AI Modules/Procedures/Sampling

SA 530 · ପ୍ରକ୍ରିୟା· विधि

Sampling

The formula shown. Selections seeded. Peer reviewers obtain identical ଭାଉଚର୍.

Start the free trial →Book a walk-through

CORAA Sampling, per-WP plans with SA 530 formula rendered

When a peer reviewer asks why the sample size is 62, 'the engine picked them' is not an ଉତ୍ତର. CORAA renders the SA 530 formula explicitly, sample size equals ceiling of (confidence factor × population value) divided by performance materiality. Confidence factors derive from the Poisson distribution. Selections are seeded, peer reviewers running the same seed obtain identical ଭାଉଚର୍.

Per-Working-Paper plans, not one ଏଙ୍ଗେଜମେଣ୍ଟ୍-wide sample
Formula rendered on every plan: ceil((CF × Population value) / PM)
Confidence factors: 3.0 at 95% (high risk), 2.3 at 90% (medium), 1.6 at 80% (low)
Force-census rows for 100% testing, year-end manual JEs, related parties, employee benefits
Seeded selection per SA 230 Para. 8, peer reviewers obtain identical ଭାଉଚର୍
Lock requires the rationale captured at sign-off

Two paths, one ledger

The old way, and ours.

Two paths to the same audit conclusion. One leaves traces; the other doesn't.

Traditional

The old way

-Sample sizes set on judgment, 'around 30 ଭାଉଚର୍ per balance head'
-Confidence factor and risk assessment rarely documented
-Selection done by picking 'every nth' ଭାଉଚର୍ or visually significant ones
-Peer reviewer cannot retrace the selection logic
-Adjustments mid-ଏଙ୍ଗେଜମେଣ୍ଟ୍ (more ଭାଉଚର୍ added) not documented

Defensible only on the ଅଡିଟର୍'s word. SA 230 reproducibility test rarely satisfied.

CORAA

On the Ledger

Sample size derived from a formula visible on the Working Paper
Confidence factor explicit, 3.0/2.3/1.6 mapped to risk level
MUS for high-value populations; simple random for low-value, homogeneous
Force-census rows configurable: year-end manual JEs, RPT, employee benefits
Selection seeded, ଏଙ୍ଗେଜମେଣ୍ଟ୍ ID + balance head code derives the seed
Peer reviewer with the same seed obtains identical ଭାଉଚର୍

Every sample defensible. SA 230 reproducibility satisfied. NFRA-inspection ready.

How it works

Three steps. Every trace logged.

Step 01

Set the risk level per balance head

For each Working Paper, the ଅଡିଟର୍ sets the Risk of Material Misstatement, High (95% confidence), Medium (90%), or Low (80%). The confidence factor follows: 3.0 / 2.3 / 1.6 from the Poisson distribution at -ln(α).

Step 02

Sample size renders

Sample size = ceil((confidence factor × population value) / Performance Materiality). Worked example: Trade Payables population ₹5 crore, PM ₹35.8 lakh, high risk = ceil((3.0 × 5,00,00,000) / 35,80,000) = 42 ଭାଉଚର୍. The formula is rendered on the Working Paper.

Step 03

Lock the plan

Locking generates the deterministic ଭାଉଚର୍ selection per SA 230 Para. 8. The seed is derived from ଏଙ୍ଗେଜମେଣ୍ଟ୍ ID + balance head code. A peer reviewer running the same seed obtains the same ଭାଉଚର୍. Locked plans are read-only; unlock requires a documented note.

Inside the module

What you actually get.

Formula visible, ନା black box

Every Working Paper renders the sample-size formula explicitly: ceil((Confidence factor × Population value) / Performance Materiality). The confidence factor, the population value, and the PM all show on the page. The ଅଡିଟର୍ can defend the sample size on the spot.

›Formula shown on every plan
›Confidence factor explicit with derivation
›ଅଡିଟର୍'s adjustment (with reason) supported
›କଭରେଜ୍ percentage shown alongside count

Confidence factors from the Poisson distribution

ICAI SA 530 Para A11 + Appendix 3 derives confidence factors from -ln(α) where α is the acceptable sampling risk. High risk → 95% confidence → CF 3.0. Medium → 90% → CF 2.3. Low → 80% → CF 1.6. CORAA renders the math.

›High risk: 95% confidence, CF = 3.0
›Medium risk: 90% confidence, CF = 2.3
›Low risk: 80% confidence, CF = 1.6
›ICAI Appendix 3 table referenced

Force-census rows for 100% testing

Some rows demand 100% testing regardless of sample math, year-end manual journals (last 14 ଦିନ), related-party ଲେଣଦେଣ, employee benefits over PM threshold. Configure once per ଏଙ୍ଗେଜମେଣ୍ଟ୍; CORAA enforces.

›ବର୍ଷ-end manual JEs (last 14 ଦିନ)
›Related-party ଲେଣଦେଣ (when RPT list is set)
›Employee benefits above PM
›ଅନ୍ୟ high-risk patterns configurable

Seeded reproducibility per SA 230

Selection is seeded from a deterministic value: ଏଙ୍ଗେଜମେଣ୍ଟ୍ ID + balance head code, hashed via SHA-256. A peer reviewer running the same seed obtains the same ଭାଉଚର୍. SHA-256 algorithm and numpy version are pinned, so identical seeds yield identical selections ବର୍ଷ later.

›Deterministic seed derivation
›SHA-256 and numpy version pinned
›Peer reviewer can retrace
›ଅଡିଟ୍ trail preserves seed value

Frequently asked

Answers, up front.

Per ICAI SA 530 Para A11 and Appendix 3: sample size equals ceiling of (Confidence factor × Population value) divided by Performance Materiality. The confidence factor depends on risk level: 3.0 for high risk (95% confidence), 2.3 for medium (90%), 1.6 for low (80%). Worked example: Trade Payables population ₹5 crore, PM ₹35.8 lakh, high risk = ceil((3.0 × 5,00,00,000) / 35,80,000) = 42 ଭାଉଚର୍.

Each random selection is seeded with a deterministic value derived from the ଏଙ୍ଗେଜମେଣ୍ଟ୍ ID and balance head code, hashed via SHA-256 and pinned to a specific numpy version. A peer reviewer running the same seed obtains the same ଭାଉଚର୍. The seed is rendered explicitly on every Working Paper. This satisfies SA 230 Para. 8 reproducibility.

MUS (Monetary Unit Sampling) is used for high-value populations where the top items concentrate most of the monetary value, for example, Trade Payables where 80% of value sits in 20% of ଭାଉଚର୍. MUS picks ଭାଉଚର୍ in proportion to their amount, achieving high coverage with fewer items. Simple random is used for low-value, homogeneous populations. CORAA picks the method automatically based on the population's distribution; the choice is rendered on the Working Paper.

ହଁ, the ଅଡିଟର୍ can adjust the sample size up or down. Every adjustment requires a reason captured in the ଅଡିଟ୍ ଟ୍ରେଲ୍ per SA 230. Typical adjustments: +N for prior-year ନିଷ୍କର୍ଷ, -N for highly automated processes, +N for sub-category coverage. The adjusted number, the reason, and the ଅଡିଟର୍ name are all logged.

See it on a real ledger

Run sampling on one of your engagements.

Bring a Trial Balance and a General Ledger. We'll walk through procedures end-to-end on your data, not a sandbox.

Start the free trial →Book a walk-through