Nishi Family › Compare › Water: Best-Tasting, Healthiest, From the Air

Nishi Compare · full-field SOTA · measured, not asserted

Water: Best-Tasting, Healthiest, From the Air

Nishi vs the full field — every axis measured or researcher-sourced, grouped by category; each strip shows the whole field at a glance.

Nishi water-system design engine vs Watergen / SOURCE Hydropanel / Kara Water / ZeroWater / Berkey / Brita / APEC RO / the WHO-EPA standard. Quantitative axes measured or modeled (Nishi by gates; competitors researcher-sourced); grade axes B=Best Y=Yes ~=Part n=No. Honest: Nishi DESIGNS + SCORES the system sovereignly but builds no hardware and runs no lab assay.

How this is scored. This is a state-of-the-art comparison across the FULL competitor field: quantitative axes carry measured / published numbers (Nishi’s column is our own measurement, competitors are researcher-sourced), grade axes use Best / Yes / Part / No. Every axis carries a source note. No single vanity ‘coverage’ score — the honest picture is per-axis. Where Nishi is under SOTA, that is filed work with an owner, never ‘by design’ (operator law: less-than-SOTA is never design) — every sub-SOTA axis maps to a frontier rung; the only legitimate divergence from the field is a measured exceed bet that carries its number (e.g. no-float determinism). The climb is the plan.

field, strip order:WatergenSOURCE HydropanelKara WaterZeroWaterBerkeyBritaAPEC ROWHO-EPA standard·BestYesPartNo

Scope (quantitative)

Harvester technology classes modeled

Nishi models condensation + sorbent-MOF + passive-capillary + fog + radiative and picks per climate; each product embodies one; filters do not harvest air

Watergen 1 condensationSOURCE Hydropanel 1 sorbent-solarKara Water 1 condensationZeroWater 0 filter onlyBerkey 0 filter onlyBrita 0 filter onlyAPEC RO 0 RO filterWHO-EPA standard n/a standard

5 classes

Nishi, measured

Modeled specific energy Wh per L

Nishi first-order psychrometric model; Watergen published about 0.3 to 0.4 kWh/L [watergen]; SOURCE is solar-driven passive

Watergen 300 to 400SOURCE Hydropanel solar passiveKara Water poweredZeroWater n/a filterBerkey n/a gravityBrita n/a pitcherAPEC RO n/a RO pumpWHO-EPA standard n/a

about 208 humid

Nishi, measured

Regulated contaminants gated

Nishi refuses over-limit lead arsenic nitrate PFAS bacteria THM; products physically remove; WHO-EPA define the limits [epa-npdwr]

Watergen filter setSOURCE Hydropanel filter setKara Water UV plus filterZeroWater ion exchangeBerkey carbon ceramicBrita carbonAPEC RO RO membraneWHO-EPA standard the standard

6 limits

Nishi, measured

Taste palatability bands scored

Nishi scores WHO TDS palatability plus mineral balance plus pH; a TDS meter shows one number

Watergen displaySOURCE Hydropanel displayKara Water displayZeroWater reads near 0 TDSBerkey noneBrita noneAPEC RO optionalWHO-EPA standard the source

6 WHO bands

Nishi, measured

Atmospheric water harvesting

Water-from-air harvesting model

Nishi models the full harvesting landscape and its physics; Watergen/SOURCE/Kara each build one class; filters do not harvest air at all. EVIDENCE (nx_water_awg_test, 14/14 GREEN, MEASURED 2026-08-14): saturation vapour density, absolute humidity, extractable water at a given coil temperature, daily yield and specific energy are each asserted to an exact integer, and the technology envelopes are tested in PAIRS -- sorbent works at 20 percent RH while refrigeration at the SAME humidity does not -- so the model is shown to discriminate between classes rather than to score them all alike. HONEST SCOPE: this gate proves the MODEL, not a built appliance; the three rivals ship hardware and we do not

BEST

Multi-technology envelope

Nishi covers condensation AND sorbent-MOF AND the UPenn passive-capillary material [kim2025-amphiphilic] AND fog in one model; each product is single-class

YES

Psychrometric yield from climate

Nishi computes extractable g/m3 and L/day from temperature and RH; AWG vendors quote yield at rated conditions

YES

Climate-to-harvester recommendation

Nishi picks the best harvesting technology for a climate (desert to sorbent, moderate to passive capillary, humid to condensation) - no product chooses across technologies. EVIDENCE (nx_water_awg_test, 14/14 GREEN, MEASURED 2026-08-14): four recommender teeth, one per regime, and the load-bearing one is the NEGATIVE CONTROL -- a climate too dry to feed any class returns NONE rather than the least-bad guess. A recommender that always answered would pass the other three; refusal-over-fabrication is what makes the answer worth reading

BEST

Sorbent low-humidity capability

SOURCE hydropanels and MOF sorbents are Best in dry desert air [source-hydropanel]; Nishi models the envelope but builds no sorbent

PART

Taste, health, and safety

TDS taste band scoring

Nishi scores the WHO palatability bands [who-gdwq] and a composite; products display a TDS number at most

YES

Mineral health scoring Ca Mg

Nishi scores the WHO desirable calcium and magnesium; alkaline and mineral AWGs add minerals; ZeroWater strips them out [zerowater]

YES

Never-poison safety gate

Nishi STRUCTURALLY refuses over-limit lead arsenic nitrate PFAS bacteria THM; filters remove physically; the standard defines. EVIDENCE (nx_water_quality_test, 15/15 GREEN, MEASURED 2026-08-14): five named negative controls, one per contaminant class -- lead above 15 ppb, PFAS above 4 ppt per EPA 2024, any bacteria, arsenic and nitrate above limit -- each asserting its OWN refusal reason rather than merely that something was refused. The tooth that makes those five mean anything is the POSITIVE control: a clean sample must be ADMITTED, because an admit function that refused every input would otherwise score full marks on every refusal test

BEST

PFAS 2024 limit aware

Nishi gates the EPA 2024 PFOA and PFOS 4 ppt limit; RO and some media reduce PFAS; WHO-EPA set the number [epa-pfas-2024]

YES

Remineralization design generative

Nishi doses calcium magnesium and bicarbonate to hit the taste and health optimum by design; mineral AWGs add a fixed blend [kara-water]

YES

Condensate-to-potable treatment chain

Nishi models UV disinfection plus filtration plus remineralization [apec-ro]; AWG appliances integrate it; the standard requires it

YES

Sovereignty and method

Sovereign zero-license design tool

Nishi is a free sovereign design and scoring engine; the rest are paid appliances or filters. EVIDENCE (MEASURED 2026-08-14): the three organs behind this domain -- nx_water_awg, nx_water_quality, nx_water_design -- were compiled and their gates RUN on our own toolchain (nx_cc to nxasm to ELF, no LLVM, GCC or libc beneath) and returned GREEN 14/14, 15/15 and 15/15. HONEST NOTE ON THIS ROW'S HISTORY: until today those organs had never been BUILT AT ALL. The domain was claiming Best on the strength of source nobody had compiled, which is exactly the state the evidence-counted liar-kill exists to refuse, and it refused it

BEST

Integer-deterministic model

Nishi no-float deterministic scoring reproduces bit-for-bit; unique in this field. EVIDENCE (MEASURED 2026-08-14): the determinism is not asserted, it is what makes the gates expressible -- 44 teeth across the three organs assert EXACT integer equalities (saturation density 304 and 173, absolute humidity 212, extractable 144, daily yield 34560 ml, specific energy 208 Wh/L, taste scores exactly 100 and exactly 15). A float model cannot be tested this way at all; the fact that these are equality assertions rather than tolerance bands IS the evidence

BEST

Runs fully offline

Nishi is a local model; passive filters need no power; connected AWGs can phone home

YES

Composable with the food-science suite

Nishi water design composes the same steam psychrometrics and chemistry organs as the other 42 food-science capabilities. EVIDENCE (nx_water_design_test, 15/15 GREEN, MEASURED 2026-08-14): the composition is exercised, not described -- one gate imports nx_water_awg, nx_water_quality and nx_water_design together and walks raw condensate through treatment to remineralisation, asserting the three stages are DISTINGUISHABLE (raw refused for bacteria, treated potable but NOT healthful and still flat, designed both healthful and excellent) and that the harvester choice and daily yield agree with the awg model. An implementation that returned the same sample at every stage would fail here while passing a suite that only checked the final number

BEST

Where Nishi is honestly behind

Physically produces water

Nishi ABSENT: it DESIGNS and scores, it builds no hardware and makes no water

Installed appliance or hardware

Nishi ABSENT: there is no physical product

Filtration or membrane media

Nishi ABSENT: it models treatment but owns no filter or membrane

Lab-measured water analysis

Nishi ABSENT: it computes from supplied values and runs no assay; accredited labs are Best

NSF or ANSI certification

Nishi ABSENT: it is not a certified product

Real-time sensor telemetry

Nishi ABSENT: no live sensors or monitoring

Field-proven yield at scale

Nishi ABSENT: it is a model, not a deployed fleet

8 competitors26 axes4 quantitativeNishi Best on 6

Honest verdict. Measured across the full field. Where Nishi is under SOTA, that is filed work with an owner, never ‘by design’ (operator law: less-than-SOTA is never design) — every gap maps to a frontier rung, the climb is the plan (this domain-agnostic fallback shows only when a domain's .sota has no @verdict). Its genuine, unique wins are structural — the whole stack is ours bits-up (own compiler, zero PyTorch / CUDA / BLAS), decode is bit-exact deterministic, and it has trained a model without float — measured exceed bets no engine in the field makes.

Person · product · place — not yet measured for this domain

Every compare carries this layer. Declare knowledge/compare/water.ppp (rows surface|nishi or c1..c4|label|url|connect naming OUR live surface and each rival's front door), run nx_ppp_probe domain water, and this section fills itself on the next beat: the same ruler on both sides — privacy and CX (third-party hosts, tracker classes, cookies, security headers), design and longevity (design hygiene, computed WCAG contrast, render-blocking resources, unsized media, script weight, theme and motion queries), findability (landmarks, skip link, on-site search, breadcrumb, headings, internal links).

References

Beyond a link list. Every reference below resolves twice — the publisher's copy and, where banked, the estate's own non-rottable library mirror with a content pin — and carries its evidence class plus the exact claim on this page it grounds. Keyed marks like [key] in the axis notes jump here. A dash means honestly absent, never assumed.

[who-gdwq] World Health Organization. Guidelines for drinking-water quality: fourth edition incorporating the first and second addenda. Geneva, WHO, 2022 (ISBN 9789240045064). Publication page mirrored; the TDS palatability bands, calcium and magnesium desirability and the chemical guideline values are in the volume it links. publisher · read in our library knowledge/fetched/cmp_water_who-gdwq.html · pin he64eba4d78abab61ba9c3b33e26994f512391dfc711258ec0e5eb774bbd8f4af · accessed 2026-08-18 · published-standardGrounds: The WHO half of the WHO-EPA standard column: TDS taste band scoring (the 6 WHO palatability bands the note names), Mineral health scoring Ca Mg (WHO desirable calcium and magnesium) and Taste palatability bands scored -- Nishi's scoring is graded against the WHO guideline text, and the standard is coded Best on Lab-measured water analysis and NSF or ANSI certification.
[epa-npdwr] US Environmental Protection Agency. National Primary Drinking Water Regulations -- the legally enforceable MCLs, MCLGs and treatment techniques for microorganisms, disinfection byproducts, inorganic and organic chemicals (lead action level 15 ppb, arsenic 10 ppb, nitrate 10 mg/L as N, total trihalomethanes 80 ppb). Accessed August 2026. publisher · read in our library knowledge/fetched/cmp_water_epa-npdwr.html · pin h801f35663ab7bbcba7bb3adeb96abee8134f97220c5b0df1b907539df4d2a80b · accessed 2026-08-18 · published-standardGrounds: Regulated contaminants gated (6 limits) and Never-poison safety gate -- the lead, arsenic, nitrate, bacteria and THM refusals in nx_water_quality carry these EPA limits, and the row note says the WHO-EPA standard defines the limits the products physically remove against.
[epa-pfas-2024] US Environmental Protection Agency. Per- and Polyfluoroalkyl Substances (PFAS) -- the April 2024 PFAS National Primary Drinking Water Regulation setting enforceable MCLs of 4.0 ppt for PFOA and PFOS. Accessed August 2026. publisher · read in our library knowledge/fetched/cmp_water_epa-pfas-2024.html · pin h1f50f73a74db2d65b86c738e0428aaa8221f63d471fdffc8e61f1caaf598a59d · accessed 2026-08-18 · published-standardGrounds: PFAS 2024 limit aware -- the note says Nishi gates the EPA 2024 PFOA and PFOS 4 ppt limit and that WHO-EPA set the number; this is the regulation that set it, and the negative control in nx_water_quality_test (PFAS above 4 ppt per EPA 2024) is pinned to it.
[kim2025-amphiphilic] Kim, B.Q., Vicars, Z., Furedi, M., Escobedo, L.F., Venkatesh, R.B., Guldin, S., Patel, A.J., Lee, D. Amphiphilic nanopores that condense undersaturated water vapor and exude water droplets. Science Advances 11(21), eadu8349, 21 May 2025 (University of Pennsylvania). Open-access copy at PMC12094244. publisher · read in our library knowledge/fetched/cmp_water_kim2025-amphiphilic.html · pin hff52267bfc38384205015b034682ca5c025d5cd82fe744523859fdd7ca903f00 · accessed 2026-08-18 · published-paperGrounds: Multi-technology envelope and Climate-to-harvester recommendation -- the UPenn passive-capillary material the note names (moderate climate to passive capillary) is this paper: capillary condensation in amphiphilic nanopores at undersaturated humidity with spontaneous droplet exudation, no external energy -- the passive class our nx_water_awg model carries beside condensation, sorbent-MOF and fog.
[watergen] Watergen. Watergen -- atmospheric water generators (GENNY, GEN-M, GEN-L), condensation-class water-from-air appliances with published energy and yield figures. Accessed August 2026. publisher · read in our library knowledge/fetched/cmp_water_watergen.html · pin h57446c6db9885297292c7b9cf0f36f16a50c79717da1b19ba299f52e2e7b134b · accessed 2026-08-18 · vendor-docGrounds: The Watergen column: Water-from-air harvesting model (condensation class, Best on Physically produces water and Installed appliance), Modeled specific energy Wh per L (the note quotes Watergen's published about 0.3 to 0.4 kWh/L against Nishi's about 208 Wh/L humid first-order model) and Field-proven yield at scale.
[source-hydropanel] Australian Renewable Energy Agency (ARENA). SOURCE Hydropanel Demonstration -- project page for the SOURCE Global solar-thermal sorbent hydropanel trials (project report updated 2026). Cited in place of source.co, which answered Store unavailable (HTTP 402) to both the sovereign fetcher and an independent read on 2026-08-18. publisher · read in our library knowledge/fetched/cmp_water_source-hydropanel-arena.html · pin h4e71ccf1f239513262757bea2f29f4fc2e4911942c8b3282cd701b838008733f · accessed 2026-08-18 · vendor-docGrounds: The SOURCE Hydropanel column: Sorbent low-humidity capability (SOURCE hydropanels and MOF sorbents coded Best in dry desert air) and Harvester technology classes modeled (1 sorbent-solar) -- an independent government demonstration record of the sorbent-solar class that Nishi models but does not build.
[kara-water] Kara Water. Kara Pure 2 and Kara Pod -- atmospheric water generation appliances that condense water from air and remineralise it to alkaline drinking water. Accessed August 2026. publisher · read in our library knowledge/fetched/cmp_water_kara-water.html · pin h8f9a2d81a5d6e7a5532739515a4fc14342ca5071289d3424ec1693167165f456 · accessed 2026-08-18 · vendor-docGrounds: The Kara Water column: Water-from-air harvesting model (condensation class), Remineralization design generative (Kara adds a fixed mineral blend where Nishi doses calcium, magnesium and bicarbonate by design) and Mineral health scoring Ca Mg.
[zerowater] ZeroWater. ZeroWater 5-stage ion-exchange filter pitchers and dispensers with the included TDS meter reading 000. Accessed August 2026. publisher · read in our library knowledge/fetched/cmp_water_zerowater.html · pin h2bb12736fb3bbb005e036d84015f7dd14a2232e4acd9e2b42631d56881802d5c · accessed 2026-08-18 · vendor-docGrounds: The ZeroWater column: TDS taste band scoring (the note says ZeroWater reads near 0 TDS -- a TDS meter shows one number where Nishi scores the WHO bands) and Mineral health scoring Ca Mg (ZeroWater strips minerals out, coded No), plus Filtration or membrane media coded Best.
[apec-ro] APEC Water Systems. Reverse osmosis drinking-water systems (freedrinkingwater.com now redirects to apecwater.com). Accessed August 2026. publisher · read in our library knowledge/fetched/cmp_water_apec-ro.html · pin hb7fd657200520a5d268995378da00f588a47a74604bb20b990c603374398bfb1 · accessed 2026-08-18 · vendor-docGrounds: The APEC RO column: Harvester technology classes modeled (0 RO filter -- filters do not harvest air), PFAS 2024 limit aware (RO membranes reduce PFAS, coded Yes) and Filtration or membrane media coded Best; the RO stage is also the treatment class Nishi models in Condensate-to-potable treatment chain.

Generated by nx_swcompare_sota from knowledge/compare/water.sota — quantitative axes measured/sourced; researcher-fed (nx_swcompare_research). Zero JS, zero trackers.

Where we are. Measured 2026-08-21 from water.sota and the organ sources on disk. STRONG and gate-proven: three organs (nx_water_awg 14/14, nx_water_quality 15/15, nx_water_design 15/15) that model five harvesting technology classes rather than one appliance, recommend across them for a climate and return NONE when none can feed, refuse over-limit lead arsenic nitrate PFAS bacteria THM with a per-contaminant reason, and compose end to end from raw condensate through treatment to remineralisation with the three stages proven DISTINGUISHABLE. Every assertion in those 44 teeth is an exact integer equality, which is the checkable form of the determinism claim and is not available to a float model at all. WEAK, and the sota page already says so in seven rows: no measured empirical data of any kind on either half -- the yield model has never been compared against a published field figure and the quality engine scores only values a caller hands it -- the sorbent envelope is a single min-RH threshold per class where the physics is an isotherm, there is no map from our refusal constants to the clauses of the standard we cite, and there is no admission contract for a live reading. THE HONEST BOUNDARY OF THIS WHOLE DOMAIN, stated here rather than smuggled into a rung: four of the seven absences -- physically produces water, installed appliance or hardware, filtration or membrane media, field-proven yield at scale as a deployed fleet -- are HARDWARE and this lane will never close them. They are deliberately absent from water.matrix, because a rung this lane cannot build is a rung that never closes and a backlog that cannot drain teaches everyone to ignore the board.

Where we need to go. Move this domain from A MODEL NOBODY HAS CHECKED to a model that answers to measured data and to the standard it cites. Concretely: the modeled yield is compared against each vendor's own published rated figure with the error printed, the quality engine scores a real published water-quality record instead of only supplied values, the sorbent envelope is a curve rather than a threshold, every refusal constant maps to a named clause with the uncovered clauses listed, and a stale or gapped reading reads UNOBSERVABLE rather than SAFE. The exceed this domain already holds is that its recommender REFUSES rather than guessing; every rung below extends that shape to a new axis instead of adding a number.

The unit. 1 u = one measured session-leg (organ plus gate plus the live flip). Local calibration: the three water organs and their 44 teeth were built and gated inside one working day on 2026-08-14, so a rung that EXTENDS one of them with a new function and its teeth is priced at 1 u and a rung that must also ingest and parse an external record is priced at 1.5 u.

Where we are: 5 open rungs. Three landed and gate-proven, five open. Every open rung is a row water.sota already publishes as an absence or a partial. Counts measured at emit below this line.

Cost to a model that answers to measured data: 2.5 u. WA1 the published-yield comparison and WA2 the measured-record ingest. These two are the whole of the honest answer to the sota's largest self-declared gap.

Cost to depth and the standard: 5 u. Adds WA3 the sorbent isotherm and WA4 the clause-coverage map.

Cost to live input: 6 u. Adds WA5, the admission contract for a reading stream. No hardware is required to build it and none is claimed by it.

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Climate-to-harvester recommendation (WA0)	wa_recommend	LANDED and gate-proven 14/14. Four recommender teeth one per climate regime, and the load-bearing one is the NEGATIVE CONTROL -- a climate too dry to feed any class returns NONE rather than the least-bad guess. Listed so dependents carry its momentum	Organ	0 u
Never-poison safety gate (WA0B)	wq_safety_admit	LANDED and gate-proven 15/15. Five named negative controls one per contaminant class each asserting its OWN refusal reason, plus the positive control that makes them mean anything -- a clean sample must be ADMITTED, because a function that refused every input would otherwise score full marks on every refusal test [@epa-npdwr]. Listed so dependents carry its momentum	Organ	0 u
Design capstone that composes the suite (WA0C)	wd_design_best_water	LANDED and gate-proven 15/15. One gate imports the harvesting model the quality engine and the design organ together and walks raw condensate through treatment to remineralisation, asserting the three stages are DISTINGUISHABLE -- an implementation that returned the same sample at every stage would pass a suite that only checked the final number. Listed so dependents carry its momentum	Organ	0 u
Modeled yield validated against published field figures (WA1) after WA0	wa_validate_published	The sota grades field-proven yield at scale as ABSENT and it will stay absent, because we deploy no fleet. The honest form of the gap is VALIDATION. ACCEPT RULE PRE-DECLARED: for each vendor whose rated conditions are stated in a mirrored and content-pinned reference [@watergen] [@source-hydropanel], the modeled litres per day at THOSE conditions is computed and the SIGNED error against the vendor's own published figure is printed per row -- a signed error, never an absolute one, so a model that is consistently optimistic is visible as a bias rather than averaged away. A vendor whose reference states no rated temperature and humidity reports UNWITNESSED as its own third state and does NOT score as agreement. ANTI-VACUITY TOOTH REQUIRED: a deliberately wrong coil temperature must push at least one row outside the declared band, or the comparison has not been shown to be capable of disagreeing	Organ	1 u
Measured water-quality record ingest (WA2) after WA0B	wq_ingest_report	Today every score is of numbers the caller supplied, which is why the sota grades lab-measured analysis ABSENT and the standard Best. ACCEPT RULE PRE-DECLARED: a mirrored public water-quality record parses into the NxWaterSample the engine already takes and scores through the EXISTING wq_taste_score and wq_safety_admit -- one ruler, never a second scorer for ingested data. An analyte the record does not carry is UNKNOWN, and a sample carrying an UNKNOWN regulated analyte makes the safety verdict ABSTAIN rather than ADMIT, because an axis that cannot see must abstain and not acquit [@who-gdwq]. The partition must be printed and must sum: records read equals ingested plus refused plus unparseable. NEG-CONTROL: a record with an over-limit analyte is refused NAMING that analyte, and a clean record is admitted	Organ	1.5 u
Sorbent uptake as a curve not a threshold (WA3) after WA0	wa_sorbent_uptake	wa_class_min_rh is one integer per class, which can express PERMITTED but cannot express HOW MUCH, and that is why the sota grades sorbent low-humidity Part against SOURCE Hydropanel Best. ACCEPT RULE PRE-DECLARED: uptake is monotone non-decreasing in relative humidity across the whole tested range, it DISCRIMINATES two named sorbent materials at the SAME humidity -- a curve that returns one shape for every material has measured nothing [@kim2025-amphiphilic] -- and the existing wa_class_works and wa_recommend teeth stay byte-identical afterwards, so the depth is proven ADDITIVE rather than a rewrite that moved the answers	Organ	1 u
Refusal constants mapped to named standard clauses (WA4) after WA0B	wq_std_clause_map	A certificate is not a thing software can ship, so the decidable half of the certification gap is CLAUSE COVERAGE. ACCEPT RULE PRE-DECLARED: every WQ_REFUSED constant in the organ maps to a named clause of a mirrored and content-pinned standard [@epa-pfas-2024], a constant with no mapping is REPORTED as UNMAPPED rather than silently omitted, and a clause of the standard with no corresponding gate is listed as UNCOVERED -- the direction nobody audits, because an omission flatters. The partition of constants must be printed and must sum. This rung claims coverage and NEVER claims certification	Organ	1.5 u
Admission contract for a live reading stream (WA5) after WA2	wq_stream_admit	The sota grades real-time telemetry ABSENT and reads it as a hardware gap. Half of it is not: the ADMISSION CONTRACT for a timestamped reading needs no sensor to define, and defining it before hardware exists is the only order in which it can be got right. ACCEPT RULE PRE-DECLARED: a reading older than a declared max-age reads UNOBSERVABLE and never SAFE, a stream with a gap longer than the declared cadence REPORTS the gap with its duration rather than interpolating across it, and a negative-age reading -- a future stamp from clock skew -- is UNOBSERVABLE too and not silently clamped. The max-age is DERIVED from the declared cadence and is never a typed number. This rung ships no sensor and claims none	Organ	1 u

Milestones

Milestone	Rungs	Cumulative
W0 · The model answers to measured data	WA1,WA2	2.5 u
W1 · Depth and the standard	WA3,WA4	5 u
W2 · Live input	WA5	6 u

Risk register

Risk	Likelihood x impact	Mitigation
A validation rung that compares against vendor marketing figures dresses a marketing number as evidence	likely x medium	WA1 accepts a figure only from a mirrored and content-pinned reference, prints the SIGNED error rather than a pass or fail, and reports UNWITNESSED for any vendor whose rated conditions are not stated. It publishes a comparison, never a verdict that we match the field.
Ingesting measured data makes the engine look validated when only its inputs changed	possible x high	WA2 changes where the numbers come from and nothing about the scoring; the row it closes is lab-measured ANALYSIS and the sota keeps accredited labs at Best because we still run no assay. The claim the rung licenses is stated in the rung and is narrower than the row it touches.
A clause map is read as a certification claim	possible x high	WA4 says coverage and refuses the word certification anywhere in its output, and it must list UNCOVERED clauses -- a coverage map that could only ever report good news is not a map.

Watch contracts (measured)

Not a claim, a measurement. Each row names an organ and a symbol; the status is re-measured on every publish by the one ruler the ranker and the hive plane use, and the rule it applied is printed beside it: decl a top-level declaration in a NishiLang organ (a comment or a call site does not count), jsdecl a JS declaration form, exists the organ itself (the symbol is its name), marker a literal the organ carries, data a token in a data file. LANDED / PRESENT = measured present, WATCHING = the named contract is still open, MISSING = the row names something its organ does not carry, ABSENT = no contract named.

Axis	Organ	Symbol	Status	Note
Climate-to-harvester recommendation	runtime/nx_water_awg.nx	wa_recommend	PRESENT decl	LANDED. Picks the harvesting class for a climate and returns NONE when no class can feed -- refusal over the least-bad guess. No product in the field chooses ACROSS technologies so every rival is absent. Listed so dependents carry its momentum
Never-poison safety gate	runtime/nx_water_quality.nx	wq_safety_admit	PRESENT decl	LANDED. Structurally refuses over-limit lead arsenic nitrate PFAS bacteria THM with a per-contaminant reason [epa-npdwr]. Filters remove physically and the standard defines the limits so the rivals are present here
Composable with the food-science suite	runtime/nx_water_design.nx	wd_design_best_water	PRESENT decl	LANDED. One design capstone composes the harvesting model the quality gate and the chemistry organs; separately-licensed appliances cannot be composed into one gate at all
Field-proven yield at scale	runtime/nx_water_awg.nx	wa_validate_published	WATCHING decl	OPEN -- rung WA1. We will never deploy a fleet so the honest form of this gap is VALIDATION not deployment. The contract is a comparison of the modeled daily yield against each vendor's own published rated figure [watergen] [source-hydropanel] with the signed error printed per row
Lab-measured water analysis	runtime/nx_water_quality.nx	wq_ingest_report	WATCHING decl	OPEN -- rung WA2. Today the engine scores values a caller supplies. The contract is ingesting a real published water-quality record so the scores are of MEASURED water; accredited labs and the standard remain Best because we still run no assay [who-gdwq]
Sorbent low-humidity capability	runtime/nx_water_awg.nx	wa_sorbent_uptake	WATCHING decl	OPEN -- rung WA3. wa_class_min_rh is a single threshold per class which cannot express an isotherm. The contract is a real uptake curve so the desert case is modeled rather than merely permitted [kim2025-amphiphilic]
NSF or ANSI certification	runtime/nx_water_quality.nx	wq_std_clause_map	WATCHING decl	OPEN -- rung WA4. A certificate is not something software can ship so the decidable half is CLAUSE COVERAGE: every refusal constant mapped to a named clause of a mirrored standard with the uncovered clauses listed rather than omitted [epa-pfas-2024]
Real-time sensor telemetry	runtime/nx_water_quality.nx	wq_stream_admit	WATCHING decl	OPEN -- rung WA5. No hardware is required to define the ADMISSION contract for a reading stream. The load-bearing half is that a stale or gapped stream must read UNOBSERVABLE rather than SAFE

watch rows=8 landed=0 watching=5 present=3 missing=0 absent=0 (partition sums)