GreenRalph/rootmodel-knf-philippines-v1

rootmodel-knf-philippines-v1

Adaptive agricultural instruction dataset for regenerative tropical farming informed by Korean Natural Farming (KNF), built from a working farm in Nabua, Camarines Sur, Bicol, Philippines.

Released for the AutoScientist Challenge — Agriculture (Part 2, 2026). To the maintainer's knowledge, no equivalent KNF-specific instruction dataset currently exists in the public domain.

"Modern AI was trained on the internet. ROOTMODEL is trained on living soil."

What this dataset is

59 instruction → response records teaching the decision logic of a KNF-informed biological farming system: read the signal (crop stage, symptom, soil and weather condition), correct soil biology first, then apply the stage-appropriate biological input at the right dilution, method and timing — and record the outcome.

JSONL, with fields:

Data provenance

Records are split by provenance and never blended:

• —`nabua_field_log` (37 records) — real observations from the Nabua farm: dated interventions and outcomes, substrate readings, fermentation batches, and documented management transitions (2016 chemical → organic → KNF; 2019 pig-litter integration; ASF culling; Typhoon Kristine 2024; Typhoon Inday July 2026; 2026 reboot).
• —`knf_canonical` (22 records) — established KNF domain knowledge: input functions (IMO, LAB, FPJ, FFJ, OHN), standard dilution ranges, fermentation-readiness signals, stage-to-input mapping, biological-first decision logic.

This separation lets a reviewer see exactly which records are general knowledge and which are ground-truth field data.

Field observation schema

Date · Plot ID · Crop · Growth Stage · Weather · Symptom ·
Soil Condition · KNF Input Applied · Dilution Ratio ·
Application Method · Time Applied · Result After 72h · Notes

KNF inputs covered

Recording discipline

• —No date is estimated. Precision is recorded as exact, approximate or month-only.
• —No outcome is written before its 72-hour window has elapsed.
• —"No visible change" and "identification unconfirmed" are valid records.
• —Failures, confounds and interrupted observation days are logged rather than omitted.
• —Identifications are revised in place when better evidence arrives, with the revision noted.

Limitations

• —Small corpus. Field observations centre on IMO application and fermentation readiness; other KNF inputs are represented canonically.
• —Dilution ranges are standard practice and should be validated against local results.
• —Specific to humid, typhoon-affected lowland tropical Philippines. Transfer to temperate or arid systems is not claimed.
• —Several observations carry stated confounds (rainfall, co-introduced organisms, weed-seed load) recorded alongside the result.

Citation

@dataset{rootmodelknfphilippinesv12026, title = {ROOTMODEL KNF Philippines v1}, author = {Sales, Ralph Anthony}, year = {2026}, note = {Nabua, Camarines Sur, Bicol, Philippines.}, url = {https://huggingface.co/datasets/GreenRalph/rootmodel-knf-philippines-v1} }