How every figure is computed — and what this product does not claim

1. Data sources and the hierarchy

Every important data point is stored with its value, source, source URL, observation date, collection date, frequency, method, confidence and version. Sources are ranked; a lower tier never overrides a higher one for the same fact.

Today only AMFI is ingested: scheme identity, the AMFI category heading, the daily NAV and the NAV history for 69 scheme codes across 34 funds (1,95,407 dated observations). Fields that depend on factsheets or index data are shown as Insufficient data. Nothing is estimated in their place.

2. Returns

• Point-to-point. NAV at the end of the series against the NAV on or just before the date the stated number of months earlier (up to 7 calendar days earlier, otherwise Insufficient). Under one year the figure is the simple change; one year and longer it is also annualised.
• CAGR. (NAV_end / NAV_start)^{1 / years} − 1, with years = calendar days / 365.25.
• "Since 2015". From the first NAV held in the dataset, not from launch; the label says so.
• Growth vs IDCW. An IDCW option's NAV falls on each payout, so its NAV series understates total return; no return is computed for it and the page says why.

3. Rolling returns and consistency

For each window length (1, 3, 5 years) every start date stepping one week from the first NAV is taken; the annualised return to the date one window later is recorded. The page shows the number of windows, the worst, the 25th percentile, the median, the 75th percentile, the best, and the share of windows that were negative. At least 12 windows are required; otherwise Insufficient.

4. Risk

• Volatility. Standard deviation of daily log returns over the trailing three years (and over the whole history), annualised by √250. Requires at least 30 daily returns.
• Downside deviation. Root-mean-square of the negative daily log returns only (target 0% a day), annualised by √250.
• Maximum drawdown. The largest fall from any running peak to a later NAV; the peak and trough dates are shown.
• Recovery period. Calendar days from that trough to the first NAV at or above the previous peak; "not yet recovered" when it has not happened.

5. Benchmark and the reference fund

The scheme's official benchmark and its total-return series are not on file (the index provider's public endpoint refused automated download when this was built), so no benchmark-relative figure is shown and the field says Insufficient data. For equity, index and hybrid schemes a reference fund — the Direct plan of a Nifty 50 index fund in the set — is compared on shared dates over three years: CAGR difference and the share of rolling one-year windows in which the fund was ahead. It is labelled a proxy for the broad market every time it appears and is never called the benchmark. When the official series is ingested it feeds the same function.

6. Direct vs Regular

Both plans of a scheme hold the same portfolio; the Regular plan's NAV grows more slowly by roughly the distributor commission. The gap is measured directly from the two NAV series as the three-year CAGR difference on shared dates. It is the observed gap, not the published expense-ratio difference, which is not on file.

7. Data freshness

Every dated observation is classed against a window for its kind — NAV: fresh ≤ 3 days, recent ≤ 10; portfolio: fresh ≤ 45, recent ≤ 75 — as FRESH, RECENT, STALE or INSUFFICIENT. Stale data is labelled as older than the preferred window; it is never shown as current.

8. Versioning

NAV snapshots are immutable: a re-run of the ingestion writes a new snapshot only when AMFI's latest date has moved on, and the previous file is kept under historical/. The engine has a version (1.0.0); the dataset has a version (2026-09-17); both are printed on every page.

9. Goal and retirement arithmetic

• Projection. Lump sum today plus a monthly contribution at the start of each month, compounded monthly at the equivalent of the stated annual rate; the monthly amount (and any annual lump sum) rises by the step-up every twelve months.
• Scenarios. 8%, 10% and 12% a year, labelled illustrative. They are not expected returns, forecasts or promises; you may add your own rate.
• Required SIP. The level monthly amount (with the same step-up) that reaches the target under the scenario, by linear scaling of a unit SIP. Required step-up is found by bisection between 0% and 50%.
• Inflation. Never applied silently. The default (6%) is an assumption shown in the input. A nominal target is also shown in today's purchasing power: target / (1 + inflation)^years.
• Retirement corpus. Today's monthly expenses × 12, inflated to retirement, less any pension; the present value at retirement of that amount growing with inflation for the chosen number of years while the corpus earns the post-retirement rate (growing-annuity formula, annual). A sensitivity table moves inflation ±1 point and duration ±5 years.

10. Riskometer and categories

Categories are AMFI's own headings, which follow SEBI's categorisation circular; the engine never reassigns a scheme. The riskometer uses SEBI's six levels — Low, Low to Moderate, Moderate, Moderately High, High, Very High — and is shown only when a scheme's own label has been ingested from its factsheet or SID. No level is assigned by this site.

11. What "best" means here

A question like "best midcap fund" is treated as a research question. The answer is "it depends on what you are optimising for", followed by a transparent shortlist: each fund in the category is listed with the dimensions on which it stands in the top third of the set — rolling 3-year median, share of positive 3-year windows, shallower drawdown, lower volatility, faster recovery, larger Direct-plan advantage — with the value beside each. Nothing is combined into a score, no fund is ranked first, and the dimensions with insufficient data are named.

12. What this product does not claim

• It does not recommend buying, selling or holding any scheme, nor an allocation.
• It does not forecast returns; every rate in the planners is a scenario you can change.
• It does not estimate or fill missing data — expense ratio, AUM, holdings, manager, riskometer and benchmark figures are absent until the Tier-1 record is on file.
• It does not compute tax in this release; the tax data architecture is versioned by framework (Income-tax Act 1961 / AY 2026-27 and Income-tax Act 2025 / TY 2026-27) so the two are never mixed when rules are added.
• It stores nothing you type, sends nothing anywhere, and asks for no PAN, Aadhaar, bank or broker details.
• Personalised investment advice in India is regulated (SEBI Investment Adviser and Research Analyst regulations); this site provides research, education and scenario arithmetic, and the boundary is deliberate.