Features

Draw and evaluate an audit sample

Draw a monetary unit, random or systematic sample from a recorded seed, enter the results in Excel and evaluate a monetary unit sample.

Updated November 18, 2026 · 7 min read

Audit sampling draws a sample from the lines of an analysis, keeps every draw in its history and evaluates the results you record for each line. The analytics engine computes the plan, the draw and the bound. The assistant can prepare a draw, but it never chooses the seed and never fills in results.

Open an analysis and go to the Samples tab, then click New sample. Drawing or evaluating a sample needs full access to an open analysis. Members with read-only access can read the draws and download them.

Methods

MethodWhat it draws
Monetary unit sampling (by amount)Lines in proportion to their amount, from a plan Blast computes first.
RandomThe number of lines you enter under Lines to draw, each with the same chance. A line is drawn once at most, unless you tick A line may be drawn more than once under Advanced settings.
Systematic (regular interval)The number of lines you enter under Lines to draw, one line every N, from a start drawn from the seed.

Choose the Population first: the general ledger lines, or a table the analysis produced (under Advanced settings, Other table of the analysis (optional)).

For a monetary unit sample, enter the Tolerable misstatement. Advanced settings holds the Confidence level (95 % unless you change it), the Expected misstatement, the Amount tested and the negative amounts. The confidence level is how sure you want to be that the misstatement does not exceed the bound: 1 minus the risk of incorrect acceptance.

The Sample plan beside the form computes by itself whenever a setting changes: the most lines to check, the sampling interval and the lines left out. Zero amounts and lines without an amount cannot be selected. Negative amounts (credit notes, reversals) are left out, or sampled on their own when you choose Sample them separately. The plan uses the gamma method without an expansion factor: when the expected misstatement is above 0, a table built with an expansion factor gives a larger size. Amounts show without a currency symbol, because a ledger does not state its currency.

Under Advanced settings, Placement of the draws chooses how the points fall:

Fixed interval: one point every interval, from a start drawn from the seed.

Cell: one point drawn at random inside each interval.

Random monetary units: points drawn at random over the whole book value.

A line whose amount is at least the sampling interval is always in the sample.

The seed

Every draw starts from a seed, a whole number from 0 to 9,007,199,254,740,991 (2^53 − 1). The screen never asks for one. When you click Draw up to N lines (Draw the N lines for a random or systematic sample), Blast draws the seed with the server's cryptographic random generator and records it with the draw. When the assistant prepares a draw, Blast proposes the seed and you approve it with the draw.

The same lines, the same settings and the same seed give the same sample, so a reviewer can redo the draw. On the sample's page, Details, Redo this draw identically shows the seed and the draw's entry in the History tab.

Abandoned draws

Every draw stays in the history, including the ones nobody evaluates. The Samples list shows each one with its author and its date. When a later draw takes the same population, the earlier one shows Drawn again, not evaluated. If someone draws again to avoid some lines, anyone who reads the analysis sees both draws.

Record the results

Each line gets a status, chosen from the list in its cell:

StatusWhen to use it
AuditedYou checked the line against its documents. Enter the amount they support, the same as the book value if there is no error.
Not testableYou could not check the line. Leave the audited value empty: the evaluation counts its whole amount as an error.
Isolated anomalyYou have shown that this error does not happen elsewhere. Enter the audited value and a written justification. The evaluation adds the error as it is and does not extrapolate it. When in doubt, choose Audited.

The sheet shows these words. A hidden sheet, Codes Blast, maps them back to the codes Blast reads: do not change or delete it.

On the sample's page, Enter the audited values offers two ways.

Fill in Excel: in the add-in, click Analytics at the top of the agent. Evaluate a sample takes four steps: Choose the sample, Create the sheet, Fill in, Send and evaluate. The add-in lists monetary unit samples only, and writes a sheet of 10,000 lines at most: for a larger sample, fill a file on the web. Fill in the blue columns Status and Audited value (and Justification for an isolated anomaly), then click Send and evaluate. The add-in writes the evaluation sheet to the workbook; if Excel refused it, Write the evaluation sheet writes it again without sending anything.

Fill a file: click Download the file to fill, fill in the workbook in Excel and save it without changing its columns. Drop it as it is under Drop the filled file. Do not save it as CSV: the web refuses a CSV that keeps the source_row column.

Results that do not match the draw are refused, never corrected: a missing line, a line named twice, a line that is not in the sample, a changed book value, a changed Codes Blast sheet or the file of another sample. The message says what is wrong. Fix the sheet and send it again.

Evaluate a monetary unit sample

Under Evaluation settings, choose the Evaluation:

Standard calculation (Stringer bound), with Poisson factors, for any placement of the draws.

Calculation by cell (cell bound), only for a sample drawn by cell.

Then choose how to Compare with the tolerable misstatement. Total before offsetting the two directions (gross upper bound) compares the upper bound of each direction with it. Total after offsetting the two directions (net upper bound) first subtracts the most likely misstatement of the other direction. The add-in offers the same choices under Evaluation settings.

Click Evaluate the sample. The evaluation shows overstatements and understatements apart, and positive and negative amounts apart when negatives were sampled separately. For each one it gives the projected misstatement, the upper bound (Estimated maximum misstatement on the web, Upper misstatement limit in the add-in) and the conclusion: Satisfactory result when it stays below the tolerable misstatement, Unsatisfactory result when it exceeds it. When a misstatement is larger than its book value, that direction gets no bound: review that line. On the web, Download the report saves the evaluation report (.html) and Export to Excel saves it as a workbook.

A random or systematic sample can be drawn and downloaded (Workbook (.xlsx)). Blast does not evaluate it yet.

Validity domain

Some estimators give a reliable bound only from a certain sample size and number of misstatements found. Blast calls that range the validity domain and takes it from its coverage study. Outside the domain, the evaluation returns no bound and its conclusion is Outside the method’s domain.

The engine also holds the classical PPS evaluation and the variables estimators (mean per unit, difference, ratio and regression). In the current version of the study, none of them met the criterion at any sample size up to 1,000, so their domain is empty and the Samples tab does not offer them. The Stringer and cell bounds have no domain: the study accepted them at every sample size it measured, at 95 % confidence, the Stringer bound on fixed-interval samples and the cell bound on cell samples.

The sampling coverage study gives the populations, the criterion and the results.

Downloads

On a sample's page, Download the file to fill gives the workbook (.xlsx). Under Details, CSV for a spreadsheet gives the lines drawn and CSV to draw it again gives a CSV for a replay. The replay CSV keeps every value exactly as Blast read it: do not open it in a spreadsheet.