Translation
Translated from the Spanish source, which is the normative text. Status: draft pending human review. Differences are resolved in the Spanish text.
Research: from requirements to evidence¶
Status: research and application proposals · 17 October 2026. This study does not amend ASILO 0.1, MILPA 0.1 or SCAFFOLD 0.2, accept profiles or establish implementation conformance. The source register records URLs, editions, consulted sections and limitations. A mapping is analysis, not endorsement by a cited organization.
Question and method¶
What evidence is needed between a methodological obligation and a system action? Accepted requirements, proposed profiles and declared implementation limits were compared. External searches used generic questions and primary documentation without submitting private code, drafts, credentials or case files. Technical counterexamples and local data stay in the implementation dossier; this page provides a summary suitable for public documentation.
Distinguish four results: valid declaration, performed observation, authorized action and accepted outcome. Passing one does not establish the others. An existing file does not prove that a test ran or examined the current revision.
Mappings and remaining work¶
| Area | Existing methodological basis | Application proposal and evidence |
|---|---|---|
| Authority and provenance | MILPA-R-001/002/007/008; SCAFFOLD-R-033. | Link requirement, revision, control, negative case, result, environment and owner. W3C PROV-DM supplies entities, activities and derivations; it does not establish truth or legal authorship. |
| Identity and effects | ASILO-R-001/004/005/006; SCAFFOLD-R-037. | Obtain identity through the protocol and enforce current permissions before effects. NIST SP 800-207 rejects implicit trust based on location. As an application to development, distinguish Git checks from effective execution controls; loopback does not authenticate the process. |
| Tools and extensions | ASILO-R-001/003/004/008; SCAFFOLD-R-037. | Inventory skill, plugin and server revisions, provenance, capabilities, launch, data and egress. MCP distinguishes transport risks; Agent Skills relies on the host to enforce restrictions. No universal proxy is prescribed. |
| Delivery and recovery | SCAFFOLD-R-025 through R-029 and R-034 through R-036. | Integrate language, infrastructure and data checks, retaining revision-bound results. NIST SSDF 1.1 informs secure development; SLSA 1.2 informs artifact-provenance verification. |
| Retries and concurrency | ASILO-R-007; SCAFFOLD-R-028/035/037. | Test uncertain outcomes, concurrent updates, compatibility and recovery. RFC 9110 §9.2.2 limits non-idempotent retries; an append-only record does not implement these properties itself. |
| Measurement and evaluation | ASILO-R-008; SCAFFOLD-R-023/024/032. | Define source, window, unit, environment, baseline and limitations. NIST AI 600-1 addresses AI risks; DORA measures service delivery. Tokens or node counts do not prove value or productivity. |
The IDs refer to ASILO, MILPA, SCAFFOLD 0.1 and its 0.2 additions. Most needs concern implementation or verification profiles rather than new norms.
Practices documented by large organizations¶
Microsoft SDL documents threat reviews, supply-chain and engineering-environment security, testing, monitoring and response. It helps distribute evidence across the lifecycle; it does not establish that this ecosystem uses Microsoft's process or achieves its results.
Google SRE explains canary deployments and evaluation based on traffic, duration and service objectives. The proposal is to record project criteria and measurements without imposing 48 hours, a universal percentage or a no-outage promise.
Language, regional data and machine contracts¶
Human text is necessary; risk arises when its wording decides permissions or states. The proposal is to separate
stable codes, typed parameters and translated presentation messages. RFC 9457 §3.1.4
discourages extracting control information from HTTP's human-readable detail; extending
that principle to a CLI is a project design decision.
Record catalog versions, translation revisions and CLDR/TZDB versions for regional reproducibility. Language, data location and jurisdiction are distinct. The consulted W3C string guidance is a Working Draft, not an accepted Recommendation. OpenTelemetry GenAI v1.37.0 is also under development in that revision: pin versions and minimize telemetry rather than recording private prompts and results for convenience.
Open normative questions¶
Amendment R-017 is not accepted: a consumer does not control an external provider's declaration on its own and cannot assume it exists. The pending proposal needs a versioned public contract, consumer accountability, evidence and explicit absence of provider approval through MILPA's change process. Pretending that a bilateral declaration exists does not resolve it.
Previously recognized coverage gaps, including data/model poisoning and unexpected code execution, need bounded threats, conditions and evaluation before proposing an ASILO amendment. The engineering and architecture profiles remain proposed.
SCAFFOLD remains scoped to software projects. It does not require abandoning Scrum, Jira or meetings, or convert MILPA metaphors into agent classes or maturity scales. People, tools and responsibilities retain the scope of each accepted requirement.
Privacy, licensing and compliance¶
A project applicability record is proposed, covering activity, data, processing roles, jurisdiction, official legal source and revision, owners, controls and evidence. Technical assessment does not decide legal applicability. A standard, local execution or exclusion from training alone does not resolve retention, access, subprocessors or transfers.
Sources are linked and explained through original summaries; paywalled standards, long texts, graphics and brands are not reproduced. A signature or RFC 3161 timestamp alone does not prove originality, ownership or illicit training use. AI review helps compare evidence; it is not independent human review or certification.