Benchmarking engine
An arm’s-length range you can trace back to its rules.
Rule configuration, method, comparable search, PLI and range run as separate stages, each with a result you can open. The range rule follows the jurisdiction, and every figure is computed in code.
The pipeline
Six stages, each with its own result.
A stage hands the next one a stored result, so any step can be opened and checked on its own.
- 0Rule configurationThe jurisdiction’s range rule and tolerance band
- 1FARThe tested party’s functions, assets and risks
- 2Method selectionMethods weighed in order, CUP through profit split
- 3Comparable searchAn accepted set, every rejection coded
- 4PLIMargins for the tested party and each comparable
- 5RangeThe arm’s-length range and a risk status
The funnel
Every company that leaves says where it left.
The search narrows through numbered filters in a fixed order, and each one records how many companies went in and how many came out. The AI review runs last and judges business activity only, never size.
- A company is rejected unless its description confirms it matches.
- India follows SBC’s own screen order: 8 compulsory screens, 7 optional ratios, then functional similarity and the AI review.
- F1Geographic3,960
- F2Industry1,284
- F3Size702
- F4Independence418
- F5Data availability311
- F6Functional similarity126
- F7Profitability screen94
- F8Extraordinary events88
- F9AI qualitative review18
The margin
Every PLI from the same three lines.
Operating revenue, operating cost and operating profit are derived once from the tested party’s financials. Other income stays out of revenue and finance cost stays out of cost, so each ratio shows the figures it was divided from.
- The IRD, Instant PLI and a new study all use the same engine.
- The arithmetic is code, never a language model.
The range rule
The jurisdiction decides the bounds.
The range rule is set by the study’s jurisdiction and year, and carries the citation printed in the report. In India, the 35th and 65th percentiles are read by data place under Rule 10CA, so each bound is a real comparable’s margin or the mean of two.
- The data places are stored with the result and printed in Annexure 5.
- India with fewer than six comparables has no range: the arm’s-length price is the arithmetic mean.
- India35th to 65thRule 10CA, by data place
- UAE25th to 75thArticle 34, Federal Decree-Law 47 of 2022
- Everywhere else25th to 75thOECD Guidelines, paras 3.57 to 3.62
India: the bound is a place in the set
- 14.1%·
- 25.8%·
- 36.6%·
- 47.9%·
- 58.8%P35
- 69.4%Med
- 710.1%Med
- 811.7%P65
- 912.6%·
- 1013.9%·
- 1115.2%·
- 1217.8%·
12 × 35% = 4.2, so the 35th percentile is place 5 (8.8%). 12 × 65% = 7.8, so the 65th is place 8 (11.7%). 12 × 50% = 6, a whole number, so the median is the mean of places 6 and 7.
The result
Six statuses, each with a written test.
The tested party’s PLI is placed against the range and given one status. Each status has a fixed definition, so the same numbers always give the same answer.
- Within RangeInside the arm’s-length range.
- Low RiskOutside the range, inside India’s tolerance band of 1% for wholesale trading or 3% otherwise.
- Above RangeAbove the range by no more than twice its width.
- Below RangeBelow the range by no more than twice its width.
- High RiskOutside the range by more than twice its width.
- Operating LossAn operating loss that falls outside the range.
Coverage
What it covers.
- Stages
- 0 to 5
- Rule configuration, FAR, method, search, PLI, range
- Orbis funnel
- 9 filters
- AI review runs last
- India screens
- 15
- 8 compulsory, 7 optional
- India range
- 35th to 65th
- Rule 10CA data places
- UAE and elsewhere
- 25th to 75th
- Interquartile range
- Risk statuses
- 6
- Within range to operating loss
Questions
Before you ask.
- Does the screen show the same India range as Annexure 5?
- Yes. Both read the 35th and 65th percentiles by data place under Rule 10CA, so the range on screen is the range that is filed.
- What if an India study has fewer than six comparables?
- There is no range at that size. The arm’s-length price is the arithmetic mean, and the tested party is judged against it.
- Is any of the math done by AI?
- No. PLIs, percentiles, the range and the risk status are computed in code. AI reads documents and reviews company descriptions, and never produces a figure.
