Editorial independence
Who publishes this.
The Water Journal is an editorial and data initiative established by Prana Spring, a bottled water brand sold in the United Arab Emirates.
No bottled water indexWe publish none, and no ranking of brands.
CorrectionsPublished in a log, never made silently. We have already corrected ourselves once, on Dubai.
No claim of being firstGlobal tariff datasets and city price comparisons exist. What does not exist is one that shows its arithmetic.
Append onlyA superseded figure is closed with a date and kept. No historical value is recalculated under a later methodology.
The corrections log
2026-08-16dubai · “AED 143.83” → “AED 143.85”. The scenario converted cubic metres to imperial gallons using the physical constant 219.9692. DEWA bills with 220 — the meter multiplication factor printed on the invoice, and the factor that turns each statutory per-gallon rate into the per-m³ rate DEWA publishes. Source precedence puts the billing formula above a physical constant. Corrected after obtaining Executive Council Resolution 16/2011.
2026-08-17 · re-stampall · No figure moved. “v1.0” → “v1.1”. Three frozen rules had stopped describing the engine that produces the figures. v1.0 Rule 3.2 required both headline figures to be published together; the site ships Dubai's water supply while holding its total water services. v1.0 §5 graded a disclosed scope the same as an unresolved assumption, which graded London and Dubai B; the engine grades on uncertainty alone and returns A. The v1.0 §10 fixtures asserted Singapore Grade A at S$ 35.15, London B and Dubai B with per-cubic-metre slabs and VAT on sewerage. None of that was what ran. The change policy in §0 requires an increment; it was not invoked at the time, so it is invoked now.
2026-08-17 · articlejudgement-goes-first — “the part of the game that fails first when water runs short is not the swing — it is the reading of the shot” → “the measure that changed most proportionally was not shot distance — it was the reading of the shot”. The published version asserted an order in which faculties fail as a water deficit deepens. Smith, Newell and Baker 2012 compared two hydration states at a single deficit; it cannot establish an order of failure, and the article's own headline claim went beyond its source. The claim is withdrawn. What the study supports — which measure had changed most in proportional terms by the time body-mass loss reached 1.5% — is stated instead, together with the sample size n = 7, which the published version omitted. No figure in the article changed.
2026-08-17 · articlethe-last-step-of-every-breath — “measurements have mapped perivascular pO2 falling away from the vessel wall, showing the drop-off as a physical fact rather than an inference” → “measurements have reported such gradients, but their magnitude is disputed because the phosphorescence technique can consume oxygen as it measures”. Tsai et al. 1998 was cited as settling the size of the oxygen gradient from vessel into tissue. Golub and Pittman 2008 show that the phosphorescence-quenching method can itself consume oxygen and so inflate an apparent transmural drop. The source is now cited for the existence of the gradient and not its size, the dissenting paper is added, and the animal model is named in the body text. No figure in the article changed.
2026-08-17 · re-stampall · No figure moved. “v1.1” → “v1.2”. Four rules added while extending the engine to ten new cities. §2.5 defines the reference connection, which turns Tokyo's 20 mm and Seoul's 15 mm from assumptions barring Grade A into printed scope. §2.6 fixes one year length at 365.2425 with two narrow exceptions. §5.4 requires both a payable and a structural price where temporary relief is live, and puts the structural one in the basket. §7.6 forbids an unknown component defaulting to zero: every component is observed, confirmed_absent with a source, or unresolved. The index is renamed to the TWJ Urban Water Tariff Index.
2026-08-17newyork · No figure moved. “no fixed component modelled, unstated” → “unresolved — publication withheld”. Rule 7.6, introduced the same day, asked what every silent zero in the base was. New York's record models no fixed component at all. The Water Board Rate Schedule is understood to set minimum charges by meter size for metered accounts, and that has never been read from the archived schedule. Either it sets none for a residential account — confirmed_absent, with the page cited — or it does and must be modelled even where it does not bind at 180 m3 a year.
2026-08-17 · re-stampall · No figure moved. “v1.2” → “v1.3”. Rule 9.3, the two-person review, is withdrawn at the publisher's direction and replaced by a publication checklist of seven machine-checkable conditions. The review had one reviewer in the whole system: it held three fully evidenced cities indefinitely while catching nothing, and eight of ten cities showed withheld to a reader when only one was genuinely blocked on evidence.
2026-08-17singapore · “S$ 1.42 / m3” → “S$ 1.40 / m3”. A transcription error. PUB's Annex A prints 1.40 for the block above 40 m3 and the component's own label said 1.40 while the stored rate said 1.42. Found by the v1.3 verbatim check on its first run.
2026-08-17london · No figure moved. “published” → “withheld”. London has been published since collection without ever being checked against a figure its publisher produced. Every other city reaching a publishable grade has at least one such reconciliation; London had none and nobody had looked. Found by the v1.3 publication checklist on its first run.
2026-08-17 · re-stampall · No figure moved. “v1.3” → “v1.4”. §5.3 rewritten. The old rule showed every refusal publicly with its reason, which across ten cities produced a catalogue that was mostly the word withheld — and presented our own unfinished capture as though it were a property of the city's water system. A refusal is now public only where its reason is itself a verified characteristic of the tariff system, the regulatory framework or the source record. Three publication statuses and a required withholding_origin field.
2026-08-17 · re-stampall · No figure moved. “v1.4” → “v1.5”. The publication gate treated three levels of proof as one. Source validity and calculation reproducibility are ours to require; external reconciliation is not always available, because a utility publishes a tariff rather than a worked example of somebody else's benchmark. The v1.3 checklist made it a gate condition, which amounted to holding that an official tariff is insufficient until the utility computes our scenario. Blockers are now classed material or validation gap, and only a material blocker holds a number. Validation is reported beside the figure.
2026-08-17 · re-stampall · No figure moved. “v1.5” → “v1.6”. Three normative additions, each demanded by a city already in the set. §3.8 tariff state: where a tariff selects between published rates on the measured state of the system, the applicable rate is the rule plus a dated authoritative state observation, and a tariff state is structural rather than a temporary policy adjustment. §3.8a trigger provenance: a trigger must come from the instrument governing that tariff. §7.6 gains non_standardizable, because unresolved and known-but-not-standardizable were sharing one word. §5.5 separates publication from index eligibility.
2026-08-17No figure moved. “median of standardized bills converted to USD” → “100 x exp(sum of w_i ln(P_structural_i,t / P_structural_i,base)) in local currency”. The previous construction was wrong in a way collection quality could not repair: a water tariff index must not move because a currency moved. With no tariff changing anywhere and the dollar down ten per cent, a USD-based index reports a ten per cent change in the price of water. Each relative is now a ratio of one city's currency to itself, so there is nowhere for an exchange rate to enter. Geometric because the quantities are ratios: the arithmetic mean of a doubling and a halving is 1.25, reporting a 25% rise where nothing happened.
2026-08-17newyork · No figure moved. “withheld” → “published”. The minimum-charge question that withheld New York is closed. The Water Board sets a minimum for meter-billed customers of USD 0.49 per day for water service, plus wastewater at 159% of water charges, frozen since FY2017 and maintained for FY2027. It is not a meter-size charge, which was the hypothesis the city was withheld on — wrong about the form and right that something existed. Modelled as a floor on the bill rather than a component of it, and not_applicable at the reference volume where usage of USD 340.08 a year exceeds the floor of USD 178.97.
2026-08-17 · re-stampall · No figure moved. “v1.6” → “v1.7”. Three changes the collected cities proved rather than suggested. §2.7: the observation unit is a tariff jurisdiction, not a city — Riyadh and Jeddah are two display cities on one national schedule, and a metropolitan market with several providers is the inverse. §3.9: the price curve is the primary data object and 15 m3 its canonical reference point, because Hong Kong crosses every other city between 5 and 25 m3, Perth and Sydney change places in one currency, and New York looked flat and is not. §7.7: the component ontology is enumerated and a minimum bill is a floor rather than an item.
2026-08-17seoul · No figure moved. “withheld” → “published (supply metric)”. The VAT question that withheld Seoul is closed on the supply side by statute: art. 26(1)2 of the Value-Added Tax Act exempts tap water, and Basic Ruling 26-0-1 defines it as water supplied through conduits by a supplier under the Water Supply Act — which attaches the exemption to Arisu. It was never going to be closed by the silence of a rate table, and Tokyo is why: its table shows no tax line either while its formula multiplies by 1.10.
2026-08-17sydney · No figure moved. “withheld” → “published”. The first city whose applicable rate required two pieces of evidence rather than one: the tariff rule from Sydney Water and a dated state observation from WaterNSW, the storage authority. Greater Sydney storage is 92%; the drought rate engages below 60%, so the normal rate of A$3.41/kL applies.
2026-08-18No figure moved. “tests read dist/site.json” → “tests import the payload from scripts/site.js”. npm run build is `node --test && node scripts/site.js && node scripts/render.js`, so the tests run before anything is built, and dist/ is not checked in. test/cross-currency-display.test.js and test/jurisdiction.test.js read dist/site.json anyway. On a clean checkout both failed on a missing file, which means the suite could only pass where a build had already succeeded and CI, which checks out clean on every push, failed on its first step and had been failing on order rather than on anything in the data.
2026-08-18No figure moved. “manifest.completed carried grade and held_by” → “manifest.captures carries city, date and note; grade and holds are computed”. completed was an append-only log printed by npm run list as a statement of current state. Both readings sat in the file at once: Seoul and Sydney each appeared twice with contradictory lines — Sydney as grade B held by two rules and, four lines later, as grade A held by nothing — and five entries were still held by Rule 9.3, the two-person review withdrawn in v1.3. The generated list closed with a paragraph asking for a second reader nobody wanted, and nothing in the repository could notice, because a hand-written sentence about a rule is checked by nothing.
2026-08-18No figure moved. “v1.7” → “v1.8”. Abu Dhabi raised the question and Hong Kong answered it. Research on the Abu Dhabi tariff proposed that a document dated on the valuation date be required to establish that a captured rate was applicable. Hong Kong water rates have not been revised since 16 February 1995, so no 2026 Hong Kong tariff document exists and none is expected. A rule requiring one would exclude a tariff on the ground that its publisher has had no reason to reissue it, which measures publishing frequency rather than evidence. The practice of treating an open-ended tariff as valid until superseded was already applied to Hong Kong, Dubai, New York and Singapore and had never been written down, so it could not be checked and could not be argued with.
2026-08-18No figure moved. “Rule 13 covered evaluative claims about prices” → “Rule 13.1 covers evaluative claims about evidence as well”. Rule 13 forbade calling a city cheap, efficient or best on the strength of a small bill, and covered half the exposure. The v1.8 document as first written contained a section headed "The weakest instance is named" about Abu Dhabi, and a correction entry describing one record as the best-evidenced in the Index while others were the least stable. Each statement was accurate and each was about TWJ's own capture rather than about any water system, and each was still wrong to publish: the grammatical subject was a place name, and the place name is the part that leaves the document when somebody quotes it.
2026-08-18No figure moved. “v1.8” → “v1.9”. Rule 13.1 as written in v1.8 governed "anything TWJ ships". The check that enforces it reads the Index records, the corrections log and the methodology document, and has never read an article, so the rule claimed a reach its own guard did not have — the same defect this project keeps correcting, in the direction that is hardest to notice, because a rule that overclaims looks stricter rather than looser. The choice was to widen the guard or narrow the rule, and the wider rule was wrong on the merits: the Index is an instrument and the journal is journalism. A comparative adjective attached to a city inside an Index record is the instrument editorialising about its own subject; an article is an argument, and an argument forbidden from characterising anything is not an argument.
2026-08-18No figure moved. “How we check — Two readers, a primary document, an archived copy with a hash” → “the publication checklist, counted from the constants”. The public methodology page described the two-person review, withdrawn in v1.3, and it survived six versions to v1.9 because the sentence was typed onto the page instead of read from the rule. Everything the repository could check agreed with the rule; the one line no test read did not, and it was the line a reader actually sees. The same defect had already been found in the capture manifest, in three city records and in two tests — this was its fifth site.
A restatement that leaves every figure unchanged is logged as a re-stamp rather than a correction, so the correction count stays a measure of wrong figures rather than of wrong descriptions. Both are published. An article whose claim went further than its source is a correction even when no number moved — the overreach is the error.