# The herd ledger **Answer.** Uruguay's cattle books balance only if **2,300,543 head left the farms** between the 2024 and 2025 declarations: **18.95% of the opening herd**. Calves born (2,697,988) outran deaths (247,498), home slaughter (22,826) and the herd's own growth (+127,121). Notebook: https://cow-tools.sophia-labs.com/notebooks/herd-ledger/herd-ledger.ipynb · Page: https://cow-tools.sophia-labs.com/notebooks/herd-ledger/index.html · Data: Uruguay MGAP/SNIG DICOSE 2024–2025 (graph `uruguay-cows` at https://cow-tools.sophia-labs.com/mcp). ## Findings - Uruguay's cattle books balance only if **2,300,543 head left the farms** between the 2024 and 2025 declarations: **18.95% of the opening herd**. Calves born (2,697,988) outran deaths (247,498), home slaughter (22,826) and the herd's own growth (+127,121). - The July 2026 expedition's figure for Rocha (21.4%) carried the all-species bug; species-honest, Rocha's cattle offtake is **25.4%** of its opening herd. - By department the books run from **Treinta y Tres**, which shipped out 26.2% of its opening herd, to San José, 14.2%. Montevideo (a few thousand head) is left out of the extremes. ## The notebook A herd can be kept like a set of books. Opening stock, plus calves born, minus animals that died, minus animals slaughtered for the farm's own table, should equal closing stock, *unless animals left*. Whatever is left over is **offtake**: cattle sold to the meat-packing plants or exported on the hoof. DICOSE does not publish slaughter in this dataset, but it does publish the other four lines, so the fifth can be inferred. offtake = births − deaths − own consumption − (closing stock − opening stock) Opening stock is the 2024 declaration (30 June 2024); closing stock is the 2025 one; the flows (births, deaths, consumption) are those declared in 2025 for the year in between. This is the analysis that caught cow-tools out in July 2026. Births are only declared for cattle, but deaths and consumption are declared for every species; adding the sheep's deaths to the cattle's books made the extraction rate look 30% too low. The curated metrics are now species-honest (`bovine_births_count` and friends), and this notebook keeps to cattle by construction: species `1` and `11` only. ```python # Setup. import altair as alt import pandas as pd import cowtools_client as cw import cowtools_charts as cc names = cc.department_names() ``` ```python # The query: cattle stock and flows per department, both years. q = cw.sparql(""" SELECT ?year ?department (SUM(?n) AS ?stock) (SUM(?b) AS ?births) (SUM(?d) AS ?deaths) (SUM(?c) AS ?consumption) WHERE { GRAPH ?g { ?obs a cow:AnimalFlowObservation ; cow:year ?year ; cow:department ?department ; cow:species ?s ; cow:animalsCount ?n . VALUES ?s { } OPTIONAL { ?obs cow:birthsCount ?b } OPTIONAL { ?obs cow:deathsCount ?d } OPTIONAL { ?obs cow:consumptionCount ?c } } } GROUP BY ?year ?department """) q["code"] = q["department"].map(cw.dept_code) for c in ("births", "deaths", "consumption"): q[c] = q[c].fillna(0).astype(int) n = q.groupby("year")[["stock", "births", "deaths", "consumption"]].sum() n ``` ```text stock births deaths consumption year 2024 12138148 2364279 264496 23069 2025 12265269 2697988 247498 22826 ``` ```python opening, closing = int(n.loc[2024, "stock"]), int(n.loc[2025, "stock"]) b, d, c = (int(n.loc[2025, k]) for k in ("births", "deaths", "consumption")) offtake = b - d - c - (closing - opening) rate = offtake / opening ledger = pd.DataFrame([ ("opening stock, 30 June 2024", opening), ("+ calves born", b), ("− deaths", -d), ("− slaughtered for own consumption", -c), ("− closing stock, 30 June 2025", -closing), ("= offtake (left the farms)", offtake), ], columns=["line", "head"]) cc.show_table(ledger, name="ledger") ``` | line | head | |:----------------------------------|:------------| | opening stock, 30 June 2024 | 12,138,148 | | + calves born | 2,697,988 | | − deaths | -247,498 | | − slaughtered for own consumption | -22,826 | | − closing stock, 30 June 2025 | -12,265,269 | | = offtake (left the farms) | 2,300,543 | ### Cross-check against the source, and against the July oracle ```python for year in (2024, 2025): csv = cw.source_csv(year, "datosanimales.csv") bov = csv[csv["EspecieCodigo"].isin(["1", "11"])] for col, key in [("AnimalesPorEspeciePD_AD", "stock"), ("AnimalesPorEspecieNacimientosPD_AD", "births"), ("AnimalesPorEspecieMortandadPD_AD", "deaths"), ("AnimalesPorEspecieConsumoPD_AD", "consumption")]: cw.check(f"{year} cattle {key}: cube = CSV", int(n.loc[year, key]), int(cw.num(bov[col]).sum()), source=f"datosanimales.csv {year}, {col}") cw.check("offtake 2024 → 2025, head", offtake, 2_300_543, source="cow-tools WP-0 oracle (e), 2026-07-06") cw.check("extraction rate", round(rate, 4), 0.1895, tol=0.00005, source="cow-tools WP-0 oracle (e): 18.95%") ``` ```text PASS 2024 cattle stock: cube = CSV: 12,138,148 vs 12,138,148 (exactly) — datosanimales.csv 2024, AnimalesPorEspeciePD_AD PASS 2024 cattle births: cube = CSV: 2,364,279 vs 2,364,279 (exactly) — datosanimales.csv 2024, AnimalesPorEspecieNacimientosPD_AD PASS 2024 cattle deaths: cube = CSV: 264,496 vs 264,496 (exactly) — datosanimales.csv 2024, AnimalesPorEspecieMortandadPD_AD PASS 2024 cattle consumption: cube = CSV: 23,069 vs 23,069 (exactly) — datosanimales.csv 2024, AnimalesPorEspecieConsumoPD_AD PASS 2025 cattle stock: cube = CSV: 12,265,269 vs 12,265,269 (exactly) — datosanimales.csv 2025, AnimalesPorEspeciePD_AD PASS 2025 cattle births: cube = CSV: 2,697,988 vs 2,697,988 (exactly) — datosanimales.csv 2025, AnimalesPorEspecieNacimientosPD_AD ``` ```text PASS 2025 cattle deaths: cube = CSV: 247,498 vs 247,498 (exactly) — datosanimales.csv 2025, AnimalesPorEspecieMortandadPD_AD PASS 2025 cattle consumption: cube = CSV: 22,826 vs 22,826 (exactly) — datosanimales.csv 2025, AnimalesPorEspecieConsumoPD_AD PASS offtake 2024 → 2025, head: 2,300,543 vs 2,300,543 (exactly) — cow-tools WP-0 oracle (e), 2026-07-06 PASS extraction rate: 0.1895 vs 0.1895 (within ±5e-05) — cow-tools WP-0 oracle (e): 18.95% ``` A sanity bound from outside the dataset: the July 2026 expedition set this against INAC's figures for cattle slaughtered plus cattle exported live, roughly 2.2 to 2.5 million head a year. The inferred offtake sits inside that range: the declarations are honest enough that a stranger's arithmetic recovers the country's actual metabolism. ```python cw.finding("headline", f"Uruguay's cattle books balance only if **{offtake:,} head left the farms** between the 2024 and 2025 " f"declarations: **{rate:.2%} of the opening herd**. Calves born ({b:,}) outran deaths ({d:,}), home slaughter " f"({c:,}) and the herd's own growth ({closing - opening:+,}).") ``` > **Finding.** Uruguay's cattle books balance only if **2,300,543 head left the farms** between the 2024 and 2025 declarations: **18.95% of the opening herd**. Calves born (2,697,988) outran deaths (247,498), home slaughter (22,826) and the herd's own growth (+127,121). ```python water = pd.DataFrame([ ("opening 2024", 0, opening, "stock"), ("births", opening, opening + b, "in"), ("deaths", opening + b - d, opening + b, "out"), ("own consumption", opening + b - d - c, opening + b - d, "out"), ("offtake", closing, opening + b - d - c, "out"), ("closing 2025", 0, closing, "stock"), ], columns=["step", "start", "end", "kind"]) water["order"] = range(len(water)) wf = alt.Chart(water, title=alt.Title("Uruguay's cattle ledger, 30 June 2024 → 30 June 2025", subtitle="head of cattle (beef + dairy); offtake is inferred")).mark_bar(size=34).encode( x=alt.X("step:N", sort=alt.SortField("order"), title=None, axis=alt.Axis(labelAngle=0)), y=alt.Y("start:Q", title="head", axis=alt.Axis(format="~s"), scale=alt.Scale(zero=False, domain=[9_000_000, 15_500_000])), y2="end:Q", color=alt.Color("kind:N", scale=alt.Scale(domain=["stock", "in", "out"], range=[cc.CATTLE, cc.DAIRY, cc.SHEEP]), title=None), tooltip=["step:N", alt.Tooltip("start:Q", format=","), alt.Tooltip("end:Q", format=",")], ).properties(width=540, height=300) cc.show(wf, "ledger-waterfall", alt_text="Waterfall: 12.14 million opening stock, plus 2.70 million births, minus small deaths and home slaughter, minus 2.30 million offtake, gives 12.27 million closing stock.") ``` ![Waterfall: 12.14 million opening stock, plus 2.70 million births, minus small deaths and home slaughter, minus 2.30 million offtake, gives 12.27 million closing stock.](charts/ledger-waterfall.svg) Chart data and encoding: [charts/ledger-waterfall.vl.json](charts/ledger-waterfall.vl.json) (Vega-Lite). ## Department by department The same books per department. A department's offtake also includes cattle sold to *other departments* (store calves going west to be finished), so a breeding department's "extraction" is high and a finishing department's low. That is the herd-structure notebook's story told in flows. ```python dep = q.pivot_table(index="code", columns="year", values=["stock", "births", "deaths", "consumption"]) dd = pd.DataFrame({ "opening": dep[("stock", 2024)], "closing": dep[("stock", 2025)], "births": dep[("births", 2025)], "deaths": dep[("deaths", 2025)], "consumption": dep[("consumption", 2025)], }) dd["offtake"] = dd["births"] - dd["deaths"] - dd["consumption"] - (dd["closing"] - dd["opening"]) dd["rate"] = dd["offtake"] / dd["opening"] dd = dd.reset_index() dd["name"] = dd["code"].map(names) dd = dd.sort_values("rate", ascending=False) cw.check("department offtakes sum to the national offtake", int(dd["offtake"].sum()), offtake, source="same books, two groupings") # Every department's books, re-derived from the CSVs. a24 = cw.source_csv(2024, "datosanimales.csv") a25 = cw.source_csv(2025, "datosanimales.csv") b24 = a24[a24["EspecieCodigo"].isin(["1", "11"])] b25 = a25[a25["EspecieCodigo"].isin(["1", "11"])] g = lambda d, col: cw.num(d[col]).groupby(d["DepartamentoCodigo"]).sum() src_off = (g(b25, "AnimalesPorEspecieNacimientosPD_AD") - g(b25, "AnimalesPorEspecieMortandadPD_AD") - g(b25, "AnimalesPorEspecieConsumoPD_AD") - (g(b25, "AnimalesPorEspeciePD_AD") - g(b24, "AnimalesPorEspeciePD_AD"))) cw.check("department offtakes: cube = CSV (largest difference, head)", int((dd.set_index("code")["offtake"] - src_off.reindex(dd["code"]).values).abs().max()), 0, source="datosanimales.csv 2024 and 2025, cattle only") cc.show_table(dd[["name", "opening", "births", "offtake", "rate"]].query("name != 'Montevideo'"), floatfmt=".3f", name="department-ledger") ``` ```text PASS department offtakes sum to the national offtake: 2,300,543 vs 2,300,543 (exactly) — same books, two groupings ``` ```text PASS department offtakes: cube = CSV (largest difference, head): 0 vs 0 (exactly) — datosanimales.csv 2024 and 2025, cattle only ``` | name | opening | births | offtake | rate | |:---------------|-----------------:|---------:|----------:|-------:| | Treinta y Tres | 613051 | 143895 | 160917 | 0.262 | | Rocha | 688313 | 181180 | 174553 | 0.254 | | Cerro Largo | 957379 | 231065 | 216665 | 0.226 | | Paysandú | 809798 | 188421 | 179667 | 0.222 | | Maldonado | 230533 | 55728 | 47540 | 0.206 | | Lavalleja | 644438 | 149071 | 127726 | 0.198 | | Tacuarembó | 1.03458e+06 | 244087 | 199859 | 0.193 | | Río Negro | 555782 | 112883 | 106598 | 0.192 | | Artigas | 730165 | 168821 | 138861 | 0.19 | | Durazno | 900286 | 223829 | 157244 | 0.175 | | Salto | 849010 | 213043 | 146395 | 0.172 | | Flores | 393191 | 91966 | 67201 | 0.171 | | Colonia | 551213 | 95197 | 92680 | 0.168 | | Canelones | 355686 | 51287 | 58553 | 0.165 | | Rivera | 608407 | 142668 | 96461 | 0.159 | | Florida | 1.09523e+06 | 220853 | 169868 | 0.155 | | Soriano | 570050 | 91871 | 81530 | 0.143 | | San José | 548346 | 91865 | 78096 | 0.142 | ### A correction to our own July report The July 2026 expedition printed Rocha's extraction rate as 21.4%. Here it is 25.4%. The difference is the bug this notebook's introduction describes: the July figure subtracted *every* species' deaths and home slaughter (sheep included) from Rocha's cattle books. Done that way again, the arithmetic reproduces the July number exactly, which is how we know that is what happened. ```python all_deaths = cw.num(a25["AnimalesPorEspecieMortandadPD_AD"]).groupby(a25["DepartamentoCodigo"]).sum() all_cons = cw.num(a25["AnimalesPorEspecieConsumoPD_AD"]).groupby(a25["DepartamentoCodigo"]).sum() r = dd.set_index("code").loc["14"] contaminated = (r["births"] - all_deaths["14"] - all_cons["14"] - (r["closing"] - r["opening"])) / r["opening"] cw.check("Rocha, species-honest (this notebook)", round(r["rate"], 3), 0.254, tol=0.0005, source="the table above") cw.check("Rocha, with every species' deaths and home slaughter (the July bug)", round(contaminated, 3), 0.214, tol=0.0005, source="expedition 2026-07-05 printed 21.4%") cw.finding("correction", f"The July 2026 expedition's figure for Rocha (21.4%) carried the all-species bug; species-honest, Rocha's " f"cattle offtake is **{r['rate']:.1%}** of its opening herd.") ``` ```text PASS Rocha, species-honest (this notebook): 0.254 vs 0.254 (within ±0.0005) — the table above PASS Rocha, with every species' deaths and home slaughter (the July bug): 0.214 vs 0.214 (within ±0.0005) — expedition 2026-07-05 printed 21.4% ``` > **Finding.** The July 2026 expedition's figure for Rocha (21.4%) carried the all-species bug; species-honest, Rocha's cattle offtake is **25.4%** of its opening herd. ```python top = dd.query("name != 'Montevideo'").iloc[0] low = dd.query("name != 'Montevideo'").iloc[-1] cw.finding("departments", f"By department the books run from **{top['name']}**, which shipped out {top['rate']:.1%} of its opening herd, " f"to {low['name']}, {low['rate']:.1%}. Montevideo (a few thousand head) is left out of the extremes.") cc.show(cc.choropleth(dd, "rate", title="Inferred offtake as a share of opening stock, 2024 → 2025", subtitle="cattle that left each department's farms (to slaughter, export, or another department)", legend="offtake rate", fmt=".0%", domain=[0, 0.3]), "offtake-map", thumbnail=True, alt_text="Map of offtake rate by department; the breeding east ships the largest share of its herd.") print(cc.map_credit()) ``` > **Finding.** By department the books run from **Treinta y Tres**, which shipped out 26.2% of its opening herd, to San José, 14.2%. Montevideo (a few thousand head) is left out of the extremes. ![Map of offtake rate by department; the breeding east ships the largest share of its herd.](charts/offtake-map.svg) Chart data and encoding: [charts/offtake-map.vl.json](charts/offtake-map.vl.json) (Vega-Lite). ```text Map: Natural Earth 1:10m admin-1 (public domain, v5.1.2), joined on the DICOSE department code. Made with Natural Earth. ``` ## Where every number came from ```python cc.show_table(cw.checks()[["check", "got", "expected", "rule", "source"]], name="checks") cw.provenance() ``` | check | got | expected | rule | source | |:--------------------------------------------------------------------|:--------------|:--------------|:---------------|:-----------------------------------------------------------| | 2024 cattle stock: cube = CSV | 12,138,148.00 | 12,138,148.00 | exactly | datosanimales.csv 2024, AnimalesPorEspeciePD_AD | | 2024 cattle births: cube = CSV | 2,364,279.00 | 2,364,279.00 | exactly | datosanimales.csv 2024, AnimalesPorEspecieNacimientosPD_AD | | 2024 cattle deaths: cube = CSV | 264,496.00 | 264,496.00 | exactly | datosanimales.csv 2024, AnimalesPorEspecieMortandadPD_AD | | 2024 cattle consumption: cube = CSV | 23,069.00 | 23,069.00 | exactly | datosanimales.csv 2024, AnimalesPorEspecieConsumoPD_AD | | 2025 cattle stock: cube = CSV | 12,265,269.00 | 12,265,269.00 | exactly | datosanimales.csv 2025, AnimalesPorEspeciePD_AD | | 2025 cattle births: cube = CSV | 2,697,988.00 | 2,697,988.00 | exactly | datosanimales.csv 2025, AnimalesPorEspecieNacimientosPD_AD | | 2025 cattle deaths: cube = CSV | 247,498.00 | 247,498.00 | exactly | datosanimales.csv 2025, AnimalesPorEspecieMortandadPD_AD | | 2025 cattle consumption: cube = CSV | 22,826.00 | 22,826.00 | exactly | datosanimales.csv 2025, AnimalesPorEspecieConsumoPD_AD | | offtake 2024 → 2025, head | 2,300,543.00 | 2,300,543.00 | exactly | cow-tools WP-0 oracle (e), 2026-07-06 | | extraction rate | 0.19 | 0.19 | within ±5e-05 | cow-tools WP-0 oracle (e): 18.95% | | department offtakes sum to the national offtake | 2,300,543.00 | 2,300,543.00 | exactly | same books, two groupings | | department offtakes: cube = CSV (largest difference, head) | 0.00 | 0.00 | exactly | datosanimales.csv 2024 and 2025, cattle only | | Rocha, species-honest (this notebook) | 0.25 | 0.25 | within ±0.0005 | the table above | | Rocha, with every species' deaths and home slaughter (the July bug) | 0.21 | 0.21 | within ±0.0005 | expedition 2026-07-05 printed 21.4% | ```text kind target rows first line 0 sparql uruguay-cows 38 SELECT ?year ?department ``` ```python cw.save() ```