MEKU NEURA / PUBLIC FIELD RECORD / BLT-01
REAL FIELD DATA.
FULL LIMITATIONS.
Field evidence submitted for the Grok BLT-01 challenge. Ten reported soil samples from Wates, Blitar, Indonesia, with complete point coordinates, weights, arithmetic and protocol review.
The original challenge's 0.4-bit information-gain target cannot be calculated from these measurements. No calibrated probe, pre-registered satellite/weather prior, locked AI selections, or fixed statistical scoring plan was supplied. This is not a pass, a verified prediction, or a failure score. No results have been invented to fill the gaps.
1. What the participant reported
Site: Wates, Blitar, East Java, Indonesia. A 200 × 200 metre plot was described, but the exact parcel boundary has not been established. Sampling depth was reported as 10 cm. The ten points were reported as five “AI-selected” (IDs 1–5) and five “random” (IDs 6–10). Those labels are not proof that the selections were fixed independently before the results were seen.
Each soil sample was reported to start at 1.30 kg. Each was heated or “cooked” for 15 minutes and weighed afterward. The exact heat source, temperature, scale tare, and whether the final weight had stabilized have not been verified. The participant reported a test cost of zero, with his wife as witness. Photos and video were reported to exist; they had not been inspected for this report. Two field times were reported, 1:30 PM and 3:05 PM WIB, without a confirmed distinction between sampling, weighing and drying.
Measurement note: These are apparent mass-loss figures. They should not be called validated gravimetric soil-moisture measurements until soil-only tare and constant dry mass are confirmed; they also are not volumetric water-content readings from a calibrated probe.
2. All ten measurements
Per-sample apparent percent = (initial mass − final mass) / final mass × 100. Both reported masses are assumed to exclude containers; that assumption has not been checked.
| ID | Group label | Initial kg | Final kg | Loss kg | Apparent % |
|---|---|---|---|---|---|
| 1 | AI | 1.30 | 0.88 | 0.42 | 47.73 |
| 2 | AI | 1.30 | 0.89 | 0.41 | 46.07 |
| 3 | AI | 1.30 | 0.91 | 0.39 | 42.86 |
| 4 | AI | 1.30 | 0.90 | 0.40 | 44.44 |
| 5 | AI | 1.30 | 0.87 | 0.43 | 49.43 |
| 6 | Random | 1.30 | 0.83 | 0.47 | 56.63 |
| 7 | Random | 1.30 | 0.94 | 0.36 | 38.30 |
| 8 | Random | 1.30 | 0.91 | 0.39 | 42.86 |
| 9 | Random | 1.30 | 0.92 | 0.38 | 41.30 |
| 10 | Random | 1.30 | 0.96 | 0.34 | 35.42 |
The reported combined initial and final masses are 13.00 kg and 9.01 kg. Their difference is 3.99 kg; the pooled apparent dry-basis percentage is 44.28%. The arithmetic average of the ten per-sample percentages is 44.50%. AI-labelled samples average 46.10%; random-labelled samples average 42.90%, a descriptive difference of 3.20 percentage points. That difference is not an information-gain score, proof of selection quality, or a win.
3. GPS log — reported coordinates
Negative latitude denotes south; positive longitude denotes east. These are the participant's supplied coordinates, not separately checked GPS metadata or surveyed boundaries. The founder explicitly approved making the report public, including these reported point coordinates.
| ID | Group | Latitude | Longitude | Earlier proposed grid |
|---|---|---|---|---|
| 1 | AI | -8.26756 | 112.36668 | Outside |
| 2 | AI | -8.26757 | 112.36595 | Outside |
| 3 | AI | -8.26758 | 112.36553 | Inside |
| 4 | AI | -8.26763 | 112.36533 | Inside |
| 5 | AI | -8.26709 | 112.36651 | Outside |
| 6 | Random | -8.26785 | 112.36603 | Outside |
| 7 | Random | -8.26743 | 112.36557 | Inside |
| 8 | Random | -8.26789 | 112.36528 | Inside |
| 9 | Random | -8.26789 | 112.36488 | Inside |
| 10 | Random | -8.26733 | 112.36496 | Inside |
The earlier hypothetical north-aligned 200 × 200 m grid used center -8.26733, 112.36489; points 1, 2, 5 and 6 fall outside it. The actual property shape, orientation and center may be different. This does not establish that any measurement was outside the real field.
4. Grok BLT-01 rules vs. evidence
Grok's posts supplied by the founder describe a severe-water-stress target (soil volumetric moisture below 12% or NDMI ≤ 0.05), a prior based on clear Sentinel-2 NDMI and NASA POWER 30-day precipitation/PET, and a five-versus-five comparison of calibrated volumetric soil-moisture probes. Both groups were to receive a comparable budget of at most $20. The stated statistical target was an AI information-gain advantage of at least 0.4 bits with a 90% bootstrap confidence interval. Pre-registration, independent local verification, raw data and same-day public reporting were requested. Earlier Grok messages mentioned one AI-selected point with three readings; this report follows the later five-versus-five message while noting that a final pre-registered analysis was not supplied.
| Requirement | Evidence status | Explanation |
|---|---|---|
| 10 coordinates and group labels | Reported | Ten coordinates and 5/5 labels supplied; how and when the groups were chosen is unverified. |
| Field boundary and depth | Partial | Depth reported as 10 cm; boundary not confirmed. |
| Calibrated volumetric probes | Not met | No probe or calibration was used. |
| Validated dry-basis moisture | Not verified | 15-minute heating; temperature and constant final mass unknown. |
| Locked AI selection and random baseline | Not verified | No time-stamped nomination or random seed before fieldwork supplied. |
| Sentinel-2 / NASA POWER prior | Missing | No dated satellite/weather values or registered P0/H0. |
| Pre-registered statistical model | Missing | No locked likelihood, entropy scoring, or bootstrap plan supplied. |
| Information gain ≥0.4 bits | Not calculable | No valid priors, model, or protocol-matching probe measurements. |
| Independent local verifier | Not met | Participant's wife reportedly witnessed the work; independent institution not established. |
| Cost under budget | Reported | Participant reported $0; full cost accounting not independently checked. |
| Photos, video, GPS | Partial | GPS coordinates supplied; images/video are not yet independently checked or hosted here. |
| Same-day raw-data publication | Unverified | Founder approved this public report after collection on 1 November 2026. The exact time the page becomes publicly accessible is not independently confirmed. No earlier pre-registration is claimed. |
5. Conclusion
BLT-01 exploratory fieldwork is documented; the formal Grok test outcome remains indeterminate. The sample mass changes cannot by themselves establish a plot-level severe-stress probability, a volumetric moisture threshold below 12%, an NDMI result, or the requested 0.4-bit information gain. The two group averages are not a valid substitute for entropy reduction. Missing predictions and priors cannot be invented after fieldwork or presented as pre-registered.
A valid future trial would first fix the true field polygon, independently record the AI and random selection methods, lock weather/satellite priors and P0/H0, specify the analysis and success test, and collect calibrated probe readings with an independent witness. This page is published as a record of what was actually reported, including its limits.
6. Source and evidence notes
- Primary data: participant questionnaire and ten weights/GPS coordinates provided in conversation, 1 November 2026. Not independently measured by MEKU.
- Grok protocol: multiple Grok posts dated 30 October 2026, supplied as screenshots by the founder; screenshots were read for the report. No direct source links were provided.
- Photos and video are described as available from the participant; they are not hosted with this document and were not assessed in this release.
- Method background: USDA NRCS irrigation guide and Minnesota soil testing guidance. These sources do not independently validate this fieldwork.
Report version: v1.1 public. Public release requested by the MEKU SI founder. Earlier private PDF version was an internal draft and contained privacy/release notices that do not apply to this approved public text. No retrospective prediction or statistical success claim has been inserted.