Methodology

The rulebook.

Every flag in this system traces back to one of the entitlement rules below. Nothing is inferred, weighted by a model, or learned from historical outcomes — each rule is a direct encoding of a statutory provision or binding precedent.

BNSS s.479(1) — HALF-CUSTODY RULE

Undertrial has served ≥ 50% of maximum sentence

Where an undertrial (not a first-time offender, and not facing an offence punishable by death or life imprisonment) has undergone detention for a period extending to one-half of the maximum term of imprisonment specified for that offence, the court shall release them on bail.

BNSS s.479(1), Proviso — FIRST-TIME OFFENDER RULE

First-time offender has served ≥ 33% of maximum sentence

A first-time offender (no prior conviction) is entitled to release once they have undergone detention for one-third of the maximum sentence for the offence — a lower threshold than the general rule, reflecting reduced culpability concerns for those without a prior record.

SATENDER KUMAR ANTIL v CBI (2022) — CATEGORY A/B

Offence falls in a presumptive-bail category

The Supreme Court's binding classification places offences punishable by up to 7 years (Category A) and economic offences not covered by special statutes (Category B) into categories where bail is the norm, not the exception — custody beyond a reasonable period without trial progress weighs heavily in favour of release.

HUSSAINARA KHATOON v STATE OF BIHAR (1979)

Prolonged custody disproportionate to alleged offence

The foundational undertrial-rights judgment: detention that has already outlasted what a conviction would likely carry violates Article 21's guarantee of a speedy trial. Applied here as a proportionality check against custody duration relative to offence severity.

AGE & VULNERABILITY GROUND

Accused aged 60 or above

Judicial guidance consistently treats advanced age as a humanitarian factor favouring bail consideration, given the disproportionate health risk of continued incarceration. Encoded here as a flag for expedited review, not an automatic entitlement.

What this model deliberately does not do

No prediction. No risk score. No learning from outcomes.

This is a rules engine, not a machine-learning model. It does not estimate the probability that someone will reoffend or fail to appear for trial — those predictions, however accurate, encode historical policing and sentencing bias into a number that looks neutral but isn't.

Instead, every check in this system answers a narrower, answerable question: has this person already met a threshold the law itself set? That question has a factual answer. Whether to actually grant bail remains, as it must, a human judicial decision informed by full case facts this system does not have access to.

The tiering shown in the dashboard (Flagged / Review / Clear) reflects how many independent legal grounds apply and how far past the relevant threshold the case sits — not a composite "liberty score" produced by weighting unrelated factors together.