What changed, and when

Updates

A plain-language record of what changed on this site and when. Data on this site is revised as official releases land, and corrections are made in public rather than quietly — so this page exists to make both visible. Each entry corresponds to a published version of the site; the full technical detail, including internal refactors not worth listing here, is in the repository's CHANGELOG.

Subscribe by RSS Browse the analysis library →

2026-08-15 · v2.47.0

The explainers get a front door, and a walkthrough slide

GDP prices the rent on a flat you own and puts nothing against the meal cooked in it. That now runs through the paper, the methodology, the appendix, the summary, the citizen guide, the policy brief and the analyses, not just one page.

Added — A page at someperspective.info/care on the production boundary: what the national accounts count as production, and what they leave outside. The accounts work out a rent for the flat you own and add it to output. They put nothing against the meal cooked in it. The site had one sentence on unpaid work and nothing behind it.
Data — The National Statistical Office's two Time Use Surveys, from 2019 and 2024, are now in the dataset. Among people aged 15-59 who do it at all, women spent 289 minutes a day on unpaid domestic work in 2024 against 88 minutes for men — about 1,758 hours a year that appear in no employment statistic and no output figure.
Added — Two survey rounds allow the change to be measured rather than asserted, and the result is not a simple story either way. Men's participation nearly doubled, from 26.1% to 45.8%. But minutes fell for both sexes, so the gap among participants went from 202 minutes to 201 — one minute in five years. Over the same period women's participation in paid employment rose from 21.8% to 25.0%: more women in paid work, with the unpaid work not meaningfully redistributed.
Added — Published Indian valuations of unpaid household work run from 7.5% of GDP to 36%, a factor of nearly five driven almost entirely by the choice of valuation method rather than by measurement error. The page publishes the full range with each method named and declines to pick a number. What it does claim is narrower: every one of those estimates is large, and the reason none can be called correct is that no statistical agency has been asked to settle it.
Changed — This bears on a figure used elsewhere on the site. Female labour force participation of about 21% is usually discussed as an absence of work, while the Time Use Survey records 83.9% of women aged 15-59 doing unpaid domestic work on the reference day. Both are true. Together they show the participation figure measures an absence of work inside the boundary. That is now stated where the employment numbers are used.
Note — No figure on this site has been adjusted. Every growth series remains the official one on the official boundary. The claim is that the boundary is a choice, that India already collects the hours needed to value what sits outside it, and that a framework which imputes rent for a dwelling but records zero for the care performed inside it embeds a judgement about whose work counts as production. The research paper carries the same argument as Section 6.11.
Changed — The argument now sits in every document rather than on one page. Methodology gains a section under Limitations. The technical appendix gains A4b, with both Time Use Survey rounds tabulated by sex and the five published valuations listed with their methods. The executive summary states before its findings that the growth figures exclude this work. The citizen guide explains it in plain language. The policy brief asks for the specific thing that would fix it: a household production satellite account, using the survey hours India already collects. The media kit carries it as a quotable fact. The paper has it as Section 6.11.
Added — A new standing analysis, 'GDP counts the rent on your flat, not the cooking', with its own permalink. The library is now 37 pieces, and the count on the site is read from the data rather than typed in, so it stops going stale.
Changed — The walkthrough's limits slide now names this too. Every growth figure in the deck sits on the official boundary and none is adjusted.
Fixed — Both long-form explainers were badly buried. The GDP page was reachable from one sentence inside another section and a single footer link, and neither page was in the navigation at all. There is now an Explainers menu in the top navigation and a card near the top of the opening section carrying both. The list is held as data, so the next explainer appears in both places from one line.
Added — The guided walkthrough gains a slide on it, placed straight after the inflation slide so a reader reaches the verdict holding both caveats about what the headline figure does and does not contain. The deck is now eighteen slides. The bullet that had been added to the limits slide is removed, since the point now has a slide of its own.
Fixed — The slide count and the analysis count were typed into the page in several places and had already gone stale twice. Both now come from the data.

2026-08-15 · v2.44.0

Breathing room on phones, and a footer that dates itself

The header was crushing its own subtitle into six lines on a phone, text sat on the very edge of the screen, and three copyright lines disagreed with each other.

Fixed — On a phone the header stayed a horizontal row, so the title and its subtitle were squeezed into about half the screen while the theme controls took the rest. The subtitle wrapped to six lines. The header now stacks below 640 pixels: the title gets the full width and the controls sit on their own row underneath. The subtitle is down to three lines.
Fixed — Text was sitting 16 pixels from the edge of a phone screen, which reads as content jammed against the bezel. The page gutter is now 20 pixels below 640 pixels wide. It costs 8 pixels of line width and buys a visible margin. Tablets are unaffected and keep their wider gutter.
Fixed — Three copyright lines said three different things: 2025-2026 on the main site, 2025 in the executive summary, and 'May 2026' plus a retired project title in the research package. All three now agree. The main site's line also writes its own end year and reads the data vintage from the dataset, so it cannot go stale on the first of January or drift after a data refresh.

2026-08-15 · v2.43.0

A dedicated page on why inflation looks better after 2014

The one measure that improved now gets argued out at length, on its own page and in a new section of the paper, instead of being handled in a paragraph.

Added — A standalone explainer at someperspective.info/inflation. Consumer price inflation is the single cross-era measure on which India performs better after 2014 — 4.98% against 8.05% — and a project that reports eight findings in its own direction and disposes of the ninth in a sentence has not earned the eight. The page sets out six factors in order of how much each explains, then the strongest case that can be made against its own conclusion, then what evidence would change it.
Added — One of those six is new to this project and cuts against the government rather than for it. Central excise duty on petrol was ₹9.48 a litre and on diesel ₹3.56 in 2014. Between November 2014 and January 2016 it was raised on nine separate occasions — ₹11.77 a litre on petrol, ₹13.47 on diesel — for the stated purpose of moving the gains from collapsing world oil prices into the exchequer rather than to the pump. So the disinflation households experienced was the residual after the tax take, and to the extent fiscal policy touched fuel prices in that window it acted to keep them higher than world prices would have delivered.
Added — The research paper gains Section 6.10, which makes the same argument at full length with two sub-sections: the strongest case against the paper's own reading, and what would falsify it. A consistent measure of core inflation excluding food and fuel would settle much of it; this project does not have one going back to 2004 and says so rather than implying the test has been run.
Changed — The one-line thesis, which appears in four places on the site and ends 'the one honest exception is inflation', now links straight to the explanation.

2026-08-15 · v2.42.0

A state-level view, on 2026 population rather than the 2011 census

The national picture hides most of what matters. There is now a section that answers 'what about my state?' for sixteen states, and representation is measured on the government's own 2026 projection instead of a fifteen-year-old census.

Added — A new 'Your State' section under Data. Choose a state and it reports where that state sits on per-capita income, how many people stand behind each of its Lok Sabha seats, how many rupees come back for every ₹100 of Union tax it pays, and how many seats it is projected to gain or lose if constituencies are ever redrawn — with a sentence you can quote and three charts placing it against the other fifteen.
Added — People per MP is computed here for the first time, and it is the sharpest number in the section. Because seats were frozen on the 1971 census, representation now varies by a factor of 1.64 across these states: 1.67 million people behind each seat in Kerala against 2.74 million in Rajasthan. A vote is not worth the same everywhere in India, and this says by how much.
Changed — State information was previously scattered across three separate blocks — an income and seats table in one place, tax-return ratios in another, delimitation projections in a third — none of which could be read together. They are now one table, so a reader can see all four facts about a state at once.
Note — Two honest limits are stated in the section itself. Every population figure is from the 2011 census, because that is the most recent one India has — so representation is being measured with a fifteen-year-old ruler, which is this project's own argument arriving somewhere it cannot be avoided. And the tax-return figures come from an analyst affiliated with the same party body the author holds a fellowship in; the direction is well established, the exact ratios are contestable, and the section says so.
Data — Representation is now computed on projected 2026 population rather than the 2011 census. The figures come from the National Commission on Population's 'Population Projections for India and States 2011-2036' (Report of the Technical Group, Registrar General of India), Table 21, which gives the projected population of every state as on 1 March 2026. This matters: on 2011 numbers the gap between the best- and worst-represented state here was 1.64 times, and on 2026 numbers it is 1.85 times — 1.81 million people behind each seat in Kerala against 3.34 million in Rajasthan. A frozen seat allocation keeps compounding, so using a stale population understated the problem. The 2011 figure is kept alongside each state so the change is visible.
Note — There is an irony worth naming. The best population figure available for 2026 is a projection, and it is a projection precisely because the 2021 census was never held. Measuring whether a vote is worth the same across India now depends on an estimate rather than a count. That is this project's central argument turning up in the one place it cannot be argued around.

2026-08-15 · v2.40.0

The walkthrough gets a front door, and readable type on a phone

The guided walkthrough was reachable only from a card in the last section and one footer link; it is now in the navigation and on the landing page. Text on phones was rendering at caption size and now reads normally.

Fixed — Measured on a 390-pixel screen, 94.5% of all visible text on the front page was rendering at 12 or 14 pixels, and only 0.4% at 16. An earlier accessibility pass had raised a 9-pixel floor to 12, which stopped text being unreadable but left the whole page at the size normally used for captions. The scale now moves up a step on phones: body text reads at 16 pixels, paragraphs in the smaller style at 15, and secondary labels at 13.5. Chart labels, table cells and the tab bar are deliberately left where they were, because their size holds the layout together. After the change, 88% of visible text is 15 pixels or larger and no paragraph anywhere renders below 15.
Fixed — On a 320-pixel screen — the narrowest phone still in common use — the header row was 24 pixels wider than the display, which pushes the whole page sideways. It now wraps at that width. Checked at nine screen widths from 320 to 1440 pixels across all fourteen sections: no horizontal overflow anywhere.
Fixed — The stylesheet was requested without a version stamp, so a returning visitor could pair newly published markup with a stylesheet up to ten minutes old. It is now stamped with the site version, and a build check fails if the two ever drift apart. The guided walkthrough's data requests are stamped the same way, so the deck cannot draw yesterday's numbers from a cache.
Removed — The shareable PNG of the three indices, added in 2.37.0 for a social-media launch, has been withdrawn along with the post text. The chart itself remains on the site, where it is drawn live from the dataset and can be exported at any time from the Three Indices section.
Added — The guided walkthrough now has a place in the top navigation and a card at the top of the opening section. It is the shortest honest route through the argument — seventeen slides, about ten minutes, every figure drawn live from the dataset — and until now it was reachable only from a tile buried in the last section and a single footer link.
Correction — This page previously said that every entry here maps to a tagged release. The repository has no tags, so that was not true. The sentence now says what is actually the case: each entry corresponds to a published version of the site, with the full technical detail in the repository's changelog.
Changed — The repository's README was three versions and one project title out of date. It now carries the current version, the declaration of interests, links to the walkthrough and paper, and a table of the checks that run on every change.

2026-08-15 · v2.39.0

A declared interest, corrected V-Dem inputs, a loader that recovers, and a sitewide audit

The author's political affiliation is now stated on the site rather than denied. Three errors in this project's own figures are corrected, including a headline growth rate that had been quoted eight ways. And a full audit of every page, on desktop and phone, fixed what it found.

Correction — The footer previously said 'No political affiliations.' That was false. The author is a Manmohan Singh Fellow of the All India Professionals' Congress, a department of the Indian National Congress, and this project compares a Congress-led government with its successor. The claim is replaced by a declaration of interests, stated on the site, in the media kit, in the guided walkthrough and in the paper's Declarations. The research remains unfunded and no party body commissioned, reviewed or approved it — but that is a separate fact from neutrality, and the site no longer conflates the two.
Correction — Two of the Democratic Quality Index's four components are V-Dem series. The table computing the index held approximated rather than published values for both, and the approximations were too high in the UPA decade — a Liberal Democracy value of 0.555 for 2014 against a published 0.488, and a civil-society value of 0.87 against a published 0.669. Because the error inflated the earlier baseline and left the later years roughly right, it exaggerated the measured decline. It ran in this project's own favour. Both columns are now the published V-Dem series verbatim. The UPA-decade DQI average falls from 0.59 to 0.57 and the 2014 value from 0.54 to 0.49; 2026 is unchanged at 0.29, and the eight-of-nine cross-era result is unaffected.
Correction — The international comparison had its 2014 column on V-Dem's Electoral Democracy Index and its 2026 column on the Liberal Democracy Index — two different instruments read as one series. India's decline was shown as 0.71 → 0.26; the like-for-like figure on the Liberal Democracy Index is 0.488 → 0.260. All five countries are now on the same index throughout. India's proportional fall drops from 63% to 47%, and Turkey's is now the steeper of the two; India retains the largest absolute fall in the cohort. The multi-country chart was also labelled 'DQI', which was wrong: the DQI is constructed for India alone and cannot be computed for the others. It is now labelled as what it plots.
Added — A build check that fails unless the international table and the index's component table report the same V-Dem figure for India — which they can only do if both are on the same V-Dem index. The earlier consistency check could not have caught this, because it compares published documents against the dataset and both figures were in the dataset.
Fixed — The page was starting itself twice. Alpine calls a component's init automatically, and an extra x-init called it again: two requests for the dataset, two sets of event listeners, two chart initialisations. Because the 'live data unavailable' flag latched once set, a single dropped request out of the two was enough to raise that warning on a page whose data had in fact loaded.
Fixed — The dataset request now retries three times with backoff, and the warning banner carries a working retry button that clears itself if the retry succeeds. Previously one failed request left the banner up for the rest of the session with no way back.
Fixed — The site's main page is a single large file. If the transfer was cut part-way through — a dropped mobile connection, an intermediary that truncates — the page came up looking almost normal but with computed figures silently blank: an empty scorecard, a sentence reading 'On [nothing] cross-era measures'. There is now a check that detects a page which failed to start and says so, with a reload button, instead of leaving blanks.
Fixed — Index figures quoted in prose had drifted apart between sections: the SSI appeared as both 8.2 and 6.4, the FCI as both 0.68 and 0.80, the DQI as 0.40 in one place and 0.29 in another, and graduate unemployment as 27% and 28.5% against a dataset value of 26.5%. Every one of these is now read from the dataset at page load rather than typed into the page, so they cannot drift again.
Fixed — The animated timeline opened on 2014, where the SSI is 0.0 — so the first thing the panel showed was 'SSI Score: 0', which reads as a broken figure rather than a result. It now opens on the latest year, states which year it is showing, and explains what a zero SSI means: the index counts documented suppression episodes, and none had begun by 2014.
Correction — The headline growth figure was wrong. 'India's GDP grew at 6.8% annually since 2014' appeared in eight places across the site and the documents, and it is not what the published series says on any basis. The arithmetic mean for 2014-2026 is 6.1%, which is what the research paper has always reported — so the site's front page and its own paper disagreed. This error ran in the government's favour rather than this project's, but it is an error either way. Every instance now reads 6.1%, with the ex-pandemic figure of 7.2% given alongside it so the 2020 contraction is not quietly doing the work in either direction. The UPA comparison figure is corrected from 7.7% to 7.1% on the same basis.
Correction — The executive summary quoted India's press-freedom rank of 161 and labelled it 2024. 161 is the 2023 figure; 2024 was 159. The metric panel now carries the latest year throughout — 157 for 2026 — and the consistency check has been widened to catch value-before-year phrasing, which is the ordering that had been hiding this one.
Changed — The same work was being cited five different ways across the site, with two different titles, two different years and a period of 2014-2025 that stopped a year before the data does. Everything now cites one form: Raman, V.S. (2026), Some Perspective: India's Political Economy, 2004-2026. The research paper keeps its own distinct title, as it should.
Fixed — A sitewide audit ran every page at desktop and phone width, checking for script errors, failed requests, broken links, duplicate identifiers, sideways scrolling, skipped heading levels, images without alternative text and touch targets below the minimum size. It found the methodology page scrolling sideways on a phone (a long formula and an unbreakable URL), contents links and tags too small to tap reliably, and heading levels skipping from h2 to h4 in four documents. All are fixed. Ninety-eight of a hundred page-and-screen combinations now come back completely clean; the two remaining are elements that are empty or hidden by design.
Fixed — Seven of the downloadable documents had no description, no canonical URL and no social-preview tags, so sharing a link to any of them produced a bare URL with no title or image. All seven now carry them.
Fixed — The retired research presentation declared the walkthrough as its canonical address while actually redirecting readers to the paper. It now points both at the paper.

2026-08-15 · v2.37.0

Correcting how we read the inflation result

The one measure that improved was being described too generously. It is not evidence of good policy, and the paper now says so.

Correction — Earlier wording called the fall in inflation 'the single most defensible economic achievement of the period' and attributed it to an institution being strengthened. That reads as crediting the government, and it does not survive scrutiny. The comparison straddles a global commodity cycle: Brent crude fell from about $115 a barrel in June 2014 to under $30 by January 2016 — around 75% — driven by US shale supply and OPEC defending market share, in an economy importing roughly 86% of its crude. A shock that size, arriving exactly at the boundary between the two periods, explains more of the gap than any change of government.
Changed — Where domestic policy did contribute, the mechanism was inflation targeting — a rule that constrains the executive's discretion. If that is the channel, it supports the project's argument rather than qualifying it. And the 2026 figure sits on the reweighted CPI basket, so it prints lower for the same price movements.
Added — A shareable chart of the three indices, 2004-2026, is now available as a PNG.

2026-08-15 · v2.36.0

Accessibility: touch targets and readable text on phones

Buttons, links and controls that were too small to tap reliably on a phone, and text as small as 9px, are fixed.

Fixed — Around forty interactive elements were smaller than the 44-pixel minimum on a phone — chart download buttons and theme toggles at 28 pixels, footer links only 14 pixels tall, filter chips at 27, and the trajectory sliders with a touch area just 4 pixels high. All now meet the minimum on touch devices.
Fixed — Text rendering as small as 9px on phones is raised to a 12px floor. No text on the site now renders below 12px on a phone.
Changed — These changes apply only to touch devices and narrow screens. The desktop layout keeps its existing density — a mouse is precise, a fingertip is about ten millimetres.
Note — Links inside a sentence and the tap-to-explain glossary terms are deliberately left at text size. Enlarging them would break the line spacing of every paragraph containing one, and the accessibility guidelines specifically exempt targets positioned by the flow of text.

2026-08-15 · v2.35.0

The theory chapter moves into the paper; two presentations retired

Original theoretical work that only existed inside a slide deck is now part of the research paper, and the two hand-maintained presentations have been withdrawn.

Added — The paper gains a section setting out where this work sits in the literature and what it adds: a refinement of modernisation theory (growth can legitimise erosion, not only produce democracy), an extension of Levitsky and Way's competitive authoritarianism toward incremental and legal erosion, and a proposed concept — 'fiscal authoritarianism', the use of federal fiscal architecture to centralise power while leaving the constitution formally intact. It also gains a research agenda: state-level variation, sectoral and social disaggregation, district-level indices, and comparative work.
Changed — The two presentation documents have been retired. Their content is now either in the paper or in the guided walkthrough, both of which stay in step with the dataset. The old links still work and redirect rather than breaking.
Changed — The paper's policy section now carries the fuller set of immediate and structural reforms, stated as directions the measurements point in rather than as costed proposals.

2026-08-15 · v2.34.0

A guided walkthrough, and a responsive audit

A new ten-minute guided tour of the evidence, and fixes for layout problems that made several documents unusable on a phone.

Added — A guided walkthrough at /walkthrough/ — seventeen slides taking you from the question through the evidence to what follows, with seven live charts. Keyboard, swipe and click navigation, and each slide has its own link. Every figure is read from the dataset when the page loads, so the deck cannot fall out of step with the data the way a slide deck normally does.
Fixed — Tables in the downloadable documents had no way to scroll on a narrow screen, so on a phone they pushed the whole page sideways — by more than 400 pixels in the technical appendix. Tables now scroll inside their own container.
Fixed — The Economic Trajectory sliders had a fixed width that overflowed the smallest phones.
Changed — The site was audited at eight screen widths from 320 to 1920 pixels across every page type. Results and remaining gaps are recorded in the repository.

2026-08-15 · v2.33.0

The research paper, and a correction to a headline inequality figure

The full working paper is published, and an error in our own summary documents — one that ran in this project's favour — is corrected in the open.

Added — 'Growth Without Accountability: Measuring India's Institutional and Distributional Transformation, 2004-2026' — a 15,000-word working paper, available to read online or as a 43-page PDF. It documents the measurement problem, specifies how the three indices are built, reports the results, tests them against an independent study, and states the limitations at length.
Correction — Four of this project's summary documents stated the 2014 top 1% income share as 15%, which made inequality appear to rise by 7.6 percentage points to 2023. That figure is wrong: 15.0% is the 2014 bottom 50% share, and the two series had been crossed. The World Inequality Database series this project follows has 21.3%, making the rise 1.3 points. The documents are corrected and the framing rewritten around the two facts that are both true and stronger — the top 1% share is at its highest level since 1922, and the bottom half of the population lost roughly an eighth of its share of national income in twelve years.
Changed — An automated check now compares every year-labelled figure in the published documents against the source dataset on each change, so a contradiction of this kind cannot pass silently again.
Added — The research package now delivers what it advertised: a complete ZIP with every document, the paper, the dataset and the replication code, plus CSV and Excel exports.

2026-08-15 · v2.31.0

An independent check, August data, and three rendering fixes

Added an external study that reaches this project's governance conclusion by a different method, refreshed the data to August 2026, and fixed three display bugs — one of which had been hiding a chart's data entirely.

Added — An independent-corroboration section in Methodology covering Grier & Grier (2026), 'Promises, Promises'. They build a synthetic-control counterfactual over 1984-2013; this project builds composite indices from observed series. Different method, convergent conclusion — with the points of disagreement stated openly rather than smoothed over.
Data — CPI 2026 updated to 4.45% (MoSPI July 2026). Unemployment re-cited to the June 2026 PLFS bulletin — the headline 5.5% was unchanged.
Correction — MoSPI replaced the CPI 2012=100 index with 2024=100 from the January 2026 print, reweighting on HCES 2023-24. Food's weight fell from 45.86% to 36.75%, so headline CPI now responds less to a food-price shock than it used to. The 2026 figure is not on the same basis as 2014-2025 and is now labelled as such rather than silently spliced onto the older series.
Fixed — The Era Comparison detail chart had been rendering with no data line at all — a colour rule matched none of its points and made the whole series transparent. It now draws correctly, split into NDA-I and NDA-II.
Fixed — On phones the whole site was zooming out because one wide element stretched the page to roughly three times the viewport width. All fourteen tabs now fit a phone screen properly.
Fixed — Era shading and policy-event markers on the Interactive Data chart were never drawing.
Security — The two third-party scripts the site loads are now pinned with integrity hashes, so a substituted file on the CDN cannot execute.

2026-07-07 · v2.30.1

July data refresh

Routine refresh against the public releases available in early July 2026.

Data — Unemployment 2026 revised 5.1% to 5.5% (PLFS Monthly Bulletin, May 2026). CPI 2026 set to 3.93%.
Changed — Stale-date detection in CI widened, and the build's checkout action pinned to a single version across both workflows.

2026-06-18 · v2.30.0

Stylesheet precompiled; runtime CDN retired

The site stopped compiling its stylesheet in the browser on every visit and now ships a prebuilt one, with a guard so it cannot silently fall out of date.

Changed — Tailwind is compiled ahead of time into a committed stylesheet instead of being generated at page load. Faster first paint and one fewer third-party script.

2026-06-18 · v2.28.0

Indices rebuilt: COVID undercount, NSO merger, GST

The three constructed indices were rebuilt from a single canonical script so both political eras are computed identically.

Changed — The Statistical Suppression Index now includes the COVID mortality undercount and the NSO merger; the Fiscal Centralisation Index now treats GST as a structural shift in state taxing power rather than only counting cesses.
Method — Both the UPA and NDA decades are now generated by the same script, so no difference between the eras can come from how they were computed.

Full technical detail for every release, including internal refactors not listed here, is in CHANGELOG.md.