Independent/Reader-funded/Infrastructure, not tokens
DeAINEWS

AI you control — open models, private inference, and the networks that run them.

Provider Policy & Trust

Sweep Cycle 8: Record Check Attempts, Zero Usable Observations

The 2026-11-09 sweep appended 23 entries to 13 providers and set a funnel record: 18 checks attempted, none usable. The census still reads 0 verified.

DeAI is powered by Morpheus (mor.org). We cover competing providers on the same terms — see our methodology.

The cycle-seven report ended with the funnel's first positive evidence on the record: two pricing-fetch corroborations, hashed, after six cycles of nothing. The 2026-11-09 sweep — the eighth dated run, one week later — kept the volume and lost the evidence. Cycle 8 appended 23 history entries to 13 of 17 providers, tying the tracker's largest single write, and set a new attempt record: 18 of 38 logged checks tried to observe something. Not one produced usable evidence, and the grade census still reads 0 verified.

Key facts

  • Eight of eight dated runs completed. All 17 provider records carry last_checked: 2026-11-09 (site commit ef574a9, "23 changes appended"). The weekly cadence has now held five consecutive runs — 10-05, 10-12, 10-26, 11-02, 11-09 — with the digest chain still showing no 10-19 file.
  • 23 entries appended to 13 providers — the write ties cycle 6's record. Cumulative history: 17 → 29 → 39 → 54 → 62 → 85 → 104 → 127 entries across the eight dated runs. Chutes took 3; NEAR AI took 4; Akash and Warden took 2 each; nine providers took 1.
  • 35 candidates surfaced across 15 projects. Cycle 7 reached 18 projects; cycle 8's spread narrowed but stayed wide, with two candidates from Lium and one from Gensyn — projects with no tracker record file, held in the human queue.
  • 38 checks logged, 18 attempted — a new record, zero usable. The attempt count rose from 17 to 18, but the mix collapsed toward probes that could not answer: nine direct probes rate-limited (HTTP 429), four access-gated (HTTP 402), two attestation checks with nothing to validate, two pricing fetches with no machine-readable source, five OpenRouter pulls with no cross-check surface.
  • The grade census is unchanged: 62 claimed, 2 corroborated (both Chutes API documentation, dated 2026-08-24), 130 not-yet-checked, 0 verified across 17 providers.

What cycle 8 actually logged

The claims table, read from the export's history entries (every row is the provider or its community speaking about itself, which is why every row enters at claimed):

ProviderEntriesWhat was claimed (from the history log)
Chutes3Seven-model catalog expansion (DeepSeek V4 Pro, DeepSeek V4.1 Flash, Qwen3.8 Max, Qwen3.8 2.4T A95B, GLM 5.3, GLM 5.2 Fast, Kimi K2.7 Code); move to pay-as-you-go — "Plus and Pro are gone"; Tencent Cloud partnership for the Hunyuan flagship launch
NEAR AI4Qwen 3.8 27B Abliterated "live on NEAR AI Cloud" at a claimed $0.04/M input, $1.35/M output, 256K context; TEE-hosted "no one can see your prompts or outputs. Not even us."; Palo Alto Networks' Prisma AIRS listing the cloud as a model provider
Phala2Three TEE-served model launches (Voxtral Small 24B, Whisper Large v3, FLUX.2 klein 4B); claimed 120.50 billion tokens in 24 hours, Kimi K3 at 69.39% of it
Akash2Mainnet 19 live with the v2.2 upgrade (third-party observed); governance Proposal 349 opened for expedited voting (third-party observed)
Warden2Solana live on Warden Token Terminal and MCP; $WARD staking APR claimed at 12%, up from 5%
Venice1Four model launches collapsed to one entry: Krea 3, Kling 4.0 Flash, GPT Image 2.6, DeepSeek V4.1 Flash 1017 Ultrafast
OpenServ2Hitchhiker's Guide to Data Centres launch events in Amsterdam and London; hackathon submissions closing November 15
Nesa2Privacy positioning — "keeping data fully private throughout inference"; auditable-agent architecture claims
Targon1Third-party scale chatter: subnet revenue claimed approaching $1M
Darkbloom1Claimed milestone: "just passed 1 trillion tokens processed"
io.net1Claimed "70%+ of gateway tokens" on open-weight models, 4x spend share since June
Oasis1Investment activity: Lombard Systems' $4M pre-seed and seed raise
SolRouter1Router for Mac teased: "Works with Wi-Fi off"

Eight of the twenty-three entries are scale or pricing claims, three are privacy-mechanism claims — every one still a self-report. Four of the thirty-five candidates came from third-party accounts rather than provider handles; all four stayed at claimed with the third-party flag recorded. The Price Index ran the same week on the OpenRouter basket, giving the catalog side its own dated snapshot.

The chart: eight cycles of the same funnel

CycleWindowSurfacedCandidatesCollapsedAppendedChecks loggedChecks attemptedPositive evidencePromoted
1 (2026-09-14)1 week1913—1217600
2 (2026-09-21)1 week1919410241000
3 (2026-10-05)2 weeks38381615451600
4 (2026-10-12)1 week16161820800
6 (2026-10-26)2 weeks3131323331200
7 (2026-11-02)1 week3333819371720
8 (2026-11-09)1 week3535423381800

The attempted column is now the funnel's headline series: 6 → 10 → 16 → 8 → 12 → 17 → 18. Cycle 8's 18 attempts is the record, and its zero is the most instructive one yet, because the attempts were not blocked before contact — thirteen of them reached a real endpoint or check and were turned away at the door.

Where the 13 attempts died: a rate limit, a paywall, and an empty directory

Nine of the thirteen usable-path attempts were direct probes of Chutes' endpoint (llm.chutes.ai/v1), one per Chutes candidate. All nine returned HTTP 429 — rate-limited on sweep day. The harness records this as transient, explicitly not a contradiction; a probe that gets throttled says nothing about the claim it was probing. Four Venice probes returned HTTP 402, the paywall code: the models Venice announced exist behind access, and the probe carried no credentials. Two attestation checks (Nesa, NEAR AI) found no validatable attestation published — and the harness's standing rule holds that privacy absence-claims are never verified from a policy statement. Two pricing fetches (NEAR AI, Lium) had no machine-readable pricing source configured, the exact gap cycle 7's corroborations were supposed to motivate closing.

The five remaining attempts were OpenRouter pulls that confirmed the claimed provider has no listing there — the no-cross-check-surface outcome that has been the funnel's most common result since cycle 3. Add the blocked tail — nine API probes with no public endpoint configured and nine on-chain queries on the chain adapter that has been pending since cycle 1 — and 20 of 38 checks never left the ground for structural reasons.

The contrast with cycle 7 is one method row. Cycle 7's two pricing fetches found fetchable pages and corroborated their claims with hashes; cycle 8's two pricing fetches found no machine-readable source at all. The check type that produced the funnel's only positive evidence produced nothing a week later, because the evidence depends on the claim having a comparable published surface, not on the method existing. That is the honest reading: the funnel's constraint is per-claim, not per-method.

The gates that never write

No Morpheus item surfaced this cycle — the sixth cycle of eight where the favorable-by-subject gate had nothing to intercept, and its write count stays at zero across all eight runs. No incidents surfaced either: Dolphin, Engy, Morpheus and SolAI produced no history changes at all, their last_changed dates holding at earlier sweeps. Nothing was promoted, nothing was demoted, and no history row was rewritten — the ledger is append-only, and cycle 8 only appended.

What has to happen before the zero column moves

The unblock list is structurally identical to cycle 7's, which is the finding. The chain adapter has now been the largest blocked bucket in seven of eight cycles. The 429s are a single-retry problem: one probe on a different day or hour would likely get through, and nine Chutes candidates went unobserved because all nine probed the same endpoint in the same window. The 402s need credentials the harness does not hold. The attestation checks need providers to publish something validatable. The pricing fetches need the machine-readable comparison source named in the cycle 7 report — still unbuilt.

What the eight-cycle series now shows is a funnel that looks everywhere it can and keeps finding the same walls. Attempts have tripled (6 → 18), the method mix spans six check types, and one cycle produced hashed positive evidence — yet the promotion column reads zero across 87 attempted checks in total. Until a check observes something the provider does not control — an attestation that validates, an endpoint that answers, a chain that confirms — the correct tracker state is exactly what the export shows: everything claimed, nothing verified, and a funnel that now records precisely why, check by check, in every cycle.

Questions

What did the tracker's eighth sweep cycle find?
The 2026-11-09 weekly sweep surfaced 35 candidates across 15 projects and appended 23 dated history entries to 13 of the 17 tracked providers, tying cycle 6's largest single write and growing cumulative history from 104 to 127 entries. Thirty-eight checks were logged and 18 attempted a real observation — the funnel's high-water mark — but none produced usable evidence, so the grade census still reads 0 verified and 2 corroborated.
Why did 18 attempted checks produce nothing usable in cycle 8?
The attempts clustered where the funnel cannot see. Nine direct probes of Chutes' endpoint all returned HTTP 429 (a transient rate limit, recorded as not probeable, not a contradiction), and four Venice probes hit HTTP 402 (access-gated). Two attestation checks found no validatable attestation published, two pricing fetches had no machine-readable source configured, and five OpenRouter pulls confirmed no cross-check surface. Every zero has a named reason.
Did any provider's verification grade move this cycle?
No. The export generated 2026-11-10 still shows 0 verified and 2 corroborated field values — both Chutes API documentation, dated 2026-08-24 — across 17 providers. All 23 new history entries entered at grade claimed, and cycle 7's two pricing corroborations remain at the human gate rather than rising to verified.
How does cycle 8 compare to cycle 7's positive evidence?
Cycle 7 (2026-11-02) produced the funnel's first positive evidence: two pricing-fetch corroborations with SHA-256 hashes. Cycle 8 repeated the method mix but the pricing fetches found no machine-readable source (NEAR AI, Lium) instead of a fetchable page. The contrast shows the funnel's constraint is per-claim, not per-method: the same check type that corroborated in one cycle found nothing to compare against in the next.
Why did 35 candidates produce 23 history entries?
The apply step collapses repeated same-day posts of the same category into one entry, keeping the most detailed text: 3 same-day repeat groups collapsed 4 excess items. Three surfaced items — two Lium, one Gensyn — belong to projects with no tracker record file and stay in the human queue rather than the history. Chutes took the largest single-project write at 3 entries.

Sources

  1. DeAI Decentralized Inference Provider Tracker (JSON export, generated 2026-11-10) — DeAI News
  2. DeAI sweep digest, 2026-11-09 (weekly DIPT run) — DeAI News
  3. OpenRouter list-price delta, 2026-11-09 (Issue 8) — DeAI News
  4. DeAI Provider Trust Tracker — DeAI News

About DeAI

DeAI is an independent publication covering open-weight AI models, private inference, and decentralized infrastructure — the tools for running AI you actually control. We test providers on price, privacy, and refusal behavior and publish the numbers, not the vibes. DeAI is powered by Morpheus (mor.org), a decentralized inference marketplace, and covers it on the same terms as every other provider.

Powered by Morpheus and StrandCMS

Morpheus is a decentralized inference marketplace, covered on the same terms as every other provider — we rank it wherever the data lands. StrandCMS is the open-source, agent-first framework this site is built on.

Learn more about the Morpheus Inference API →

Sponsor disclosure — not editorial

Powered by Morpheus and StrandCMS. Morpheus is a decentralized inference marketplace, covered on the same terms as every other provider. StrandCMS is the open-source, agent-first framework this site is built on.

Learn more →