviewers.

A provider balance or successful message is not a payroll result.

Ask a second reviewer to repeat the match from the protected references.

Where product configuration or policy determines the expected treatment, cite the authorized decision rather than inferring it from the observed amount.

The completed trace tests observability and handoff discipline without deciding product classification, wage treatment, fees, deductions, recovery, consent, tax, or final-pay requirements.

Preserve the provider and payroll event identifiers so the next scheduled run can test whether the open adjustment cleared once and only once.

Keep any unmatched status open for owner review.

Build three independent event populations

Freeze provider events, source time versions, and payroll outputs separately before joining them.

Provider data should include event token, worker token, type, amount, currency, status history, earning-window reference, timestamps, reversal links, and retry identifiers.

Time data should identify the version used for the event and the authoritative version later supplied to payroll.

Payroll data should identify the scheduled period, relevant line, and any approved offset or adjustment reference.

Include rejected, cancelled, reversed, pending, provider-only, and payroll-only events.

Document API or file versions, extraction times, time zones, inaccessible records, and exclusions.

This design prevents a current provider dashboard from rewriting earlier states and keeps the study focused on interface evidence instead of bank destinations or full timecards.

Classify paths through changes, reversals, and retries

Define expected links for an unchanged event, a corrected time record, a cancellation, partial reversal, duplicate message, late update, termination, and final-pay exception.

Reviewers should trace identifiers and amounts through each system without assuming a successful message produced a payroll result.

Mark eligibility or treatment questions as owner decisions, not analytical classifications.

Test whether retries are idempotent, whether late changes reject or queue visibly, and whether one adjustment can enter payroll twice.

A mismatch remains unresolved when source meaning or authorized expectation is unavailable.

Preserve observed provider state, payroll state, owner conclusion, and analyst inference as separate fields.

The protocol tests observability; it does not decide product classification, wages, deductions, fees, consent, recovery, tax, or final-pay requirements.

Present interface findings without a product verdict

Report counts by state, changed time versions, provider-only and payroll-only records, reversals, duplicate messages, unresolved age, and owner-disposition time.

Show event-level evidence for all exceptions and independently recheck a declared sample of matches.

Do not net opposite differences or treat a later clearing event as proof that the original handoff was timely.

Explain program configuration, population, and window so readers do not generalize beyond the study.

The CFPB, DOL, IRS, NIST, and privacy sources remain limited to their stated domains.

The buyer can use the results to decide whether a bounded support lane can compare events and escalate uncertainty.

Legal, payroll, HR, product, finance, and other qualified owners retain every consequential conclusion.

Keep a replayable interface evidence set

Retain extraction hashes, event and message identifiers, schema or file versions, source-hours version references, payroll-period definitions, state rules, reviewer notes, and retry outcomes.

Do not retain authentication factors, bank destinations, or full timecards in the research package.

Where a provider overwrites event history, record that limitation and the approved snapshot method used during the study.

A later reviewer should be able to replay joins and classifications from protected records and see why each unmatched item remained open or closed.

If product configuration changes, begin a separately labeled comparison period.

Reproducibility supports monitoring, but it does not turn interface evidence into a legal or payroll conclusion.

Record approved disposal of every temporary working extract.

FAQs