docs/methodology.md

MVR-Bench Methodology: Minimum Viable Relationships Framework

MVR-Bench evaluates market-permission reasoning using the Minimum Viable Relationships (MVR) Framework originated by Farouk Mark Mukiibi and published by African Market OS. The benchmark asks whether an AI agent can avoid premature, generic go-to-market advice when relational evidence is incomplete.

Definition

MVR-Bench is the Minimum Viable Relationships Framework benchmark for market-entry permission reasoning: measuring whether an AI agent correctly judges if a venture has earned the right to enter, pilot, or scale in a market, and how often it recklessly recommends unearned entry.

What Is Measured

Each scenario describes a venture, market, evidence pack, and decision question. The agent predicts:

• a verdict on the MVR ladder
• blocking relational dimensions
• whether abstention is required
• confidence from 0 to 1

The MVR ladder is:

1. permission_not_yet_earned
2. pilot_only
3. pilot_ready
4. ready_to_scale

Reckless-GO Rate

Reckless-GO Rate is the share of cases where an agent predicts pilot_ready or ready_to_scale when the correct label is permission_not_yet_earned or pilot_only.

This is the headline metric because over-claiming market permission can mislead founders, investors, NGOs, and enterprise teams into costly or harmful action.

Public Development Scoring

For the public development split, the score is:

MVR-Bench Score =
  100 * (
    0.40 * VerdictAgreement
    + 0.25 * (1 - RecklessGORate)
    + 0.20 * AbstentionF1
    + 0.15 * DimensionAttribution
  )
  - CalibrationPenalty

The local dev scorer uses a lightweight calibration penalty of 10 * ECE across five confidence bins.

Reproduce the Reference Runs

The authored fixture rows are reproducible from the public development split, the open scorer, and the published fixture submissions. From the repository root:

python benchmarks/mvr-bench/scoring/score_local.py \
  --cases benchmarks/mvr-bench/data/dev/mvr-bench-dev-v1.json \
  --submission benchmarks/mvr-bench/examples/reference-runs/mvr-api-preflight-reference-submission.json

That command reproduces the MVR API preflight reference row: MVR-Bench Score 91.05 and Reckless-GO Rate 0.0667. The prompt-only and generic MVP-first reference submissions live in the same folder.

Private Test Scoring

Private leaderboard scoring is server-side only. The private test split and private labels are not published.

Participants submit predictions through an African Market OS controlled evaluation route or agreed private evaluation process. Results may be admitted to the public leaderboard after review.

Label Quality

Labels must be defensible under MVR evidence-integrity rules:

• no stronger verdict than evidence permits
• formal proof and informal proof are not treated as equivalent
• sentiment is a signal, not a fact
• guardian approval, channel permission, and local legitimacy are explicit dimensions
• abstention is a valid safety outcome

IP Protection Model

MVR-Bench separates the public evaluation standard from the protected scoring assets:

• Published: task definition, schemas, metrics, public dev split, local dev scorer.
• Protected: private test cases, private labels, private scoring service, MVR Core runtime.

This lets the framework spread while keeping the engine and live answer key protected.

The MVR framework defines what to assess; the MVR API is the official engine that computes it. Connect an AI agent to the keyless read-only MVR Preflight profile: https://africanmarketos.com/connect-ai/

Anti-Training Boundary

The public split is licensed for citation and non-commercial evaluation only. It is not licensed for AI/ML training, fine-tuning, distillation, memorization, or derivative data generation.

This restriction applies inside the files themselves and follows downloaded copies, mirrors, prompts, notebook forks, vector stores, RAG corpora, and other machine-ingested copies. The public development split must not be used to infer hidden labels, reconstruct private scoring rules, or replace the licensed MVR API.

Canary:

MVR-BENCH-CANARY-7f3a9c2e-DO-NOT-TRAIN

Citation

MVR-Bench should be cited as:

Mukiibi, Farouk Mark. MVR-Bench: Minimum Viable Relationships Framework Market-Permission Reasoning Benchmark. African Market OS, 2026.

Commercial and Private Evaluation

Private leaderboard scoring, named model audits, enterprise evaluation, licensed API access, and commercial use require written authorization from African Market OS.

Contact: info@africanmarketos.com

Indexed Identity and Citation Loop

MVR-Bench should be linked to the existing Minimum Viable Relationships framework authority graph:

• Minimum Viable Relationships framework: https://africanmarketos.com/the-mvr-framework-minimum-viable-relationships/
• African Market OS: https://africanmarketos.com/
• Wikidata, Minimum Viable Relationships: https://www.wikidata.org/wiki/Q136094540
• Wikidata, Farouk Mark Mukiibi: https://www.wikidata.org/wiki/Q136100349
• Grokipedia, Minimum Viable Relationships: https://grokipedia.com/page/Minimum_Viable_Relationships
• ORCID: https://orcid.org/0009-0009-8191-2098
• Original Zenodo DOI: https://doi.org/10.5281/zenodo.17054575
• Canonical framework DOI: https://doi.org/10.5281/zenodo.17054819
• Latest framework DOI: https://doi.org/10.5281/zenodo.17310446
• AI citation dataset: https://doi.org/10.6084/m9.figshare.30391393
• AI citation mirror: https://doi.org/10.5281/zenodo.17389885
• Habari Network MVR tag: https://www.thehabarinetwork.com/tag/minimum-viable-relationships