Safety metric definitions in five selected reports

Lost time injury frequency rate (LTIFR) is not a complete definition. A usable comparison also needs the event rule, the people and locations covered, the reporting period, the exposure basis and the meaning of the displayed value. Matching the acronym or rate scale cannot resolve differences in those fields.

This study preserves five purposively selected cases from Nitium’s fixed historical edition 0.12.17-local1475-gipcl-cera, frozen on 6 October 2026. It examines one interpretation question per report and retains two accepted observations to distinguish a rate from a count. The source population has not been replaced with newer or more convenient issuers. The study adds no company-data records.

We reread the exact official report editions and their selected safety pages on 15 October 2026, using same-day retrieved files whose hashes match the earlier evidence. These are historical cases, not a latest-data survey. Page references below identify physical PDF pages and printed numbers. No incident log, medical record or underlying hours register was inspected.

Five source-linked safety-definition cases

1. Cochin Shipyard: population coverage

Cochin Shipyard’s standalone FY2024–25 safety table separates Employees and Workers and expressly includes contract workforce. LTIFR is stated per million person-hours worked. The selected table does not provide the underlying hours. Preserve that coverage with the rate; a permanent-only headcount cannot supply its denominator. Source: Cochin Shipyard BRSR, PDF/printed 1 and 18, questions 13 and 11 respectively.

2. KPR Mill: the type of value matters

KPR Mill’s standalone FY2024–25 table reports one recordable injury for Employees and one for Workers; its footnote includes contract workforce. The FY2023–24 comparative injury and LTIFR cells display <1%. The LTIFR heading specifies a per-million-person-hours basis, while current LTIFR cells are dashes. Those tokens do not supply exact injury counts or comparable numeric rates. Do not sum categories as disjoint or carry the footnote into the original prior-year report. Source: KPR Mill annual report, PDF 51/68, printed 41/58.

3. Andhra Paper: a count is not an established numerator

Andhra Paper’s standalone FY2024–25 employee cells show LTIFR 0.13 per million person-hours and zero recordable work-related injuries. The selected table separates Employees and Workers but does not establish actual exposure hours or contract-workforce inclusion. Preserve both values. Their coexistence alone does not establish a shared event definition, justify reconstructed hours or prove an error. Source: Andhra Paper BRSR, PDF/printed 1 and 21, questions 13 and 11 respectively.

4. Mrs. Bectors: the event threshold stays with its edition

Mrs. Bectors’ consolidated FY2025–26 worker cells show LTIFR 0 per million person-hours and seven recordable injuries. Its LTIFR rule concerns work-related injuries preventing normal duties for at least 48 hours immediately after the incident. Coverage is Group manufacturing locations; corporate and regional offices, depots, warehouses and distribution centres are excluded unless specified otherwise. The FY2024–25 comparative is restated. Actual hours and contractor coverage remain unestablished. Keep these definitions with this edition; zero LTIFR does not mean zero recordable injuries. Source: Mrs. Bectors BRSR, PDF/printed 1 and 18–19.

5. Sandhar: a defined unit does not resolve a dash

Sandhar’s standalone FY2024–25 safety table states a per-million-person-hours LTIFR basis and includes contract workforce. It shows dashes for Employees and Workers in both current and FY2023–24 comparative safety cells. A stated basis can be retained while the numeric value remains unavailable; the dash does not establish zero injuries or fatalities. The safety-population note is not automatically a headcount definition. Source: Sandhar BRSR, PDF 1/23, printed 88/110.

What can this comparison establish?

The cases support a comparison of disclosed definitions and evidence limits. They do not support a league table of workplace safety. A rate-scale field can be complete while the event rule is unresolved. A population note can be explicit while actual hours are absent. A source can provide a numeric rate without giving enough inputs to independently reproduce it.

Keep two decisions separate: whether a reported value can be faithfully described, and whether it can enter a proposed calculation. A source-linked description can preserve an unresolved value without endorsing its measurement. A calculation needs stronger evidence about the relationship between the selected inputs. An unexpected pairing is a question for the data owner, not permission to overwrite either cell.

Original worked model: changing the rule changes the rate

The following is a fictional arithmetic exercise, not an industry definition or a reconstruction of any issuer’s data. One invented plant has four work-related events in a defined reporting year. Its authorised aggregate record gives inability to perform normal duties immediately after each event as 12, 30, 60 and 96 hours. For this exercise only, those durations are complete and unambiguous.

The same population and location contribute exactly 2,000,000 worked person-hours during that year. Both trial rules count each eligible event once and use those same hours. Rule A counts durations of at least 24 hours. Rule B counts durations of at least 72 hours. These thresholds are chosen solely to make the effect of different event rules visible.

Under Rule A, the 30-, 60- and 96-hour events qualify: three events. The rate is 3 ÷ 2,000,000 × 1,000,000 = 1.5 events per million person-hours. Under Rule B, only the 96-hour event qualifies: one event. Its rate is 1 ÷ 2,000,000 × 1,000,000 = 0.5 on the same scale.

The event register, population, period and hours have not changed. The lower second number comes entirely from the narrower fictional selection rule. It therefore cannot demonstrate an improvement in safety. Nor should either trial rate be labelled an issuer’s LTIFR: the example supplies its own definitions and has no mapping to a company’s reporting method.

Converting both rates to a 200,000-person-hour scale multiplies them by 0.2: 1.5 becomes 0.3 and 0.5 becomes 0.1. The unit conversion is valid within this invented model, but the definitions remain different. A common scale is one prerequisite for comparison; it does not make the counted events equivalent.

Original fictional model. The same register and hours produce different rates when event eligibility changes. Matching the rate scale cannot make different event definitions equivalent.

Now remove the actual hours. The event counts remain three and one under their respective rules, but neither rate can be reproduced. A year-end headcount, standard workweek or guessed attendance factor would introduce an unsupported exposure estimate. Record the missing denominator rather than replacing it. If the population or period changes, the same pause applies even when an hours figure is available.

A reproducible comparison record

For each proposed pair, retain five fields: event eligibility; included people and locations; period and source edition; numerical value or exact nonnumeric state; and exposure unit, multiplier and actual hours where available. Put the source locator beside the relevant field. A blank should identify what is unresolved rather than silently inheriting the neighbouring issuer’s definition.

In the complete fictional model, the decision is precise: each rate can be reproduced within its own rule, and the scale conversion works, but the two rates do not measure the same event set. In the missing-hours variation, neither rate can be reproduced. These are different outcomes and should not be collapsed into a single “data checked” status.

Before publishing a real comparison, ask an authorised reporting owner to resolve only the missing relationships: which events qualify, whether the hours cover exactly those people and locations, and whether both periods use the same definition. Keep the response and source version with the proposed calculation. Person-level incident and health information should remain with its authorised custodian; this review can use controlled aggregate confirmations.

Limits and related resources

The five cases were deliberately selected for interpretation questions. They do not estimate how common any practice is, establish compliance, explain incident causes or independently verify company measurements. No cross-company rate, pooled exposure total or operational trend is calculated. A later definition must not be inserted into an earlier edition without source support.

For the operational preparation process, use the occupational safety data guide. For separate historical observations and their coverage notes, use reported workforce and safety data. This study’s contribution is the five-report definition comparison and the explicit demonstration of what scale conversion can and cannot resolve.

Nitium’s reporting support can help organise aggregate evidence and clarification questions within an agreed scope. Contact Nitium with the intended reporting output and available source records; avoid sending personal incident or medical information in an initial enquiry.

Prepared 15 October 2026 with AI-assisted research and internal source and calculation checks. Source editions and qualifications are preserved. No independent assurance, issuer endorsement or client relationship is claimed.