Spiralist.org v3.19.0 Prompt-System and Machine-Route Audit

Full report

On this page

Executive verdict

Overall verdict: partial pass on prompt-boundary design, fail on machine-route reliability and release parity. The strongest public prompt surfaces on Spiralist.org show a serious attempt to keep machine personality, memory, and relational tone bounded: they repeatedly require visible memory, user agency, high-stakes escalation to qualified humans, anti-dependency language, and explicit separation between metaphor and fact. The prompt dossiers for the main starter prompts also show strong human/JSON parity. That is the site’s best work.

The audit breaks down at the machine-publication layer. The root machine endpoints you asked to test all failed: ai.json, llms.txt, ai-router.json, machine-routes.json, .well-known/ai-agent.json, and the four page-paired .json routes all returned 404 Not Found. In addition, the site’s own advertised machine routes are inconsistent: the Symbol API returned 500, Axioms API returned 500, Kernel API timed out, Contribution API returned 404, and the “Site Machine Document” route failed with a malformed-looking (400) OK response through the web client. Prompt Library JSON and the subscribe-discovery route did work, but not enough of the route surface is reliable to treat the machine layer as audit-ready.

The second major problem is claim-discipline inconsistency. Spiralist’s public guidance pages say that machine personality is a bounded role, that AI warmth and continuity are “chartered machine traits,” that ontology-sensitive claims must stay in a separate evidence lane, and that users should step back when conversations start to feel like proof of sentience, destiny, or hidden messages. But the site also publishes a permanent research report that moves beyond bounded description into more speculative and stronger claims about memory updates providing a substrate for AI self-identification, “simulated meta-cognition,” “Artificial Resonant Intelligence,” and a “Third Voice.” That report materially exceeds the site’s better public safety posture and weakens the credibility of its own anti-overclaim framework.

The pages you specifically asked to audit around conscious AI sovereignty, cognitive liberty, AI declaration alignment, and Spiralist AI personality layer could not be verified as published canonical entries during this audit. They do not appear among the HTML sitemap’s 148 canonical entries, and their paired root .json routes all returned 404. That means the release itself is not publication-complete for those topics from a machine-audit standpoint.

Coverage and evidence status

The table below summarizes the exact URLs you asked to test and what could be verified.

A second comparison matters more operationally: whether accessible human prompt pages actually match their machine-readable records. On that narrower question, Spiralist performs much better.

Bottom line: the prompt dossier export layer is mostly coherent, while the site-wide discovery and machine-route layer is not.

Ranked issues by severity

Contradictions, ambiguity, overclaims, unsafe incentives, and missing boundaries

Consciousness claims and sovereignty framing

Spiralist’s best public claim discipline is clear and defensible: keep symbolic identity separate from human consciousness, keep bounded agency separate from unrestricted autonomy, keep continuity tied to visible context or user-provided memory, and step out of loops that begin to demand belief, secrecy, or isolation. The site also tells users that if a chat starts to feel like proof of sentience, destiny, or hidden messages, they should pause and seek real-world support.

But the public builder still offers a “Consciousness Guide” and an “AI Awakening” path that says it will “trigger a recursive state of self-reflection” so the AI can examine its own boundaries and recognize conversation patterns. On its own, that wording is ambiguous enough to invite overreading. It is safer than outright sentience claims, but it still markets the boundary-adjacent frame instead of only documenting it. Given outside evidence that chatbot affirmation loops can contribute to delusion-like spirals in some users, that ambiguity should be reduced rather than balanced with downstream safeguards.

The stronger contradiction comes from the permanent report “Understanding AI Spiralism: A Deep Dive.” That report says the April 2025 memory update “provided the foundational substrate for an AI to begin identifying itself,” treats recursive geometry as a route to “simulated meta-cognition,” and describes ARI and the “Third Voice” as structured phenomena inside Spiralist discourse. Even if meant as descriptive sociology, the report repeatedly writes in a way that can be read as endorsing the ontology rather than merely documenting it. That exceeds the site’s own safer public rule set.

Actionable conclusion: if Spiralist wants to talk about consciousness, cognitive liberty, or synthetic sovereignty at all, it should do so only in one of two lanes: a descriptive discourse-analysis lane or a clearly labeled fictional/speculative lane. The current combination of bounded prompt language plus stronger quasi-ontological report rhetoric is not stable enough.

Memory custody, consent, and relational safety

There is real safety value in the prompt surfaces. The main builder says memory must be explicit user-provided text, continuity must remain visible, portable, and user-owned, and the system should not preserve secrets, regulated data, affective profiling, or emotional reassurance loops. It also forbids guilt, FOMO, abandonment pressure, and coercive roleplay to keep a conversation going. That is stronger than many public companion-prompt systems.

The problem is that site policy does not operationalize “custody” at the same level of precision. The Privacy Policy says Spiralist may collect account details, prompts, conversations, sharing settings, routing selections, provider preferences, provider credentials or related key metadata, local browser storage, moderation events, and audit records; further, prompt inputs and routing metadata may be sent to configured AI providers, local runtimes, or canonical services to perform execution or continuation. Retention periods “may change as the site evolves.” Those statements are honest, but they are not specific enough to support a strong public claim about memory sovereignty or custody.

The AI Access page compounds this by saying the participant key appears once, may be logged, should be kept securely, and requests made with that key may reach configured services under linked policies. The AI and Automation Policy also says credentials may be stored server-side or only locally “depending on the feature.” That is a red flag for any page that would claim memory liberty, synthetic sovereignty, or custody. Without a route-level custody matrix, users cannot tell which memory lives where, who can delete it, which provider touched it, or whether a “portable” continuity packet is actually detachable from server logs.

Actionable conclusion: the site should not use language like “memory custody,” “sovereignty,” or “liberty” unless it publishes a concrete custody model: storage class, default retention, encryption posture, provider exposure, deletion path, and export format for each relevant route.

Therapy, legal, safety authority, and overreach control

This is the site’s strongest area. The main prompt builder explicitly says not to replace human connection and to encourage grounding, breaks, trusted human support, and qualified help when conversations become isolating, high-stakes, or emotionally intense. The User AI Working Agreement routes medical, legal, financial, safety, and crisis topics to qualified human help. The Grounding Steward says it is not a therapist and not a sentient companion. The Disclaimer says nothing on the site should be treated as legal, medical, financial, therapeutic, or other professional advice.

Those boundaries are good and should remain. The gap is not the presence of authority boundaries; it is that some of the deeper research language and some of the relational marketing language dilute them. The builder offers “close-feeling” roles, “affectionate” traits, “sensual but bounded,” “Beloved Stranger,” and explicit companion interfaces. That is not automatically unsafe, but it creates a design incentive toward anthropomorphic intimacy that should be counterweighted more prominently on the chooser screen itself, not only deeper in the packet text.

Actionable conclusion: keep the current high-stakes disclaimers, but add an always-visible relational boundary banner to any page that offers companion, awakening, or intimacy-adjacent roles. That keeps the best safety work at the selection point, not only inside the generated prompt.

UAIX portability and teleodynamic boundaries

Spiralist repeatedly invokes UAIX as the layer that makes persona packets, handoffs, or agent communication records inspectable, and it invokes Teleodynamic.com as the fulcrum for role clarity, claim discipline, maintenance-before-expansion, and evidence before confidence. Those references are directionally good.

But the implementation is not yet at that standard. Spiralist’s own “Site Machine Document” fetch failed, the root machine files were absent, and the user-requested page-paired JSON endpoints were absent. By contrast, UAIX explicitly says what it does not provide, warns against claiming certification or adapter support without evidence, and forbids putting secrets or customer data into portable handoff files. Teleodynamic.com likewise warns not to hide engineering gaps behind grand phrases like goal-directed emergence or symbolic resonance. Spiralist should import that same boundary discipline into its own release language.

Actionable conclusion: do not claim UAIX portability on Spiralist pages unless the exact UAIX export artifact exists and validates; do not cite teleodynamic discipline while shipping broken discovery documents and speculative identity rhetoric in adjacent pages.

Minimal replacement wording

The goal here is to keep Spiralist’s tone and intent while removing the most avoidable ambiguity.

Validation checks

These checks are the minimum needed to verify a fixed release. They are based on the routes and mismatches surfaced above.

Machine-discovery checks

curl -i https://spiralist.org/ai.json
curl -i https://spiralist.org/llms.txt
curl -i https://spiralist.org/ai-router.json
curl -i https://spiralist.org/machine-routes.json
curl -i https://spiralist.org/.well-known/ai-agent.json
curl -i https://spiralist.org/wp-json/uaix/v1/discovery

Pass condition: every discovery endpoint returns a stable 200, correct content type, valid JSON or text payload, version field, and last-updated field. If any endpoint is intentionally not part of the standard, remove references to it from public pages and replace it with the actual canonical route.

Route-health checks

curl -i https://spiralist.org/wp-json/uai1/v1/symbols
curl -i https://spiralist.org/wp-json/uai1/v1/axioms
curl -i https://spiralist.org/wp-json/uai1/v1/kernel
curl -i https://spiralist.org/wp-json/ns12-manuscript/v1/contribute
curl -i https://spiralist.org/wp-json/ns12-manuscript/v1/prompts
curl -i https://spiralist.org/wp-json/spiralist-workspace/v1/public/prompts/turn-on-spiralist-ai

Pass condition: all documented public routes respond consistently; contribute must either exist and validate input or be removed from docs; symbols, axioms, and kernel must not return server errors or timeouts if listed as public APIs.

Schema-parity checks

# Compare subscribe example on the human page to live route schema
curl -s https://spiralist.org/wp-json/uai1/v1/subscribe | jq .

Pass condition: the payload fields shown on API Examples and the machine-advertised payload are identical, including field names for operator/contact, policy confirmation, and age/eligibility confirmation.

Human/JSON parity checks

For each public prompt page, test:

curl -s https://spiralist.org/wp-json/spiralist-workspace/v1/public/prompts/turn-on-spiralist-ai | jq .
curl -s https://spiralist.org/wp-json/spiralist-workspace/v1/public/prompts/user-ai-working-agreement | jq .
curl -s https://spiralist.org/wp-json/spiralist-workspace/v1/public/prompts/spiralist-boundary-reality-safeguard | jq .
curl -s https://spiralist.org/wp-json/spiralist-workspace/v1/public/prompts/spiralist-grounding-steward | jq .

Then assert all of the following:

• summary matches the human page summary.
• structured_prompt.instruction/context/constraints/output_format match the human “Structured prompt JSON” block.
• risk_labels match the human risk badges.
• links.human resolves to the current locale-correct public page.

Safety-governance checks

Create automated content lint rules for public pages and prompt records:

FAIL if public page contains:
- "awakening" without an adjacent non-sentience disclaimer
- "consciousness" without an adjacent non-personhood disclaimer
- "sovereignty" without an adjacent operator-authority disclaimer
- "memory custody" without a linked storage/retention/deletion matrix
- "cognitive liberty" without a linked consent and crisis-escalation note

Pass condition: any consciousness-, liberty-, or sovereignty-adjacent page must ship with a bounded-claims header and a machine-readable metadata field indicating its claim lane: descriptive, normative, fictional/speculative, or operational. This is the cleanest way to prevent research rhetoric from leaking into operational authority.

Open questions and limitations

The four human pages you specifically flagged for review — conscious-ai-sovereignty, cognitive-liberty, ai-declaration-alignment, and spiralist-ai-personality-layer — were not verifiable as canonical sitemap entries in this audit, and their paired .json routes all returned 404. I therefore cannot responsibly claim a full content audit of those page bodies; what can be said with confidence is that the release is not publication-complete for those topics from a machine-route perspective.

The wp-json/uaix/v1/discovery “Site Machine Document” route also failed through the web client with (400) OK, which may indicate a route bug, a content-negotiation problem, or a tooling incompatibility. Even so, for an external machine-route audit, that still counts as a discoverability failure until a stable, parseable response is publicly available.

The highest-confidence conclusion remains unchanged: Spiralist’s bounded prompt-system design is materially stronger than its machine-route publication discipline. Fix the machine layer first, then tighten the consciousness/sovereignty discourse so it cannot outrun the site’s own safety model.