--- license: cc-by-4.0 language: - en pretty_name: Indian Credit Card Facts size_categories: - 1K. **The line this dataset draws.** It carries *facts* — what an issuer charges, what a card earns and where that earn is capped, what has changed and when, and the published transfer ratios between programmes — plus one realistic point value per card, because a reward rate read without its point value is the commonest way this data is misread. It does **not** carry CardAdvisor's *model*: editorial ratings, rupee valuations of perks and welcome bonuses, route-by-route point values, or valuations of airline and hotel currencies. Those are opinions, they change as our method improves, and they are published on the site where they can be read in context. Facts are open under CC BY; the model is shown, not shipped. --- ## Motivation **Why was this dataset created?** There is no public, structured record of Indian credit-card terms. Issuer pages state today's terms and nothing else; when a reward rate is cut or a cap introduced, the previous terms disappear from the web. Researchers, journalists and consumers who want to answer "what did this card pay last year, and when did that change" have had nowhere to look. Comparison sites hold the data but publish it as prose inside pages designed to sell applications, and none of them publish the history. **What questions is it meant to support?** Devaluation frequency and direction over time; notice periods between announcement and effect; the gap between headline reward rates and realised value once caps and exclusions apply; the structure of the points-transfer network; fee dispersion across issuers and tiers. **Who created it, and who funded it?** Devchandra Sah, operating as CardAdvisor (), an independent Indian platform. No issuer, network or bank funds, sponsors, reviews or has any input into it. The site carries affiliate links on some cards; commission status has zero weight in any published figure and no input into this dataset. --- ## Composition **What does each row represent?** - `cards` — one active credit card available to Indian consumers. Fields cover fees (joining, annual, waiver threshold, forex markup, monthly APR), stated eligibility (minimum income, age, CIBIL where published), reward rates by spend category with their caps and one point value per row, the perk types the card carries (lounge access, insurance, …), and network/tier classification. - `changes` — one recorded change to a card's terms, dated, with direction (devaluation / revaluation / neutral), magnitude, the metric affected, its before and after values, and at least one external source. Where an issuer announced a change ahead of its effective date, both dates are recorded. - `transfers` — one edge in the points-transfer network: a source programme, a destination programme, the ratio between them, and the tier gate, cap, transfer time and source where published. What a destination mile or hotel point is *worth* is not in this table. - `valuations` — one card's realistic point value in paise per point on its default redemption route, and the same for each reward programme. One number each; the route-by-route breakdown is on the site. **Is any of it personal data?** No. The dataset contains no personal data of any kind. It describes financial products, not people. Nothing in it derives from user behaviour, and the site's own recommendation engine stores no name, phone number, PAN, Aadhaar, salary figure or card number. **Is it a sample or the complete population?** `cards` aims at the complete population of actively marketed consumer credit cards in India, and is not a sample. `changes` is explicitly **not** complete: it is the set of changes we have found and verified against a source, which is bounded by what was publicly reported. Its coverage is strongest from 2023 onward and thins the further back it goes. Any analysis of change *frequency* over time must account for that, and the observation bias runs in the obvious direction — recent changes are better recorded than old ones. **What is deliberately absent?** Editorial judgement — the model. Scores, rankings, recommendations, "best card" verdicts and expert ratings; our rupee valuations of perks and welcome bonuses; route-by-route point values (cash, voucher, travel portal, transfer) and the "best route" they imply; and our valuations of airline and hotel currencies. Those are opinions, they belong on the site rather than in a dataset, and mixing them with verifiable facts would make the whole file harder to trust. The one model-derived number retained is a single realistic point value per card and reward row, kept because without it the reward-rate columns are actively misleading (see *Known risks of misuse*). No issuer artwork, logos or product photography are included or referenced. *Erratum, release v2026-08-15 (schema v2):* that release's files did carry several of the model columns listed above — `expertRating`, `welcomeBonusValueRs`, `perks[].estimatedAnnualValueRs`, the route-level `…PaisaPerPoint` values, `bestRouteName`, and the transfer graph's `valuePaisa` / `paisaPerPoint` — while this section said they were excluded. The datasheet stated the intent correctly and the files did not honour it. The release stands as published (a frozen file is not rewritten); the correction is this release, schema v3, and the erratum is recorded in the release index at . --- ## Collection process **Where does it come from?** Primary sources are the issuers' own pages, fee schedules and MITC documents. Indian banks are required to serve from the `bank.in` domain, and card records link to those URLs. Where a change predates our tracking, the event is reconstructed from contemporaneous reporting and marked `confidence: reconstructed` rather than presented as primary. **How is it verified?** Card records carry a last-verified date. Change events carry at least one external source and, where possible, an archived copy of it — we never cite our own site as the source of a real-world change. Corrections to our own data are explicitly **not** recorded as change events; only real-world changes are. **What is the known accuracy?** Published, and measured rather than asserted. For a fixed sample of 30 cards across 15 issuers we re-check every material field against the issuer's own page and publish the result, including the failures, at . On the 2026-08-15 pass: | Field | Backed by the primary source | Note | |---|---|---| | Joining fee | 24 / 24 (100%) | | | Annual fee | 24 / 24 (100%) | | | Fee waiver spend | 12 / 13 (92%) | | | Forex markup | 17 / 24 (71%) | | | Monthly APR | 3 / 22 (14%) | Published in the fee schedule, not the card page | | Stated minimum income | 8 / 20 (40%) | Many issuers never publish one | A further 36 field claims sit behind six issuer sites that cannot be machine-read at all — four render their figures with JavaScript, two refuse automated retrieval — and are counted separately rather than averaged in, because being unable to check something is not the same as checking it and coming up short. **Anyone using the APR column should treat it as the weakest in the dataset.** It is not corroborated by the card pages, and auditing fee schedules is the next planned pass. --- ## Preprocessing and derived values Most columns are transcribed, not computed. Three are derived, and the derivation is stated so it can be disagreed with: 1. **Point values** (`valuations`, and `pointValuePaisa` on each reward row) are a single source of truth per card, expressed in paise per point, based on the card's default redemption route rather than an issuer's own claimed value. This is the one model-derived number the open dataset keeps. 2. **Effective reward rate** = headline rate × point value. A card paying 5 points per ₹150 at 25 paise a point is 0.83%, not 3.33%, and the dataset carries both the raw rate and the point value so either can be recomputed. 3. **Reward caps** are applied once per reward row — monthly and annual caps are not compounded. No imputation is performed anywhere. A value the issuer does not publish is `null`, never a guess, and `null` in this dataset always means "not published", never "zero" and never "no requirement". --- ## Uses **Suitable for:** market-structure analysis, devaluation and notice-period research, fee and reward dispersion studies, teaching datasets for finance or data-journalism courses, and as a grounding source for question-answering about Indian credit cards. **Not suitable for:** determining what any individual will actually be charged. Terms vary by applicant, by channel, and by the issuer's discretion; APRs in particular are banded by credit score and the single figure here is the published headline. It is also unsuitable as a complete history of the market before roughly 2023, for the coverage reason above. **Known risks of misuse.** Reward rates read without their point values overstate returns by several multiples — this is the single most common error made with Indian card data, and it is why both columns ship together. Comparing `changes` counts across years without accounting for observation bias will show a spurious upward trend. --- ## Distribution - **Licence:** [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Free to use, including commercially, with attribution. - **Formats:** JSON and CSV for each table. - **Live endpoints:** — mutable by design; today's catalogue. - **Immutable releases:** — each dated release is frozen and never rewritten. Cite these, not the live endpoints. - **Field dictionary:** published alongside the data on and included in the JSON payloads. - **Where else it lives:** Zenodo — concept DOI (always the newest release; each release has its own DOI) · Hugging Face — . The same files, the same datasheet; the site's release index is the canonical copy. --- ## Maintenance Maintained by Devchandra Sah (). The catalogue is reconciled against issuer sources on a rolling basis; the change tracker runs daily and every event is adjudicated by a person before it is published. Releases are cut when the underlying data has moved materially. **Corrections are welcome and are dated in public.** Report one at ; changes to our own records appear in the public change log. Errata for a published release are recorded in the release index rather than by rewriting the release — a frozen file that changes is not a frozen file. --- ## Versions | Release | Schema | Note | |---|---|---| | `v2026-08-16` | 3 | Model columns removed (see *What is deliberately absent*). 280 cards, 1617 events, 153 edges, 233 valuations. | | `v2026-08-15` | 2 | First deposit. Carried model columns contrary to the datasheet — erratum recorded; superseded by v2026-08-16. | Each release keeps its own DOI; the concept DOI resolves to the newest. --- ## Citation > CardAdvisor (2026). *Indian Credit Card Facts — an open dataset*, release v2026-08-16. > https://cardadvisor.in/data/versions See `CITATION.cff` for machine-readable citation metadata.