Munsif — Performance Review Audit

How Munsif works — four steps, every cycle

Ingest

Reads this cycle's review data, whatever scale or format your company already uses.

Compute

Calculates real statistics — manager averages, deviations, score/text gaps. Not a guess.

Explain

Turns each flag into one plain-English sentence HR can act on immediately.

Certify

Produces a fairness certificate for the cycle — before reviews are released.

Review data

Edit any cell directly, add or remove rows, or import your own CSV.

Import CSV

Manager	Employee	Lang	Score	Review text

Settings

Tune the thresholds and assumptions the audit runs on.

Mismatch threshold

Thin doc max words

Outlier deviation ≥

Strong outlier ≥

Assumed team size

Avg monthly wage (PKR)

Replacement cost (months)

Positive keywords — English

Negative keywords — English

Positive keywords — Roman Urdu

Negative keywords — Roman Urdu

Runs on whatever is currently in the table above — 18 reviews. Edit, add, or import data any time and re-run.

Manager consistency

avg vs. company baseline

Consistent Outlier Strong outlier Trend (simulated) ⓘ

Flagged reviews

Review detail

click a flag to inspect

Select a flagged review to see the fairness breakdown.

Cycle fairness certificate

—

GENERATED LOCALLY · MUNSIF AUDIT ENGINE v0.3

Why this isn't "just ChatGPT"

Anyone can paste one review into a chatbot and ask if it sounds biased. That doesn't scale, doesn't remember your company's own history, and can't reliably do arithmetic across hundreds of reviews.

Computed, not guessed

Manager averages, deviations and outliers are calculated directly from your data — real statistics, not a language model estimating a number from memory.

Built for how Pakistani teams actually write

Reads English and Roman Urdu review text side by side, and checks documentation against local labour-law standards most generic tools never encode.

Never leaves the building

Runs entirely on your own machine against your own data. There's no upload, no cloud account, and nothing to negotiate with legal about data leaving the company.