How we value points and rank cards.

New to any of this? Start here first — this section assumes you already know what a point is. A card page leads with a percentage, the board leads with a rupee figure, and the two disagree on purpose. Three numbers, three different questions, defined once.

Which is why a card can print one number and pay another, in both directions. Axis Magnus for Burgundy earns 4.8% on the swipe and returns 11.24% net of fee at ₹5L a month, because its milestones and monthly bonuses do work the swipe rate never sees. It runs the other way too: HSBC Privé prints 3.3% and reads −19.59% at ₹1L a month, because its fee eats a small spender alive. The percentage is the engine. The rupee figure is the engine minus the fee, at your spend — which is the one the calculator exists to give you.

You can check us on this without taking our word for it. Every annual-₹ figure the site renders carries data-value-model in the markup — the string nav or the string ranking, on the figure itself where the server writes it and on the container where a tool fills it in your browser. View source on any card page, any board row, any calculator result. If a rupee figure anywhere carries neither, that is a bug and the report form is how you tell us.

every term, defined · start here · what a point is worth

Transferable bank currencies are anchored at the community standard of 1 partner airline mile ≈ ₹1 — the same basis the rest of the Indian points ecosystem quotes when it says "Atlas 4%, Infinia 3.3%". We use it deliberately: our headline reward rates are comparable with every number you'll read elsewhere, and the anchor itself is conservative — saver business-class redemptions routinely clear ₹2 per mile.

The single house value never tells the whole story, so every one of the 164 currencies in the Valuation Index also carries a floor (the guaranteed portal/cashback rate — what you get with zero effort) and a ceiling (the realistic best-case sweet spot — not a unicorn redemption). Discipline and sweet spots live in that range, not in an inflated headline. Marketing rates don't appear anywhere on this site.

One invariant runs through every card file

Each card's effective earn (earn.base, points per ₹100) times its currency's house ₹/point equals its headline reward rate: base × value ≈ rate. Redemption haircuts and transfer markups are baked into the base, so the headline is honest by construction. Two live examples:

• HDFC Regalia Gold earns 2.5 RP per ₹100, but its points cap out at ₹0.50–0.65 against HDFC's ₹1.00 house value — a 35–50% haircut. We book the base at 1.63, so 1.63 × ₹1.00 ≈ 1.6%. The haircut is priced in, and the card page says so.
• Axis Magnus for Burgundy earns 6 EDGE RP per ₹100 but transfers at 5:4 — double the standard 5:2 — so its base is booked at 12 against the ₹0.40 house value: 12 × ₹0.40 = 4.8%. The velvet-rope premium is priced into the card, not smuggled into the currency.

The ranking starts with a computed realistic annual value at three spend profiles — ₹1L, ₹3L and ₹5L a month — using each card's real earn tiers, milestone and monthly bonuses and fee waiver, valued at the house ₹/point. On top of the everyday rate we credit each card's best accelerator (SmartBuy, iShop, 5X online, app-hotel portals) on a realistic ~20% of spend — less for narrow or capped boosters — because a disciplined cardholder really does route some spend there. We assume 20%, disclosed, not the issuer's "up to" fantasy applied to everything; the calculator shows your everyday value and lets you tune the mix to your own wallet.

On top of the fair-value math we weigh currency quality (transferable beats closed-loop), transfer-partner strength (all 154 live routes are in the optimizer), lounge and premium benefits, and access gates — a card you can't get, or can only hold behind a relationship threshold, ranks accordingly. Hypothetical "up to" numbers count for nothing. When an issuer nerfs a card, its rank moves; the tracker is the audit trail.

Three value models, and which page runs which

The site computes a card's annual ₹ three ways, on purpose, and every annual-₹ headline says which one it is (in the markup, as data-value-model — on the figure itself where the server renders it, and on the container the tool fills where the number is computed in your browser):

• Net annual value — earn, milestones and bonuses, no accelerator (nav). Everyday spend at the card's own rate, plus milestones and monthly/quarterly bonuses, minus the fee — no accelerator credit at all. This is what every card page's three spend profiles, the calculator, the deck planner, the Portfolio Score and /best/first-travel-card print. It is the conservative number: what the card pays if you never chase a portal.
• Ranking value (ranking). The same arithmetic PLUS the uplift from routing ~20% of spend (less for narrow or capped boosters) through the card's best accelerator. It drives the called winner on /compare and the order of /best/business.

The value engine — your mix, your spend (engine). The third model, and the one the site now leads with: valueAt(card, spend) in src/lib/value-engine.ts prices a card on a STATED mix of spend — 30% groceries, 10% utilities, 20% travel, 6% fuel, 34% everything else, the calculator's own seed — at a stated monthly figure, MCC-aware, each category at the card's own published rate for that category, capped where the issuer caps it, plus milestones and monthly bonuses, minus the fee. No accelerator credit. It is what every card page's value block, the board (at your spend band's low end) and its six band pages, /best/high-spenders (at ₹5L a month on the same mix), the calculator and the wallet print, marked data-value-model="engine"; the break-even a card page ends on is the first rung of a stated ladder at which this engine's net reaches ₹0. One card, one number, everywhere it appears — npm run qa's check 19 recomputes it independently for every priced card at three spends and fails the build on a rupee of disagreement.

So the same card shows two different ₹ — IDFC FIRST Private at ₹3L/month reads ₹2,15,030 on its card page and ₹3,11,028 on the board. Neither is wrong; the accelerator credit is the whole difference, and a reader who never touches a bank portal should read the card page's number as theirs.

A third label sits next to these and means something else again: a card's headline earn rate is a percentage — what a rupee of ordinary spend earns at our house ₹/point, before milestones, accelerators and the fee. Where a page instead says "X% net of fee at ₹5L/mo spend", that is net annual value divided by the spend that produced it.

Bundled memberships are surfaced, never counted. Club Marriott, Accor Plus, Epicure and friends appear on card pages and in the membership matrix at honest ₹ values — but none of that ₹ enters net annual value or a rank. A dining club is worth ₹0 to a holder who never uses it, and folding opt-in perks into the headline would quietly reorder the rankings on usage assumptions we can't defend. The math above counts only what every holder earns.

Every card gets exactly one of three calls — no fence-sitting. Because our card pages publish machine-readable review markup, the scale behind those stars is disclosed here: the Review structured data on each card page maps our verdict onto a 5-point rating exactly as follows.

One org-authored review per card, no invented star counts, no aggregate ratings we don't have. The rating you may see in search results is this editorial verdict — nothing else.

The Portfolio Score grades your whole deck — not one card — on a 0–100 scale. It is one published formula, three pillars, no black box:

Score = round(100 × (0.60 × E + 0.20 × C + 0.20 × R))

E is earning efficiency — your deck's rupees against the best deck's. C is coverage — how much of your spend earns full rate. R is resilience — whether the points will still be worth something when you redeem them. The grade scale is below; the full derivation of all three is one tap away.

S ≥ 90 · A ≥ 75 · B ≥ 60 · C ≥ 45 · D ≥ 30 · F below

What the score does not know

The whole thing runs in your browser. Your spend, your deck, your score — nothing is uploaded, logged or stored anywhere but your own tab.

Everything the site shows lives as version-controlled, human-curated data — every edit is a reviewable diff. Official issuer terms come first: card facts are verified against T&C and MITC PDFs and program pages, not forum hearsay. The devaluation tracker holds 376 dated events spanning 1961–2026, and every single one carries a source URL you can click.

Valuation history is held to the same standard: a historical ₹/point is published only when a tracked event quantifies the change, each point names its event and shows its arithmetic, and we always take the most conservative figure in a stated range. Currencies with no quantified history get a single current point — we never invent a curve to make a sparkline look busy.

Our news desk is AI-written, and we'd rather tell you than have you guess. It is a phraser over a fact base, not a writer: the generator turns our own curated tracker events into posts and has no license to state anything that isn't already in the data — no invented cards, rates, fees, partners or dates.

Five fences, in order, every run:

1. Grounded — input is our curated tracker facts, nothing else.
2. Schema-valid — output that fails validation is discarded.
3. Cited — no source URL, no post.
4. Deduped — near-duplicates of recent items are skipped.
5. Audited — every attempt is logged with model, prompt hash and decision; rollback is a delete.

Rankings, valuations, verdicts and every number on this page are human-curated data — the machine phrases; it never decides.

No affiliate links. No display ads — not now, not later. Nobody pays for a rank, a verdict, or a kind word; rankings are derived from the data above and nothing else. That's a locked, written decision (ADR 0003), not a vibe. If that ever changes, the rule is already on the record: prominent disclosure on every page carrying a link, zero commercial influence on rankings or verdicts, and this page updates first. Until then, the only thing we're selling is the verdict.

Every program hub and the valuation index carry a nerf-risk chip: stable · watch · elevated · high. It is not a prediction and not an opinion — it is a weighted sum over the 376 dated, sourced events in the devaluation tracker, attached to each currency by the same inference that builds its nerf log. Same data, same page, one number.

In one line: score = max(0, Σ cuts − Σ buffs − quiet), where each cut-side event contributes kind × severity × attach × decay. Recent major devaluations weigh most; buffs earn credit back; a long quiet streak earns credit too. Thresholds: below 1 stable, below 2.5 watch, below 4.5 elevated, 4.5 or above high.

Today that grades the 164 currencies as 135 stable, 8 watch, 2 elevated and 19 high (anchored to 2026-09-30). Every chip's one-line "why" names the driving events — and names as "latest" only what is already in force, counting anything an issuer has announced for a future date separately, because a change that has not happened is not a cut this program has made. npm run qa recomputes every grade independently, and fails the build both when a rendered chip disagrees with this formula and when the day it was anchored to has drifted from the clock.

The honest caveat: past cuts are not future cuts. This is a track-record grade — how a program has actually treated its members, weighted for recency — not a forecast. A clean sheet can end overnight; Axis proved that in April 2026 with zero days' notice. Read it the way you'd read an issuer's history before signing up: as character evidence, not prophecy.