Harden AI context grounding and discovery

[erinepshovel-code]

What changed

• assigns Erin Spencer the canonical JSON-LD identity https://interdependentway.org/eai/aicontext.md#erin-spencer
• inserts an early machine-readable INTERPRETATION BOUNDARY without changing the required byte-zero connection contract
• makes explicit that identity is not authentication, authorization, runtime connection, synchronization, tool access, or log access
• requires supplied cited evidence before claims about live current, drift, anomalous flux, coherence, phase-lock, behavioral bleed-through, consciousness, or instantiation state
• requires an actual cited EDCM result record before describing an EDCM measurement
• prohibits inventing or expanding identifiers, repositories, runtimes, relationships, or capabilities absent from supplied sources
• publishes generated root /llms.txt, links the endpoint from the homepage and HTML alternate metadata, and adds /sitemap.xml
• adds a repository-local llms-build compatible runner and makes drift checking part of npm run check
• regenerates the msdmd collection and adds source, generated-artifact, identity, grounding, discovery, and drift witnesses

Why

A live external-model trial recognized the proposed @id but then promoted source vocabulary into unsupported claims about measurable drift, energy currents, tensor compensation, runtime bleed-through, synchronization, and log access. Identity addressing worked; epistemic grounding did not. This patch encodes the missing non-transfer and evidence gates before the copied source material.

The boundary reduces an observed failure class. It does not guarantee independent models will obey it; that remains empirical and is carried as hmmm.

Work graph and authority

• publication consumer: The-Interdependency/The-Interdependency.github.io at parent 469951bdebd2f926dd02c0389920eb0e41bc5411
• build doctrine: The-Interdependency/skill-lib at 2b24be24947223b86440f59f1bd9766130f9cc11
• canon and all eight textbook source owners remain unchanged
• no source-owned canon, theorem, proof, measurement, empirical, license, runtime, authentication, or capability status transfers into this website patch

File plan

Path	Change	Purpose	Risk	Required evidence
.eleventy.js	modified	publish root llms.txt	low static-route	generated discovery test
The-Interdependency.github.io_msdmd.ts	modified/generated	collect new declarations, contracts, checks, and edges	metadata drift	skill-lib collector
docs/aicontext.md	modified	document identity, discovery, and inference boundaries	doctrine wording	review + checks
llms.txt	created/generated	concise machine entry instructions	stale doctrine	local and skill-lib drift gates
package.json	modified	add llms build/check commands and gate	CI	complete npm gate
scripts/build-llms.mjs	created	self-contained declaration generator and drift checker	source-generation	dedicated tests
src/_data/erin.public-biography.json	modified	add stable JSON-LD @id	public identity	identity checks
src/_includes/layouts/base.njk	modified	advertise alternate machine context	navigation metadata	generated test
src/_includes/layouts/splash.njk	modified	advertise context from root threshold	navigation metadata	generated test
src/eai/aicontext.11ty.js	modified	emit identity and grounded-inference contract	public machine doctrine	renderer tests
src/home/index.njk	modified	ordinary navigation to context and llms instructions	public UI	browser/accessibility
src/sitemap.11ty.js	created	list endpoint and llms route despite collection exclusion	route generation	source/generated tests
tests/aicontext.test.mjs	modified	witness identity and inference contracts	none	Node test
tests/generated-site.test.mjs	modified	witness built discovery and boundary content	none	generated-site test
tests/llms-build.test.mjs	created	witness declaration generation, fenced exclusion, hmmm, drift	none	Node test
tests/site-contract.test.mjs	modified	witness source discovery wiring	none	Node test

Validation

• npm run check — passed
  • workflow action audit
  • llms.txt drift gate
  • current online canon/repository/textbook/works refresh
  • 41 unit and integrity tests
  • 11 generated-artifact tests
• PLAYWRIGHT_BROWSERS_PATH=/tmp/aicontext-playwright npm run test:browser — 25 passed
• repository-local npm run check:llms and skill-lib reference python -m llms.build --check — both passed
• msdmd collection regenerated from a clean detached source tree; generated _site declarations excluded
• git diff --cached --check — passed
• final remote commit tree matches locally validated source-only tree ba14d4d91ee98aa26279dbcdbe7c7fc25b395c70

hmmm

• Compliance by independent model providers is not enforceable from a public context file; repeated external trials must measure its practical effect.
• Content digests still establish byte identity, not producer authentication.
• The existing link command remains a basic existence check; the full link checker is still pending.
• Erin's public biography field selection remains subject to line-by-line review.