docs/SERIES_A_MARKET_EVIDENCE.md

Series-A market evidence — governed AI diligence map

This file turns public standards and market expectations into an evidence map for the A11oy investor demo. It is not a certification claim. It shows where the repo already has reviewable artifacts and where production work remains.

External signals used

Source	Public signal	A11oy response	Current evidence	Gap / caveat
NIST AI Risk Management Framework	AI risk programs need govern, map, measure, and manage functions for trustworthy AI.	A11oy frames governed execution as policy gates, measurement, provenance, and managed payload release.	docs/PROVENANCE.md, docs/ecosystem-readiness-report.json, npm run test:runtime	Not a NIST certification; map exact customer controls during deployment.
EU AI Act, Regulation (EU) 2024/1689	High-risk AI emphasizes risk-based obligations, technical documentation, transparency, record keeping, and human oversight.	A11oy shows receipt chains, claim-status language, and human-review/operator proof points.	packages/receipt-substrate, huggingface/DEMO_RECEIPT_SAMPLE.jsonl, docs/INVESTOR_DEMO.md	Customer-specific EU AI Act classification and legal review remain deployment work.
CISA SBOM guidance	SBOMs support software supply-chain transparency and machine-processable risk analysis.	A11oy keeps SBOM workflows, payload manifests, and deterministic bundles in the diligence path.	.github/workflows/sbom.yml, deploy/MANIFEST.json, pnpm payload:bundle:verify	SBOM assets do not equal signed deployment payloads; v0.3.x signed assets need owner-side release evidence.
SLSA framework	Provenance levels are progressive; higher claims require signed provenance and hardened build evidence.	A11oy uses honest SLSA language and avoids L3 claims unless the workflow evidence exists.	README SLSA badge, .github/workflows/slsa.yml, .github/workflows/doctrine.yml	SLSA L2/L3 should remain PR/roadmap until merged workflow evidence is green.
Model Cards for Model Reporting	Public AI artifacts should disclose intended use, performance context, limitations, and responsible-use boundaries.	The Hugging Face mirror acts as a model-card-like diligence packet while explicitly saying A11oy is not a model checkpoint.	huggingface/README.md, huggingface/VERIFICATION.md, huggingface/INVESTOR_BRIEF.md	HF is a mirror, not canonical release truth; stale remote files must be pruned on publish.

The Series-A wedge

A11oy is strongest when presented as a control fabric for regulated AI work:

1. Govern: policy gates, claim-status language, and excluded-scope tables.
2. Map: ecosystem registry and readiness report across 19 public repos.
3. Measure: doctrine tests, QEC tests, receipt tests, and payload verification.
4. Manage: deterministic operational bundle, SBOM lane, and guarded HF mirror.

This is the investor-safe story: A11oy makes agentic execution reviewable before it becomes customer impact. It does not need to pretend to be a model, a finished UDS catalog product, or a fully closed Lean proof corpus.

What must be real before production

Gate	Required evidence
Signed deployment payloads	Release assets with tarball, .sig, .sha256, public key, and verification command.
GHCR deployment path	Public or authenticated image/package pull proof tied to the same release line.
Customer controls	Deployment-specific threat model, retention policy, access control, and legal classification.
Formal proof claims	Current upstream lutar-lean proof report for exact theorem/module claims.
HF publication	pnpm payload:huggingface output published with stale remote files pruned.

Public references

• NIST AI Risk Management Framework: https://www.nist.gov/itl/ai-risk-management-framework
• EU AI Act text: https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng
• CISA SBOM minimum elements: https://www.cisa.gov/resources-tools/resources/2025-minimum-elements-software-bill-materials-sbom
• SLSA framework: https://slsa.dev/
• Model Cards for Model Reporting: https://arxiv.org/abs/1810.03993