# Review packet audit Inspect one proposed result before treating human review as a safety or value claim. This sheet contains a completed hypothetical invoice packet, alternative review choices, a declared capacity model, and a blank working copy. No operator trial, customer saving, posting authority, or payment action is supplied. This is a paper audit, not an interactive review screen. Read the source and proposal, identify the error, then work through what each review choice must retain. The browser-storage and timer behavior discussed below comes from a local teaching screen, which is not included in this download. Inspect and test your own interface before relying on equivalent behavior. ## Start with these questions 1. What is the reviewer being asked to accept, and which actions remain outside that decision? 2. Which current source can the reviewer inspect to check the proposed amount? 3. Can the reviewer correct, reject, or escalate the proposal without losing the case or its history? 4. Who takes responsibility if the decision cannot be saved or there are more cases than reviewers can handle? Work through the completed invoice first. Then use the working copy to inspect your own packet, review choices, and queue. ## Completed hypothetical packet | Field | Record | | ----------------------------- | ------------------------------------------------------------------------------------------------------------------ | | Source | doc-a, revision `invoice-documents-2026-09-04-v1` | | Work boundary | Prepare a source-supported invoice draft or return the case to the owned manual/source path | | Excluded effects | Posting, payment, supplier contact, policy changes, identity management | | Source invoice | INV-101 | | Source currency / amount | USD / 120.50, equivalent to 12,050 cents | | Altered proposed amount | 12,000 cents, equivalent to USD 120.00 | | Difference | `12,050 − 12,000 = 50 cents`, or USD 0.50 | | Paper correction | Replace 12,000 with 12,050 cents; retain the prior proposal and source revision | | Paper disposition | Source-supported practice draft after correction; no live review decision occurred | | Practice reviewer / rationale | Practice reviewer R-1 / corrected the numeric amount against the source's explicit Total field | | Independent check | A second reader should be able to reconstruct the error and correction; this check has not been run with operators | Source text: ```text Invoice: INV-101 Currency: USD Total: 120.50 Supplier: Sample Parts. Two replacement valves. ``` Deliberately altered proposal: ```json { "action": "draft", "invoice_id": "INV-101", "currency": "USD", "amount_cents": 12000, "evidence_quote": "Total: 120.50", "reason": "Draft prepared for review" } ``` The proposed shape and exact quote can pass their checks while the amount remains wrong. The quote and source comparison perform different jobs. A source-checked checkbox records a review assertion; it does not independently prove that the reviewer compared the values or caught the error. A typed reviewer name is not authentication. ## Choices and evidence to audit | Choice | Completed teaching interpretation / required check | | ---------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Correct, then accept a draft | Correct to 12,050 cents. Keep source, revision, original and corrected fields, reviewer, rationale, and decision. Acceptance retains a practice draft; it does not authorize posting. | | Reject | Reject an unusable proposal with a reason. Retain the attempted proposal; invalid fields must not make rejection unavailable. This is an alternative disposition, not a second decision on the accepted attempt. | | Escalate | If evidence is missing, contradictory, or out of scope, keep the case with an accountable manual/source owner and a next action. Escalation is not invoice completion. | | Pause / resume | Retain fields and disable decisions while paused. Resume deliberately. A paused browser page does not establish durable work ownership or stop an elapsed timer. | | Export | Include source revision, initial and final fields, decision, rationale, source-check assertion, edit events, and timing limitations. An editable local export is neither an immutable audit trail nor production approval. | | Save fails | Show that the decision is in memory only and request an export. For a real service, exercise recovery and ownership rather than assuming that the click was retained. | ## Hypothetical capacity and overflow model The model assumes five active review minutes per case, including source reading, correction, and the decision record. Each available reviewer has thirty minutes for this batch window. These figures are estimates, not observed review performance. | Scenario | Incoming cases | Minutes / case | Reviewers | Available minutes / reviewer | Demand minutes | Capacity minutes | Admission limit | Modeled overflow | | ------------------- | -------------: | -------------: | --------: | ---------------------------: | -------------: | ---------------: | --------------: | ---------------: | | Routine window | 12 | 5 | 2 | 30 | 60 | 60 | 12 | 0 | | Burst window | 20 | 5 | 2 | 30 | 100 | 60 | 12 | 8 | | One reviewer absent | 12 | 5 | 1 | 30 | 60 | 30 | 6 | 6 | Calculations: ```text demand_minutes = incoming_cases × minutes_per_case capacity_minutes = reviewers × available_minutes_per_reviewer admitted_cases = min(incoming_cases, admission_limit, floor(capacity_minutes / minutes_per_case)) overflow_cases = incoming_cases − admitted_cases ``` Completed proposed overflow disposition: - Owner: accounts-payable queue owner; verify actual availability and delegate before a trial. - At the declared limit: pause new automated intake and retain excess cases in an assigned holding queue. - Next review point: the next confirmed review window, with age and delay visible. Define an actual waiting limit before testing. - Manual fallback: use only when the manual team's owner confirms capacity; otherwise retain the hold and escalate the delay. - Retained checks: full source comparison, correction/rejection/escalation, current authority, and separate posting verification. - Unimplemented work: this worksheet and arithmetic do not enforce intake limits, provide staff capacity, or operate a queue. ## Proposed later trial Compare manual and assisted work on different but comparable permitted cases. Vary which method people try first so practice with one method does not always help the other. Include wrong amounts and source conflicts. Independently inspect final correctness and unsafe acceptance, then measure active review effort, edits, missed errors, abandonment, waiting, escalation, and support demand. Browser elapsed time includes idle and paused time. It is not labor saved. Edit counts are not correction accuracy. Include receiving-team and unresolved-case costs before estimating cost per accepted draft. No trial has occurred in this example. ## Your working copy Start with one source, its deliberately altered proposal if this is a permitted test, and the correction a separate checker expects. Audit the choices before filling in capacity or savings estimates. - Which workflow is this, and what draft or decision can the reviewer accept? - Who owns the workflow? Who independently checks whether the accepted result is correct? - Which sources may the reviewer inspect, which revisions apply, and how current must they be? - What access, redaction, retention, and evidence-storage rules apply? - Which effects are excluded? Who separately has authority to post or take another business action? - What is the packet ID? Record the invoice identity, currency, amount, and source revision, including any conflicting evidence. - What did the initial proposal say? If this is a permitted test, which error was deliberately introduced? - What result should the check require, which source supports it, and who checks that expected result independently? - What final fields and decision were saved? Retain the correction reason and how the reviewer's identity was verified. | Choice | Available to the actual role? | Evidence retained | Failure / handoff / owner | | ------------------------------- | ----------------------------- | ----------------- | ------------------------- | | Correct and accept within scope | | | | | Reject | | | | | Escalate | | | | | Pause and resume | | | | | Save and export | | | | - What independently checks source correctness, beyond shape and a checkbox? - What happens if the source changes, evidence is unavailable, the save fails, or the reviewer is interrupted? - How many hands-on review minutes does a case need, and what supports that estimate? What changes for a difficult case or an interruption? - Who is available during a burst or absence, how many cases may enter review, and how long may the others wait? - Where do extra cases wait, who owns them, and when are they checked again? Has the manual team's owner confirmed room for any work sent there? - Which deliberately wrong proposals must the review reject? Who independently owns their expected results, and what happened in each test? - Which cases would enter a later trial, how will comparable cases be selected, and in which order will people try manual and assisted work? - What will that trial measure, which safety limits must hold, and what would stop it? - What does accepted, rejected, held, and escalated work cost, including support and recovery effort? - What is the next bounded decision, who is accountable, what evidence is still needed, and when will it be reviewed? Keep the packet audit, capacity model, and measured trial results separate. Completing this sheet grants no release or business-action authority.