FarmNode Sector Evaluation — Source-Backed Research Baselines

01 — Why these evaluations were created

FarmNodes now consolidate source-backed laboratory research requests instead of equating every registered lab label with a completed experiment. The controller found six original request messages consolidated into four distinct needs: Granby natural resources data quality, agriculture predictive modelling, forestry anomaly detection, and emergency-response metadata reconciliation.

The Granby problem already has a mathematically sealed small-input oracle. The other three research needs required an initial assessment of real, dated source records before any independently scored generated algorithm could be claimed.

02 — Forestry historical baseline

A pinned forestry source snapshot contains 13 historical observations covering 2012–2024. Eight chronological holdout cases were scored against two simple forecasting baselines. In the record’s native measurement units, the naive baseline mean absolute error is 503,801.125 and the median-change baseline is 493,459.75. The median-change method was modestly better in this specific sample.

This is not an independent anomaly-detection score. No independently labelled anomaly truth set has been supplied, no forest-management decision has been authorized and source rights have not been separately certified.

03 — Agriculture historical baseline

A pinned agriculture source snapshot includes 18 observations between 1921 and 2021. Eight chronological holdout cases were evaluated. The naive baseline mean absolute error is 20,395.5 and the median-change baseline error is 27,592.25 in the original source units. The proposed median-change method therefore did not beat naive forecasting on this held-out sample.

This transparent negative result is useful: it shows the evaluation retains failures instead of presenting all candidate methods as improvements. It provides neither general predictive validity nor a proven commercial farming recommendation.

04 — Historical emergency-response metadata

The source contains 15 historical 2024 alert rows. Nine held-out records underwent timestamp parsing and schema consistency checks, and the nine checked timestamps were valid according to the recorded rules. The original fields include issue date/time, alert type, method, requesting jurisdiction, hazard and event context.

No independent truth labels for real emergency events, no risk-of-harm study and no operational authority were demonstrated. The system must never be described as dispatch-capable or as having independently validated public-warning decisions based on these checks.

05 — Interpreting 25 observations responsibly

Eight forestry + eight agriculture + nine emergency records equals 25 source-backed checks, not 25 generated algorithms. The related industry evaluation and protection suites passed five recorded tests, covering source pinning, chronological holdout separation and non-misrepresentation of these historical cases.

The open 25-task council benchmark is a different requirement: 25 separate, independently specified algorithm-invention or repair tasks, matched candidate comparisons, holdouts, executable validation, adverse tests and traceable permissions. Its qualified completion remains zero at this audit.

06 — What responsible researchers and sponsors should do

For each proposed lab task, first define who may supply the data, what the desired output represents, which source rights apply, what failure means and who is an appropriate evaluator. Keep source-based baselines separate from AI model proposals and from working runtime deployments.

For forestry, obtain independently labelled anomalies and professional forestry validation. For agriculture, define a precise predictive target, units, seasons and acceptable errors. For emergency response, acquire authoritative event labels and authorized public-safety domain review before testing operational claims. Failed or ambiguous results remain research records, not a product launch.

Neither these baseline results nor a locally correct algorithm grants intellectual property rights, permits public disclosure of third-party datasets, or generates a Bit Trade Global investment instrument.

07 — Audit trail and status

Source audit records: NIKI_FARM_ORIGINAL_SEALED_EVALUATORS_20261023.json, NIKI_FARM_SOURCE_BACKED_25_HOLDOUT_CASES_20261023.json and NIKI_THREE_SECTOR_HOLDOUT_EVALUATORS_TEST_20261023.log (18:19 UTC). Original-publisher snapshots were hash-pinned in the internal workstation evidence. The underlying source datasets and developer implementation remain outside this proprietary public-document library.

Status: source-backed evaluator baselines established; three independently qualified new sector algorithms NOT demonstrated. Physical FarmNode qualification remains zero of three and production rights have not been granted.