# BTG Foundry — Dated Engineering Evidence and Academic Evaluation Protocol

**Engineering snapshot: October 26, 2026.** Historical results only; this does not override newer separately verified model/host build evidence.

## Current local Foundry test suite

On October 26, the consolidated E-drive workspace ran the current `python -m unittest discover -s tests -p 'test_*.py' -q` suite: **121 tests in 16.084 seconds, `OK`**. This is a local multi-module engineering test pass. It does not establish full real native repair, independent physical FarmNode federation, Council model provenance, or production financial transactions.

## Measured work to date

| Test or observation | What passed/was inspected | What is not established |
|---|---|---|
| 18/18 bounded integrated tests | Original Studio static importer and Graph Doctor linked to original ADAM/BTDU logical FarmNode with negative-case checks | Arbitrary autonomous software repair |
| 2/2 bounded numeric candidates | DataChip Compute IR and independent Decimal reference interpretation agree on scoped examples | Universal invention or compiler correctness |
| 9,047/9,047 SHA-256 checks | Source snapshot copy matches the consolidation manifests; 27 component folders exist under a unified E root | 9,047 functional components or migrated production data |
| 157 HuntAR source files inventoried | Static readable source index with hashes, not executed | Functional, physically tested rebuilt HuntAR |
| 14 accepted Council responses in last observed desktop check | 10 creator plus 4 reviewer acceptances, 22 named seats and three empty | 25 independently verified operational members |
| Physical network | Original logical FarmNode Host runs bound and verifies data results | Three separate attested machines, enforceable resource metering, actual settlement |

## Distinguish three categories in publications

- **Completed system architecture:** What interfaces and workflows are specified to do upon full implementation. This can be presented as a definitive design without asserting deployment.
- **Demonstrated engineering:** Exact actual tests, source hashes, version, environment, successful/failed cases and any limitations. Never treat a descriptive model diagram as evidence.
- **Production qualification:** Original ENTITY authority, host identity, real native device behavior, independent evaluators, security and deployment approvals. Do not label a release production-ready solely from unit tests.

## Independent acceptance gates

### Gate H — Native HuntAR repair
Freeze source branch and defects first. Import the actual iOS project and safely map dependencies. Require Council-authored improvement to correct startup time, UI responsiveness, camera, AR overlays/orientation/scale, species selection, GPS, 2D maps, BSIE terrain, legal mask and offline packs. Compare baseline and repair on real iPhone and iPad with signed reproducible build logs. Record intervention time and failed experiments.

### Gate G — Generalization
Choose an unrelated application in advance. Apply the identical importer → repair → DataChip → independent tests → approved output workflow. Report precision/recall of defect identification, end-to-end success rate, manual work, introduced regressions and repeat reproducibility.

### Gate C — 25-model Council
Count only separate responding AI models after attribution checks. Keep creator/reviewer assignments, weights or provider IDs, sealed task prompts, original model outputs, independence constraints, failures, timeouts and disagreement resolution. Nonresponding aliases are not members.

### Gate F — Physical FarmNode network
Use at least three distinct owner-consenting machines. Demonstrate secure host enrollment, workload dispatch, host revocation, physical resource ceilings, job retries, real GPU profiling where applicable, wall-clock comparison, network transfer costs and signed meter receipts. Do not claim off-device memory becomes physically attached GPU VRAM.

### Gate E — ENTITY and economic use
The same authenticated user must carry valid granted scopes across Studio, Bridge, FarmNode Host and Exchange. Validate grant expiry and real revocation. Record at least one genuine completed authorized technology execution linked to its rights instrument and actual economic event. A computing contribution, a technology licence and a security are separate instruments.

### Gate P — Performance and scientific comparison
Use held-out test cases, pre-registered acceptance criteria, fixed baselines, tool versions and repeated trials. Compare against conventional IDE/AI assistants, build systems and distributed job orchestrators on quality, manual time, throughput, latency, cost and recoverability. Report negative and null results.

## Minimum reproducibility package

- Original source manifest + version hashes and owner authorization
- Frozen task description and independent oracle
- Exact code/model/driver/OS versions, target hardware and execution budget
- Full input/output logs, successful and failed cases, timeouts and repairs
- Security, privacy, rights and provenance acceptance records
- Signatures, replay references, rollback and external evaluator identity
- Comparisons with baseline alternatives, including error bars when relevant

## Important evidence interpretation

The source inventory and local integration qualify a bounded working foundation, **not** the complete self-repairing distributed technology economy. Use separate historical dates when describing earlier Council results. Published future architecture does not override failed or incomplete tests.

Current websites: https://www.blackmoretechgroup.com/platform-architecture.html and https://www.blackmoretechgroup.com/engineering-proof.html.

Historical reports are retained in [the evidence library](../evidence/index.html).