NZ Small-Community Food Self-Sufficiency — Dairy, Edible Fats, and Case Evidence
🔴 This source is a CLAIM REGISTER, not an evidence base. No figure from it may enter the model until its primary has been retrieved and verified.
This is the AI research synthesis that the 2026-07-22 batch was assembled to check. Every one of its primaries that could be retrieved was retrieved and verified independently; those primaries are the sources listed in
underlying_sources:and they, not this page, are what the model cites. The verification pass found a 3× error (a servings-per-day figure read as people-per-day — see OT_193) plus several mis-attributions and one uncited assumption (listed below).contested: trueis set for that reason. Its most durable contribution is its three declared negatives — the things it correctly reports do not exist.
Summary
An AI-generated research synthesis on food self-sufficiency for a small New Zealand community, structured around a set of prompts covering dairy at household and small-community scale, non-dairy edible oils, and New Zealand intentional-community case evidence. It is written for a roughly 30-household / 80-person community and presents “model-ready” figure tables with inline links to its sources.
It is ingested for two reasons, neither of which is its numbers. First, provenance: it is the document that generated this entire batch, and the corpus should record what was asserted so that later readers can see which assertions survived verification and which did not. Second, its negatives: a synthesis that says “no NZ source exists for X” is doing genuinely useful work, and three of its declared gaps held up.
What did not hold up is recorded in full below, because those are the errors the corpus exists to prevent.
Key claims
- claim: "FRAMING — this source is ingested as a CLAIM REGISTER, not an evidence base. Its assertions are recorded as 'what the synthesis asserts', and NO figure from it may enter the Neobiome model until its underlying primary has been retrieved and verified independently. All retrievable primaries behind its headline numbers WERE retrieved and verified in the 2026-07-22 batch and are listed in underlying_sources."
source_location: "CR_055 whole document; batch decision recorded in the project's internal notes"
- claim: "DECLARED NEGATIVE 1 (HELD) — there is NO NZ source for rendered TALLOW / LARD yield per carcass. The synthesis states its own instruction verbatim: 'Do not ingest as NZ coefficient yet.' [verified: the negative was checked and held]"
source_location: "CR_055 — edible fats section, rendered animal fat."
- claim: "DECLARED NEGATIVE 2 (HELD) — there is NO NZ source for a verified SUNFLOWER seed-yield figure. [verified: the negative was checked and held]"
source_location: "CR_055 — edible fats section, sunflower."
- claim: "DECLARED NEGATIVE 3 (HELD, BUT NARROWED AT INGEST) — no NZ community source quantifies diet share, land per person, or cooking-fat strategy. ⚠ NARROWED: Awaawaroa Bay DOES operate an olive grow-harvest-press pathway (URL_030), so the correct residual is that no NZ community source QUANTIFIES an oil output, not that no community has one. [verified and corrected at ingest]"
source_location: "CR_055 — case-evidence section; narrowing established against URL_030 (Awaawaroa Bay)."
- claim: "🔴 contested: true — VERIFICATION FOUND A 3x ERROR. The synthesis reports 98.26 as PEOPLE per day fed by a 192-bed vegetable plot. The underlying source's own Table 4 reads '98.26' as SERVINGS per day and states 'People per day ......... 33'. Had the 98.26 been applied, the vegetable land requirement would have been UNDERSTATED THREEFOLD. See OT_193. [verified error]"
source_location: "CR_055 — vegetable carrying-capacity figure; refuted against UC GEOG402 (2013) Table 4, in corpus as OT_193."
- claim: "🔴 FURTHER MIS-ATTRIBUTIONS AND UNCITED ASSUMPTIONS FOUND AT VERIFICATION, each corrected on the named source page: (a) a 90% milkfat RECOVERY fraction applied but NOT present in the FAO source cited and uncited anywhere (OT_173, RT_388); (b) FAO APHP 69 cited as 'FAO butter composition' when it is a village dairy COST-ACCOUNTING manual (OT_173); (c) an 80% screw-press oil recovery behind '806 kg recovered oil/ha' that no retrieved source supports, plus a probable paraphrase error in its '73%→80% recovery' phrasing (LIT_095, RT_391); (d) hazelnut cold-hardiness figures (-13°C dormant / -7°C open catkins) attributed to the Hazelnut Growers bulletin but belonging to the NZ Tree Crops Association hazel guide (OT_182); (e) a '4-day use-by' and 'heat to 70°C for 1 minute' attributed to the Raw Milk Regulations 2015, which contain neither (REG_031, RT_390); (f) a lifestyle-blog 'Jersey house cow 2,835 L' that is in fact a TWICE-A-DAY commercial yield mislabelled as a house-cow figure (LIT_090); (g) the Jersey NZ breed booklet used as a dairy-composition primary when it restates DairyNZ/LIC 2021-22 second-hand (OT_172 supersedes; booklet not ingested); (h) an omitted publication date on the RNZ Riverside article, whose figures are eight years old (URL_028). [all verified at ingest]"
source_location: "CR_055 whole document; each item refuted against the named in-corpus primary during the 2026-07-22 verification pass."Neobiome Intelligence relevance
Feeds d02_smart_food_systems as provenance and as a register of what was checked — NOT as a data source.
What survives from this synthesis, and where it now lives:
| Its assertion | Verified? | Where the model cites it from now |
|---|---|---|
| Dairy is the most quantifiable NZ fat pathway | ✅ upheld | ot_172_dairynz-dairy-stats-2023-24 + ot_171_dairynz-ff-ch4-cow-feed-2021 + lit_090_lopez-villalobos-2023-oad |
| Oilseed rape has a complete NZ pathway | ✅ upheld | lit_095_fasi-2012-osr-sowing-date + ot_183_pureoil-osr-grower-guide |
| Olives have NZ census yield data | ⚠ partly — per tree only | ot_180_olivesnz-2021-grove-census, ot_181_olivesnz-2024-grove-census |
| Hazelnuts have NZ yield + kernel-oil data | ⚠ partly — units incompatible | ot_182_hazelnut-growers-bulletin1, lit_096_savage-1998-nz-hazelnut |
| No NZ tallow/lard per carcass | ✅ held | RT_394 (opened) |
| No verified NZ sunflower seed yield | ✅ held | recorded as a gap below |
| NZ communities don’t quantify diet share / fat strategy | ✅ held, narrowed | url_030_awaawaroa-ecovillage-waiheke operates an unquantified olive pathway |
| Butter conversion at 90% fat recovery | 🔴 uncited | RT_388 — recovery term is empty |
| 806 kg recovered oil/ha | 🔴 unverified 80% | RT_391 — report oil-in-seed with a flagged band |
| 98.26 people/day from a 192-bed plot | 🔴 3× error — it is SERVINGS | ot_193_uc-geog402-2013-local-food-initiatives: 33 people/day |
The operating rule this page establishes: where the synthesis and a retrieved primary disagree, the primary wins, without exception. Where the synthesis supplies a figure whose primary was never retrieved (the 90% and 80% recovery fractions), the figure does not exist and must be carried as an explicit gap, not as a default.
No engine cell is auto-changed by this ingest, and no figure on this page is model-eligible.
Retrieval provenance
- Upstream sources: the synthesis carries inline URLs to its own primaries. Those that could be retrieved were retrieved and are listed in
underlying_sources:; each has its own verified page in this corpus. - Prepared by: an AI research tool. ⚠ The specific tool and the exact prompt were not recorded, and have not been inferred here. The document’s internal structure (“Prompt 1 — …”, “Prompt 2 — …”) indicates it was produced from a multi-prompt research brief, but the brief itself is not in corpus.
- Retrieved date:
2026-07-22is the date the file entered the corpus; it is not necessarily the date the synthesis was generated. The document cites a DairyNZ 2024/25 statistics release, so it postdates that. - Verification: NOT AI-extracted-and-trusted. Every retrievable primary behind its headline numbers was independently retrieved and checked in the 2026-07-22 batch, and the discrepancies are recorded above. This is the opposite of the usual
cr_posture — the synthesis is here to be audited, not consulted.
Notes
- Why ingest a source you do not trust. Three reasons. (1)
orphan == 0: the raw is filed undersource-files/02_compiled_research/and needs a page. (2) Provenance: without this page, a later reader cannot tell which of the batch’s 21 primaries were retrieved because of a specific assertion, or which assertions were refuted. (3) The negatives: its three declared gaps are real findings, and one of them (tallow/lard) is now RT_394. contradictions:lists the six pages that refute a specific claim of this synthesis — OT_193 (the 3× error), OT_180 (olive per-tree-only, not per-hectare), OT_173 (the FAO scope and the missing 90%), REG_031 (the 4-day / 70°C mis-attribution), OT_182 (the cold-hardiness mis-attribution), LIT_090 (the mislabelled house-cow yield).- The synthesis is not incompetent, and the distinction matters. Its structure, its region-by-region framing, and above all its willingness to declare negatives are good research behaviour. Its failure mode is uniform and predictable: it attaches real numbers to the wrong documents, and it fills missing coefficients with plausible-sounding assumptions that carry no citation. That is the exact failure mode that verifying every figure against its retrieved primary exists to catch, and this batch is its clearest demonstration to date.
- Vintage: cites a DairyNZ 2024/25 release; the batch verified against the 2023-24 statistics (OT_172), which is the edition retrieved. The synthesis’s 2024/25 figures (414 kg MS = 234 kg milkfat + 181 kg protein) are not verified in corpus and must not be used.
Research targets
Documents to retrieve
- RT_394 — NZ rendered tallow / lard yield per carcass. The synthesis’s own first declared negative, with its own instruction “Do not ingest as NZ coefficient yet.” Verified as a genuine gap. This is the missing animal-fat leg of the edible-fat model, alongside the dairy leg that this batch did source.
Research gaps
- Not raised as RTs (recorded here instead):
- A verified NZ sunflower seed-yield figure — the synthesis’s second declared negative, and it held. Sunflower is named in its edible-oil set but has no NZ yield behind it, so it joins walnut as a pathway blocked on the yield leg.
- The 2024/25 DairyNZ statistics edition. The synthesis cites it; the batch verified the 2023-24 edition. A later re-base would use the newer edition, but the difference is a one-season update, not a correction.
- The research brief / prompt behind this synthesis. Would complete the provenance record; not in corpus and not inferred.
Connections
Links to
Referenced by
Sources (4): OT_191 · OT_193 · URL_029 · URL_030
EDT domains (1): D02: Smart Food Systems & Agriculture
Cases (1): Tui Community