Eight hours, and everything that has to be true before you may bill them.

Nothing has been inherited yet. That is the point of this stop, and it is not a formality—it is the single assumption every later claim rests on. A timekeeping system that opens the day already filled in, with last week's hours copied forward or a schedule pre-loaded, has made the record before the work happened. Every number after that is an edit to a guess.

Federal timekeeping law is unusually blunt about this. The employee records their own time, contemporaneously, and changes leave a trail. So Arcvue will suggest—it knows which contracts EMP-1042 is authorized on, and it will offer them—but it will not enter an hour on anyone's behalf. The distinction sounds pedantic until an auditor asks who typed the eight, and the answer has to be a person.

The nothing from stop 01 becomes something here, and it becomes something in the worst possible place: a government facility with no reception. This is the ordinary case in this industry, not the edge case. If contemporaneous recording only works where there is a network, then contemporaneous recording does not work.

So the record is written on the device and held there. It carries the device's own clock reading, and—separately—the server's clock reading at the moment it is finally received. Both are kept. The gap between them is not a defect to be smoothed over; it is evidence about when the work was actually recorded, which is exactly the question at issue.

The handover happens when the app is next opened or brought back to the foreground, not the moment signal returns. There is no background listener waiting on the radio: an entry made in a basement by someone who then drives home without reopening the app is still on the phone. What that does not touch is contemporaneity—the device clock already recorded when the work was entered, which is the entire reason both clocks are kept.

The two timestamps from stop 02 arrive together with two lines of work. A system that accepts both and sorts it out later has already created the problem it will spend the month solving. EMP-1042 is authorized on SS-CTR-2026-0087. He is not authorized on SS-CTR-2026-0104, and the second line is refused at entry.

Refused, not flagged. The difference is the whole design. A flag creates a queue, a queue creates a backlog, and a backlog gets cleared under deadline by someone who was not there. A refusal happens while the person who worked the hours is still holding the phone and can say what actually happened—which is the only moment the truth is cheap.

One accepted line of 8.0 hours came out of stop 03. It is still only a claim. What converts a claim into evidence is an attestation by the person with first-hand knowledge. Most software treats that signature as a checkbox on the way to a workflow. It is the load-bearing element of the entire record.

So the attestation is specific about what is being asserted, it is captured with the entry rather than at the end of the month, and it cannot be given by anyone else. If EMP-1042 is on leave when the period closes, the correct outcome is an unattested timecard and a visible exception—not a supervisor signing in his place.

The attested entry from stop 04 goes to EMP-1007, who approves it when the timecard period closes. Approval is a period-level act on what the employee has already certified—not a daily queue somebody clears each morning.

Approval is a real control. It is recorded with an identity and a timestamp, and what it governs is the posting to the general ledger. What it does not govern, on a time-and-materials contract, is the invoice—because a T&M invoice is built from the timecard itself. The thing standing between an hour and a bill is the attestation at stop 04, not the approval at stop 05.

Approved at stop 05, the entry posts. Eight hours against SS-CTR-2026-0087—and now it stops being one person's Tuesday and becomes part of a period total. That total is where most timekeeping arguments quietly go wrong, because the period total is usually a number somebody typed.

It posts when it is approved, and there is no month-end labor batch behind it. The eight hours are on the contract and in the ledger inside the period they were worked—so the contract's cost position, the income statement and the balance sheet all read through today rather than through the last month somebody closed. Waiting on a close to find out how the month is going is a habit, not a constraint.

June 2026 has 22 weekdays. Twenty-two eight-hour days is 176.0 hours. Juneteenth falls on Friday the 19th and is charged to a holiday code rather than to the contract, so the hours chargeable to SS-CTR-2026-0087 are 168.0. Every burdened figure on this site descends from that 168.0—so it is shown here with its parents rather than asserted.

The 8.0 hours of record from stop 06 meet the wage and the three indirect rates. This is the same cascade the money chain runs on a month; here it runs on a single day, at the same rates, so you can check the arithmetic without a spreadsheet.

Twenty-one working days at $1,186.12 is $24,908.52. The month as filed is $24,908.35. The seventeen cents is not an error—it is what happens when you round once per day instead of once per month, and it is why Arcvue computes the burden at the period rather than by summing rounded days.

The cost from stop 07 is $1,186.12. The bill is $1,313.60, and the two numbers have no arithmetic relationship whatsoever—the wage comes from payroll, the billing rate comes from the executed contract, and they are never allowed to touch. The day either of them derives from the other is the day a raise silently reprices a contract.

$164.20 per hour is the LCAT-101 rate on SS-CTR-2026-0087: the GSA schedule ceiling of $178.48 less the 8% discount offered at award. This one Tuesday is 8.0 hours of the 168.0 that appear on INV-2026-0614, which the accounting page follows all the way to cash.

Everything above has been arriving at this. The billed hour from stop 08 is now a line on a closed invoice on a closed contract year, and an auditor has picked it—not because anything looked wrong, but because sampling is how this works.

The reason the answer is one screen is the thing that seemed most pedantic at the time: stop 01 insisting there was no record before the work, and stop 02 keeping both clocks. Those two facts are the first two rows below. A trail that begins at approval can tell you who signed off; it cannot tell you whether the record was made by the right person at the right time, which is the actual question.