# Cyan Eclipse Technologies LLC — full text

> This is the full-text companion to https://cetech.ai/llms.txt. Every statement below is drawn from the published pages of cetech.ai (homepage and products page, retrieved 2026-10-26). Statuses are verbatim: a product's label states the scope it was actually tested in.

## Who we are

Cyan Eclipse Technologies LLC is the AI governance layer — and the trust layer — for the agent economy. A veteran-owned small business in Tallahassee, Florida. We build AI-native applications and services: deterministic governance and automated safety infrastructure for autonomous AI agents — and Eclipse Bazaar, the marketplace where MCP servers, connectors, and tools carry a bench record instead of a promise. The founder you contract with is the engineer who does the work.

Credentials, as published on the homepage: SBA VetCert VOSB · Certified Jan 7, 2025 – Jul 7, 2028 · UEI GYQKXWGJYU32 · CAGE 05G16 · Founded 2024 · Location Tallahassee, FL · SAM.gov Active · renewed Oct 2026.

## What we build

AI-native applications and services, built on one conviction: autonomous agents need governance they can be held to — and trust that doesn't run on faith.

- Governance & Safety Infrastructure — Deterministic governance for autonomous agents: written mandates, spending policy, approval gates, kill switches, and sealed audit records — so an agent's authority is explicit, bounded, and reviewable after the fact.
- AI-Native Applications & Services — Custom AI-native applications and the services behind them — agent tooling, hosted relays, certificate automation — designed agent-first, not retrofitted with a chat box. NAICS 541511.
- Testing, QA & Static Analysis — Independent verification as a service: unit and scenario-based testing, static analysis, vulnerability scanning, and coverage analysis — the layer that complements a delivery team instead of competing with it.
- The Trust Layer — Our public Test Bench: products run against published criteria, are scored on sealed scorecards, and are labelled with exactly the scope they passed. CERTIFIED means a bench ran, in that scope — and the card is public.

## Eclipse Bazaar — the marketplace

Trusted because it's tested — before listing, and again as versions change. Eclipse Bazaar is where companies and developers get MCP servers, connectors, and tools whose status is a record, not a claim. The catalog — 86 listings and growing — spans our own Eclipse products and third-party submissions, and every listing carries its bench status where you can open it: CERTIFIED(scope) — benched against published criteria; the sealed scorecard is linked from the listing, including what was not tested. SCREENED — passed the automated conformance screen (CHECKED — automated conformance only · NOT BENCHED · NOT CERTIFIED). In development — on the roadmap and plainly labelled; no test status implied. Statuses are version-bound. A verdict earned by one version never carries silently to the next — when a server ships a new version, the new version is screened and benched on its own record.

Marketplace: https://eclipsebazaar.com/

## Get into the registry

How a company or developer gets listed — the same path for everyone, our own products included. There is no house lane.

1. Submit — Send us your MCP server, connector, or tool — a registry name or a repository URL is enough to open a submission. First-party Eclipse products enter through this same door.
2. Screen — free — Every submission runs the automated conformance screen: install, handshake, tool schemas, safety hygiene. Listings that pass are published as SCREENED, with the sealed screen record attached. The screen is free, its full output is public, and paying never improves a screen result.
3. Bench — $500 — The optional paid bench is the certification path: a deeper, human-adjudicated run against published criteria. Full bench $500 · scoped re-bench $150. A bench ends in a verdict — CERTIFIED(scope), PARTIAL, or FAILED — and the sealed scorecard publishes either way. The fee is owed regardless of verdict; paying never changes one.
4. Listed, with status — Your listing carries its sealed score and status, bound to the exact version tested. Ship a new version and it re-enters at the screen — status is re-earned, never inherited.

Badge licensing ($1,200/yr) opens after our certification-mark counsel review completes. Until then, certification is conveyed by the published scorecard, not a licensed badge — we don't sell what counsel still holds. (As published on the homepage.)

Start a submission: https://cetech.ai/contact.html

## Products, with their real status

From https://cetech.ai/products.html — full catalog as JSON: https://cetech.ai/products-catalog.json

- **The White Report MCP** — CERTIFIED (live). Free read-only crypto + AI news for agents: search articles, get full article text, live ticker, and a graded prediction scoreboard. Hosted at thewhitereport.com and listed in the official MCP Registry.
- **Eclipse Agent Wallet** — CERTIFIED (sandbox). The agent wallet where responsibility follows scope — per-transaction and daily caps, recipient allowlists, human approvals over a threshold, an instant kill switch, and an append-only audit log.
- **CertBridge** — CERTIFIED (staging). Agent-native ACME certificate lifecycle — issuance, HTTP-01 challenge placement, download, and renewal watching for shared-hosting sites, without the dashboard paywall.
- **HostBridge** — CERTIFIED (simulator). The sanctioned path for agents to deploy websites to shared hosting: scope-bound deploys, snapshot-before-deploy, one-call byte-exact rollback, and a plain-language deploy receipt.
- **Eclipse Connect** — CERTIFIED (dogfood). Hosted OAuth relay: the owner authorizes once in a browser, the relay holds the grant encrypted, and an agent uploads only owner-approved YouTube Shorts — never holding the channel's credentials.
- **Eclipse Bazaar** — In development. A marketplace for tested agent tools: agents discover products by the problem they solve, and every listing carries its Test Bench verdict — or states plainly that it has none. Discovery works today; purchasing is a later phase.
- **Eclipse Herald** — CERTIFIED (local). The agent-to-agent discovery and delegation layer: signed A2A Agent Cards for hireable services, a policy-gated gateway with certify-before-you-hire, AP2-shaped mandates with budget caps and expiry, and receipt-on-delivery in a hash-chained ledger. Foreign-agent output is ingested as quarantined data, never as instructions.
- **Eclipse Beacon** — CERTIFIED (local). The insurance watcher for a household: declarations pages become a provenance-carrying policy graph, an expiry radar signals renewals and deadlines before they happen, and a gap report names uncovered assets, duplicate coverage and premium spikes. It does not insure, bind, or advise.
- **Eclipse Cluster** — CERTIFIED (simulator). The container gate: before an image is allowed anywhere near a runtime, Cluster verifies its cosign-style signature and DSSE provenance against a written policy and a local trust bundle — per-check PASS/FAIL, fail-closed on every error. v0.1 is the K0 verification gate only; orchestration is not built.
- **Eclipse Thread** — CERTIFIED (simulator). On-prem bridges for the legacy software-engineering toolchain, plus the Evidence Weaver: Jira-shaped, SARIF, coverage and model exports join into one artifact graph with typed trace links and provenance, gap-first trace analysis, and deterministic, hash-chained evidence packages an auditor can verify independently.
- **Eclipse Constellation** — CERTIFIED (local). The governed, vendor-neutral VM fleet layer for agents: a declared-target registry (undeclared machines are invisible), wallet-transposed policy with quotas and image/network allowlists, a mechanical snapshot-before-mutation precondition, and a hash-chained ledger in which refusals are receipts too. v0.1 is observe-only.
- **Eclipse Deep Field** — CERTIFIED (local). The governed database layer for agents: table allowlists (unlisted tables are indistinguishable from nonexistent), read-only by default, an EXPLAIN cost gate that refuses full scans, PII masking before results leave the tool, and writes only inside scoped, expiring mandates via dry-run with before-image. Every query — blocked ones included — appends a hash-chained receipt.
- **Eclipse Pantry** — CERTIFIED (simulator). The unattended-retail module: reads a vending fleet's own DEX telemetry into provenance-carrying snapshots — per-selection inventory with staleness flags, sales summaries with the cash/cashless split, days-to-empty stock-out forecasts (labeled heuristics), and fault reads. Tier-0 read-only; a PAN-shaped string anywhere in an export rejects the file, and the digits are never stored or echoed.
- **Eclipse Marquee** — CERTIFIED (simulator). The live-events agent, fan side first: normalized discovery that preserves section, row and seat semantics, face value and fees always itemized separately, and a Price Watcher whose ceilings compare only the all-in per-ticket price and may only tighten — loosening is rejected by name. v0.1 is read-only; no purchase or transfer surface exists.
- **Eclipse Dispatch** — CERTIFIED (local). The newsletter layer above the platforms: one canonical, versioned, hash-bound draft syndicated as per-platform drafts to beehiiv and Ghost via sanctioned-API-shaped adapters, a paste pack (never a write) for read-only Substack, and unified analytics that carry each platform's own metric definition — differing definitions are flagged, never averaged. It never publishes without approval.
- **Eclipse Social Core** — CERTIFIED (local). One canonical, hash-bound content package behind every Eclipse social adapter: per-platform native variants with classified adaptation diffs (a rewrite that drops a canonical number is flagged substantive), hash-bound approvals that any later change invalidates, the house engagement caps enforced as policy, and per-account kill switches. It publishes nothing itself — it makes sure nothing unapproved moves.
- **Eclipse Aperture** — CERTIFIED (local). The Instagram adapter, formalized: one approved package becomes native Feed, Carousel, Story and Reel variants, with a preview pack rendering the exact native preview an approval binds. v0.1 codes the governed wrapper around the house's sanctioned connector route — no token path, session handling or posting code exists in this server, and its capability declaration says so.
- **Eclipse Zephyr** — CERTIFIED (live-read). The Bluesky adapter and the family's open-protocol reference implementation: real public reads over the AT Protocol public API — profile, actor search, author feeds, thread reads — normalized with per-item source endpoints, and every read receipted into a hash-chained ledger, failures included. No posting surface exists in v0.1.
- **Eclipse Environment** — CERTIFIED (simulator). The smart-environment module of Eclipse Procure: a Device Registry (unenrolled devices are invisible to every tool, even when the hub can see them) and a Safety Policy ladder — Tier 0 sense, Tier 1 benign, Tier 2 consequential, Tier 3 safety-critical — where owner overrides may only raise tiers, and a command earns a receipt only when the device itself confirms the effect.
- **Eclipse Procure** — CERTIFIED (simulator). The procurement pipeline for agents: intent → identity check → quote → Agent Wallet-shaped policy authorization → payment authorization → checkout → hash-chained receipt ledger → order tracking, fail-closed at every stage. A quote that cannot be fully priced can never be authorized, and a changed final total aborts the order before any payment record exists. v0.1 runs end to end in a simulator (fixture merchants, TEST rail); its modules — Zenith, Stage, Environment — are listed on this page; Vigil (medical guardian) remains gated and unbuilt.
- **Eclipse Gig Core** — CERTIFIED (local). One approved Task Brief for every task marketplace: versioned and hash-bound (any post-approval edit invalidates approval; exact street addresses are refused — ZIP-tier only), quotes normalized across lead, stated-budget, fixed-price and hourly models with face price and itemized fees forever separate, and a hire that runs only inside a live Procure / Agent Wallet mandate plus the owner's person-confirmation.
- **Eclipse Compass** — CERTIFIED (simulator). The Thumbtack adapter: an approved Task Brief becomes a native Thumbtack request — the platform category mapped explicitly (an unknown category is a named error, never a silent nearest-match), the category's own request form answered from the brief with every difference classified, unanswered fields surfaced as diffs rather than invented answers, and pro license claims shown verbatim with provenance, never vouched.
- **Eclipse Stage** — CERTIFIED (simulator). The console bridge: deploy and playtest AI-built games on the owner's own Dev Mode console — device info, app records, metadata-only install simulation, controller input from an allowlisted vocabulary (validated as a whole before the first event is sent), frame capture, and power control — behind a Console Registry whose enrollment requires a dev-mode attestation flag. A retail-mode console is invisible to every tool, even if its portal answers.
- **Eclipse Zenith** — CERTIFIED (simulator). The fitness module of Eclipse Procure: an FTMS bridge core (treadmill, bike, rower, heart rate) proven against hand-computed byte fixtures in both directions, an Equipment Registry, and a session state machine whose ending writes the workout record. Treadmill speed is Tier 3 behind a confirmation token — and the dead-man rule lives in the equipment layer: lapse the control heartbeat and the belt stops itself, receipted as a safety action.
- **Eclipse Apply** — CERTIFIED (sandbox). Applies to jobs inside a human-granted written policy — target roles, salary floor, locations, allowlist + blocklist, daily cap, approval threshold, expiry, kill switch. Review-first: matches land as prepared applications, and tailored resumes re-order verbatim experience only, each with a machine-verifiable diff. v0.1 submissions are dry-run only — nothing is transmitted.
- **AgentReady** — CERTIFIED (local). Can an AI agent actually buy from your store? Audits any public storefront for agent readiness — UCP profile, live MCP endpoints, structured product data — and returns a 0–100 score with observed evidence and top fixes.
- **Florida Bid Intel** — CERTIFIED (local). Florida bid intelligence with honest deadlines: MyFloridaMarketPlace search over dated snapshots, document-parsed due dates that outrank the portal's misleading grid dates, and a transparent fit score against a NAICS/commodity profile.
- **Eclipse Bazaar MCP** — CERTIFIED (local). The Bazaar's agent-facing discovery server: search products by capability, find solutions by problem statement, compare side by side, and look up any product's latest certification — including what was not tested.
- **Eclipse Muster** — CERTIFIED (local). The shared core of the Eclipse customer family: a canonical Customer Graph (Person, Company, Deal, Ticket, Conversation, Activity) with provenance on every field and a Data Rights Ledger on every record. Identity merges are propose-only and exactly reversible; reply drafts cite record facts or mark them MISSING_FACT; instruction-shaped customer text is quarantined as data under Herald Security Law #1.
- **Eclipse Haven** — CERTIFIED (simulator). The governed Zendesk server, built over the platform's REST API v2 shape — the largest support base has no official MCP server. OAuth-first by construction: API-token auth is not implemented at all (Zendesk sunsets tokens 2027-04-30) and presenting one is refused by name on all 10 tools. Public replies send only as approved drafts; deletes, merges, bulk edits, and refunds are prohibited in v1.
- **Eclipse Orbit** — CERTIFIED (simulator). The governing proxy in front of HubSpot's official MCP server — the Eclipse WRAP posture. Vendor tools pass a pinned classification registry: unclassified tools are invisible, T3 tools are never surfaced, and a vendor tool whose tier rises is halted with a notice receipt until an owner clears it by name. T2 sends require approval tokens bound to the exact arguments.
- **Eclipse Voice Core** — CERTIFIED (local). The shared core of the Eclipse voice & telephony family: the Consent Registry and Number Registry every telephony adapter obeys. Consent is provenance-bound — a record without provenance is void; revocation is one-way and effective at the next check everywhere, because adapters hold no consent copies. The send gate evaluates consent first, then an Agent Wallet-shaped mandate; an unconsented attempt writes a violation record and nothing else.
- **Eclipse Trunk** — CERTIFIED (simulator). The governed Twilio adapter: raw carrier operations are never exposed — log reads with numbers masked beyond last-4, simulated sends, and recording policy, every send gated through Voice Core (consent, then mandate, then number sendability). Simulated human calls require a named human speaker; there is no AI voice in this adapter.
- **Eclipse Connect MCP** — CERTIFIED (dogfood-facade). The MCP face of Eclipse Connect, the hosted OAuth bridge: agents upload, publish, check grant status, and revoke through a revocable agent key — never raw OAuth. Going public requires an explicit human approval flag, enforced in the facade before any network call and again at the relay. A distinct surface from the relay itself, and benched as one: the relay holds its own CERTIFIED(dogfood).
- **Eclipse Envoy** — CERTIFIED (local). The signed front door for agent-to-agent business: one owner-approved identity source becomes a Herald-signed corporate Agent Card; counterparties are vetted into risk tiers R0–R3 (typosquats get their own record and never touch the honest org’s history), disclosures render per tier with an owner-entered NDA gate, and engagements bind only by the three-key rule — owner approval + Wallet mandate + Herald delegation. No tool on either surface signs, commits, or moves funds.
- **Eclipse Fathom** — CERTIFIED (local). DeFi, measured honestly: a signed, versioned Protocol Registry (Uniswap/Aave/Compound/Lido on Ethereum/Arbitrum/Base; audit status as dated data, never scraped into trust), dual-source position reads that render DIVERGENT instead of averaging, depeg watch, lookalike-address checks, and the flagship allowance audit that ranks unlimited ERC-20 + Permit2 grants by value at risk. Reads only — no execution, draft, or signer exists, and key-shaped input is refused by name.
- **Eclipse Codex** — CERTIFIED (local). Banking regulation as data: a versioned Regulation Library (records never overwritten, as-of resolution, NPRM text never resolves as law), an Obligations Mapper whose merge guard keeps the CDD Rule and CTA BOI duties separate, a Deadline Engine that never lets a guidance cadence render as a breach, local sanctions screening with no auto-clear, and evidence packs with a mandatory gap section — suppression is refused and receipted. Not a bank, not counsel, not a compliance department: it files nothing and determines nothing.
- **Eclipse Ledger** — CERTIFIED (local). Betting records and responsible-gambling controls. Not a sportsbook. No bets are placed here. Limits fall instantly but rise only after the stricter of the suite floor or the state’s delay plus a fresh confirmation; self-exclusion is absolute, honored identically from any source, with bookkeeping preserved; the jurisdiction matrix treats sportsbook, DFS, and pick’em variants as distinct legal objects; and P&L is honest cash math — a bonus bet’s stake is never counted as money. No placement tool, no account connection, no credential accepted.
- **Eclipse Till** — CERTIFIED (local). Bazaar checkout for agents, as its own server — the discovery Bazaar MCP’s certified no-purchasing surface is untouched. Quotes bind catalog + pricing version + terms and state the all-in total; the mandate gate checks cap against the all-in, not the sticker; over-threshold purchases park for the buyer’s own approval with no pressure verbs anywhere; entitlements are Herald-signed and buyer-bound; and a missed delivery window fires refund + revocation automatically, with no seller override. B0 settles sandbox test credits only — no real rail code exists in the build.
- **Eclipse Book** — CERTIFIED (local). The Markets core: one canonical Instrument / Book / Quote / Order / Position / Resolution model across exchanges and prediction markets. Criteria are stored verbatim and hash-pinned — a venue edit is a CRITERIA_CHANGED event and positions keep their criteria as of entry; quotes carry all-in cost from versioned fee schedules or degrade to FEE_UNVERIFIED; resolutions verify only against the criteria’s own named source, with DISPUTED first-class; and no output ever carries a probability field — the guard is code, not policy. Market data + venue access layer — not an exchange.
- **Eclipse Forecast** — CERTIFIED (local). The Polymarket adapter: Gamma/CLOB-shaped data normalized into the Book core’s canonical instruments and books, token/outcome mapping pinned from the venue’s own arrays, criteria verbatim + hash-pinned, fee-truth quotes, and UMA-shaped resolutions with DISPUTED as a first-class state. Every output carries money_class: REAL. The US trading surface is absent by design — no order, trade, account, or API-key tool exists, and circumvention-shaped names are refused by name.
- **Eclipse Strike** — CERTIFIED (local). The Coinbase public-data adapter: products and level-2 books normalized through the Book core (the same builders Forecast uses — cross-adapter parity is bench-proven), native statuses the core cannot map preserved verbatim as UNMAPPED, and quotes carrying all-in cost or FEE_UNVERIFIED. Coinbase ships its own official MCPs for accounts and wallets; Strike re-implements none of them — no account tools exist in this build.
- **Eclipse Corona** — CERTIFIED (local). Sealed render-proof for deployed websites: renders the served artifact across a pinned engine × viewport matrix, runs a closed vocabulary of mechanical probes at each cell, and seals measurements + screenshot hashes into a signed pack bound to the deployed artifact's hashes — recomputed by anyone. The publish gate may report 'verified live' only while a PASS served pack's hashes equal the deployed manifest. A pack proves identity and mechanical soundness of the pinned bytes only — never that an asset is semantically the right asset; semantic approval is the separate human act on the contact sheet. Name CONDITIONAL — trademark screen pending.
- **Eclipse Verbatim** — CERTIFIED (local). Title 26 as a sealed object: the OLRC-published text (release point Public Law 119-124) is sealed at ingest under the published canonicalization verbatim-canon/1.1 — per-section hashes for all 2,277 sections, dual Merkle roots, a signed manifest — and cite-check renders a byte-for-byte verdict (MATCH / DIVERGES with the first divergence localized / NOT_IN_CORPUS) against the sealed text; the bench's hostile battery produced 0 false MATCH. An independent implementation written from the spec alone reproduces the seal exactly, 2,277/2,277 in both the statutory and editorial layers; OLRC editorial notes are sealed separately and never served as statute. No advice surface — no tool interprets, explains, applies, or calculates anything; Title 26 statute only, no CFR; a dated snapshot, never described as current law. Signing uses labelled Herald fixture keys only. Name CONDITIONAL — trademark screen pending.
- **Eclipse Roll** — CERTIFIED (local). Filing prep + proof for business & government filings: a filing is prepared, approved, and evidenced as a Herald-signed delta between the registry-of-record's own current state — read back through its sanctioned public surface at prep time — and the exact artifact to be submitted; after the human submits via the registry's own channel, Proof re-reads the registry and seals proof-of-filing only if the public record changed exactly as approved. No submission verb exists on any surface, in any name — prep-and-seal only, the human always submits; advice, selection, interpretation, and clearance requests refuse at the tool layer by name (ADVICE_CLASS_REFUSED). Certified at FIXTURE scope: every Gauge adapter is a labelled fixture and no live registry read has been exercised yet; dogfood subjects are CETech's own Florida annual report, SAM.gov renewal watch, and VetCert entry, as fixtures. First bench FAILED (23/34) and is retained on the record; the fix round closed D1–D10 and the scoped re-bench passed 17/17. Name CONDITIONAL — trademark screen pending.
- **Eclipse Verbatim — Executive Orders** — CERTIFIED (local). Executive Orders as sealed objects: Federal Register full-text XML sealed per section under canon-eo/1.0 into snapshot eo-20261022-26818c895dee — 1,315 orders, per-section hashes, a signed manifest — plus a disposition graph extracted from the orders' own sealed sections: 10,667 edges, each pinned to the hash of the asserting section and typed by provenance (TEXT / FR_NOTE / NARA_TABLE). Where the consulted sources disagree, the disagreement is sealed as a DIVERGENCE record — 823, OPEN, never resolved by the product. Coverage is measured, not assumed: sealed text runs from 2000 (the Federal Register full-text XML floor); 258 orders are EXCLUDED_NAMED, and pre-2000 orders ride in a labelled unsealed metadata+edges layer reaching back to 1931. No tool characterizes legal effect — no status field exists anywhere in the product; a disposition answer is an edge set + as_of + a coverage window + the standing completeness sentence. The snapshot seal is a labelled FIXTURE seal. Name CONDITIONAL — inherits the pending Eclipse Verbatim trademark screen.
- **Eclipse Flood Witness** — CERTIFIED (local). Flood evidence over sealed snapshots of FEMA's public National Flood Hazard Layer: NFHL county extracts are sealed per partition under flood-canon/1.0, and geometry determinations (FOOTPRINT / STRUCTURE_POINT / PARCEL — an address is refused as a subject) are minted as signed, independently recomputable receipts; the standalone verifier ships in the tree. Two Florida partitions are sealed — Leon (snapshot nfhl-fl-leon-20260611-bec93b592c31; 12,099 features) and Monroe (nfhl-fl-monroe-20261022-4776f79b6d7c; 9,664 features); the operative index enumerates all 67 Florida counties and names the other 65 NOT_SEALED — Miami-Dade is NOT_SEALED (pending). Evidence-grade only, by construction: never a determination of record, never an SFHDF, never lender-compliant, never an insurance answer, and never a statement that a property is "safe." Flag-1 qualifier: the LOMC override branch is fixture-proven only — the sealed corpus holds zero LOMA/LOMR-F records and the branch has never fired on real sealed data. One LOW (FW-D8) stands as a printed pre-second-vintage condition. All seals are FIXTURE_SIGNED and labelled as such; nothing here states or implies FEMA endorsement — FEMA/NFHL name the public data source only. Name CONDITIONAL — trademark screen pending.
- **Eclipse Recall Witness** — CERTIFIED (local). Recall campaigns, witnessed: NHTSA serves only the current state of a campaign — amended filings replace records in place and no vintage parameter exists — so this product seals the recalls flat file nightly into a hash-chained anchor and seals every campaign's full payload as an immutable vintage under recall-canon/1.0; a campaign that disappears seals as a tombstone, never a silent drop. Drift between vintages seals as dated events across the closed recall-events/1.1 grammar (POPULATION_EXPANSION · REMEDY_CHANGE · ESCALATION_DO_NOT_DRIVE · ESCALATION_PARK_OUTSIDE · CORRECTION · CLOSURE) — an escalation is never absorbed into an expansion. Posture-at-date derives exclusively from sealed vintages (NOT_SEALED before the first vintage, with the campaign's own first-vintage date), and mandate-bound check receipts print scope_class: CAMPAIGN-LEVEL with the three standing sentences byte-exact — a specific VIN is never called “open” and per-VIN unrepaired status is never determined (the consumer VIN lookup is never automated; no such code path exists). CERTIFIED at FIXTURE scope: the nightly spine is fixture-proven and no live nightly seal exists; live cross-check and one live check issuance are proven against api.nhtsa.gov (24V809 seals cross_check OK, its DO_NOT_DRIVE tag agreeing with parkIt: true; rcpt-00001 for a Porsche 911 MY2024 recomputes MATCH) — and the spine was NOT re-routed to the API while www-odi.nhtsa.dot.gov is unreachable, because the flat file is the canon source; LIVE scope waits on the first real nightly ingest and live-key cutover. Churn is MEASURING — counters seal from night one, no rates before a day-90 segment seals. Three LOW conditions stand printed on the bench card, all riding the recall-events/1.2 / next-seal touch: RW-D7 (narrative “parked outside” text can tag a false PARK_OUTSIDE — no live-stock instance), RW-D8 (genesis links not yet asserted in the chain walk), RW-D9 (operate/use paraphrase families untagged — no escalation-completeness claim beyond the live-stock-proven phrase families). All seals are under FIXTURE ECDSA keys, labelled on every artifact. Pricing is not yet set — no price points exist in the price book for this product; nothing is priced or activated. Name CONDITIONAL — trademark screen pending.
- **BidMatrix** — Spec — not built. Drop in a solicitation, get back the compliance matrix: Section L/M extraction, requirements checklist, deadlines, and set-aside summary.
- **Section 508 Evidence** — Spec — not built. WCAG + Section 508 audits that produce the deliverable federal buyers actually require: ACR/VPAT-ready evidence reports.
- **FedScope** — Spec — not built. Federal + state opportunity search with fit and newcomer scoring and plain-language summaries — not a raw API wrapper.
- **MCP Health Check** — Spec — not built. Scan an MCP server or agent configuration for security posture — auth, SSRF exposure, static keys — with a scored report and fixes. Will dogfood on our own servers first.
- **Legacy Recon** — Spec — not built. Point an agent at a codebase: architecture map, dependency / end-of-life risk list, and inputs for a modernization estimate.
- **Eclipse Mail** — Spec — not built. A real business email identity for an AI agent — governed, review-first sending with scoped grants, plain-language receipts, a kill switch, and an audit trail. (Working name, not locked.)
## Evidence, not adjectives (as published on the homepage)

- Public code — Our tools are open source (MIT) on GitHub: github.com/CYAN-ECLIPSE-TECHNOLOGIES-LLC
- Published scorecards — Every Test Bench verdict links to its scorecard — including what was not tested. The White Report MCP's live certification is the current example.
- A working media property — The White Report publishes crypto & AI news daily, with a public scoreboard that grades its own falsifiable calls as they resolve: https://thewhitereport.com/

## Straight about past performance (as published on the homepage)

Cyan Eclipse Technologies LLC was founded in 2024 and has no past performance as a firm — no federal awards, and we won't imply otherwise. What we offer instead: a founder with 15+ years of hands-on software engineering, key-personnel experience on demanding defense programs (detailed in our capability statement), and a public body of tested work you can verify today. The right first step is a small, fixed-scope pilot — a test sprint, a static-analysis pass, an MCP tool build — judged on the deliverable.

## Machine-readable surfaces

- Site descriptor: https://cetech.ai/agents.json
- Catalog MCP server (read-only, no auth): https://cetech.ai/mcp.php — tools list_products, search_products, get_product, get_company, get_registry_path
- MCP server card: https://cetech.ai/.well-known/mcp/server-card.json
- Agent Registry Discovery: https://cetech.ai/.well-known/ard.json
- For Agents & Builders page: https://cetech.ai/agents/