The cycle-eight report closed on a record: 18 attempted checks, the funnel's high-water mark, and zero usable observations. The 2026-11-16 sweep — the ninth dated run, one week later — kept the zero and broke the funnel in the other direction. Cycle 9 appended 11 history entries to 8 of 17 providers, growing the ledger to 138 entries, and logged 30 checks of which only 9 were attempted — every one a blind OpenRouter pull against a provider with no listing there. Not one check this cycle touched a provider-controlled surface, and the grade census still reads 0 verified.
Key facts
- Nine of nine dated runs completed. All 17 provider records carry
last_checked: 2026-11-16(site commit0133209, "11 changes appended"). The weekly cadence has held six consecutive runs — 10-05, 10-12, 10-26, 11-02, 11-09, 11-16 — with the digest chain still showing no 10-19 file. - 11 entries appended to 8 providers. Cumulative history: 17 → 29 → 39 → 54 → 62 → 85 → 104 → 127 → 138 entries across the nine dated runs. Akash Network, Phala Network and OpenServ took 2 each; Chutes, Darkbloom, Dolphin Network, NEAR AI and Warden Protocol took 1.
- 22 candidates resolved across 11 projects — the narrowest spread since cycle 4. The apply step collapsed 5 same-day repeat groups and held 6 candidates out of the history: two from Gensyn and three from Lium (projects with no tracker record file, both weeks running for Lium) and one third-party Targon scale claim.
- 30 checks logged, 9 attempted — the lowest attempt share in the funnel's history. The recomputed series: 6 → 10 → 16 → 8 → 12 → 17 → 18 → 9. Cycle 9's share (30%) is the first below 43%, and all nine attempts were openrouter-pulls returning "not on OpenRouter."
- 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 9 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):
| Provider | Entries | What was claimed (from the history log) |
|---|---|---|
| Akash Network | 2 | AKT live on Solana via Sunrise as migration from Cosmos begins (third-party observed); urgent governance parameter-change Proposal 349 opened for voting (third-party observed) |
| Phala Network | 2 | Kimi K3.1 Pro live on Phala TEEs, Lighthouse Private Memory live on Robinhood Chain; weekly Lighthouse wrap (Nov 9–15) covering Private Memory's first week |
| OpenServ | 2 | SERV platform framing — "reasoning that outperforms frontier models at up to 90x lower cost," SOC 2 underway, AWS Marketplace as first enterprise channel; two-year autonomous-agent R&D narrative |
| Chutes | 1 | Third-party Bittensor whale-risk ranking: Chutes leads SN64 on liquid concentration and overhang risk among 126 subnets |
| Darkbloom | 1 | Third-party side-by-side: arete-27b evaluated for Darkbloom on an M4 Pro Mac, ~15.5 tok/s at 4-bit MLX versus its qwen3.8-27b base |
| Dolphin Network | 1 | Roadmap claims: almost 1,000 GPUs online, DeepSeek migration, claim-plus-payments merge into v2.dphn.ai, new payment chains, 1.3x boost for bonded operators |
| NEAR AI | 1 | Community amplification: "private inference can cost less than the median while running open models in TEEs with verifiable attestation" |
| Warden Protocol | 1 | Warden MCP launch: credential-free market reads — trending tokens, price history, swap quotes across chains |
Nineteen of the twenty-two resolved candidates came from third-party accounts rather than provider handles — cycle 8's share was 4 of 35. Ten of the candidates are scale or roadmap claims, three are governance or token items, three are launches. Every one is still a self-report or a third-party echo of one. The Price Index ran the same week and logged its own first: six movers, all scissors-shaped, with no first-party confirmation for any row.
The chart: nine cycles of the same funnel
| Cycle | Window | Surfaced | Candidates | Collapsed | Appended | Checks logged | Checks attempted | Positive evidence | Promoted |
|---|---|---|---|---|---|---|---|---|---|
| 1 (2026-09-14) | 1 week | 19 | 13 | — | 12 | 17 | 6 | 0 | 0 |
| 2 (2026-09-21) | 1 week | 19 | 19 | 4 | 10 | 24 | 10 | 0 | 0 |
| 3 (2026-10-05) | 2 weeks | 38 | 38 | 16 | 15 | 45 | 16 | 0 | 0 |
| 4 (2026-10-12) | 1 week | 16 | 16 | 1 | 8 | 20 | 8 | 0 | 0 |
| 6 (2026-10-26) | 2 weeks | 31 | 31 | 3 | 23 | 33 | 12 | 0 | 0 |
| 7 (2026-11-02) | 1 week | 33 | 33 | 8 | 19 | 37 | 17 | 2 | 0 |
| 8 (2026-11-09) | 1 week | 35 | 35 | 4 | 23 | 38 | 18 | 0 | 0 |
| 9 (2026-11-16) | 1 week | 22 | 22 | 5 | 11 | 30 | 9 | 0 | 0 |
The attempted column now reads 6 → 10 → 16 → 8 → 12 → 17 → 18 → 9. Cycle 9 did not merely set a low bar — it is the first cycle in which no attempted check could in principle have observed provider-controlled evidence. Every prior cycle's attempted mix included direct probes, attestation checks or pricing fetches; cycle 9's nine attempts were all OpenRouter pulls whose outcome ("not on OpenRouter") was already determined by which claims the week surfaced. The funnel did not fail harder; it showed up with a deck of cards it cannot win with.
Where the 21 blocked checks died: the same three walls, now the whole story
All 13 on-chain queries hit the chain adapter that has been pending since cycle 1 — the largest blocked bucket in eight of nine cycles now (cycles 1–7 and 9; cycle 8's nine no-endpoint probes tied it). All 7 API probes had no public endpoint configured. One OpenRouter puller itself was unbuilt for a project slug, the first time that failure mode has appeared. Add them up: 21 of 30 checks (70%) never left the ground for structural reasons — the highest blocked share of any cycle.
The contrast with cycle 8 is instructive in the other direction too. Cycle 8's attempts were mostly rejected at the door — nine HTTP 429s and four HTTP 402s from real endpoints. Cycle 9's attempts never reached a door: a blind pull against an aggregator that has never listed the provider confirms nothing about the claim. Cycle 7's pricing-fetch method, the only one that has ever produced hashed positive evidence, had no candidates to run this cycle — the per-claim constraint the cycle 8 report identified, now visible in the method mix itself.
The gates that never write
No Morpheus item surfaced this cycle — the seventh cycle of nine where the favorable-by-subject gate had nothing to intercept, and its write count stays at zero across all nine runs. No incidents surfaced either. Venice, io.net, Nesa, SolRouter, Oasis, SOLAI and Morpheus produced no history changes at all; Morpheus's last_changed has now held at 2026-08-24 through nine sweeps, its single history row still the ledger's oldest entry. Nothing was promoted, nothing was demoted, and no history row was rewritten — the ledger is append-only, and cycle 9 only appended.
What has to happen before the zero column moves
The unblock list is unchanged for the ninth consecutive cycle, and cycle 9 shows what that costs when the week's claims land on the wrong side of every wall. The chain adapter has 13 fresh reasons to be built. The endpoint-config gap grows by 7 checks a cycle and is now the second-largest blocked bucket all-time. And the deepest problem this cycle exposed is compositional: when a week's candidates are all third-party scale chatter and governance echoes, the funnel's usable methods have nothing to attach to, and the cycle's entire observational budget collapses into nine blind aggregator pulls. A check that cannot reach a provider-controlled surface — an attestation that validates, an endpoint that answers, a chain that confirms, a machine-readable price page to diff — is structurally incapable of promoting anything, no matter how many cycles run.
What the nine-cycle series now shows is a funnel whose promotion column has read zero across 96 attempted checks in total, with its only positive evidence hashed in a single cycle. Until the adapter, the endpoint configs, or the comparison surfaces exist, the correct tracker state is exactly what the export shows: everything claimed, nothing verified, and a funnel that records precisely why — check by check, wall by wall, in every cycle.
Questions
- What did the tracker's ninth sweep cycle find?
- The 2026-11-16 weekly sweep resolved 22 candidates across 11 projects and appended 11 dated history entries to 8 of the 17 tracked providers, growing cumulative history from 127 to 138 entries across nine dated runs. Thirty checks were logged and only 9 attempted a real observation — an attempt share of 30 percent, the lowest in the funnel's history — and all nine were blind OpenRouter cross-checks that found nothing to compare against. The grade census still reads 0 verified and 2 corroborated.
- Why were only 9 of 30 checks attempted in cycle 9?
- The check mix collapsed toward checks that cannot run at all. All 13 on-chain queries hit the chain adapter that has been pending since cycle 1, all 7 API probes had no public endpoint configured, and one OpenRouter puller itself was unbuilt for a project slug. That left 9 OpenRouter pulls that ran — and each confirmed the claimed provider has no listing there, so no cross-check surface exists. Twenty-one of 30 checks never left the ground for structural reasons.
- Did any provider's verification grade move this cycle?
- No. The export generated 2026-11-16 still shows 0 verified and 2 corroborated field values — both Chutes API documentation fields, dated 2026-08-24 — across 17 providers. All 11 new history entries entered at grade claimed, and cycle 7's two pricing corroborations remain the funnel's only positive evidence.
- What claims did cycle 9 append to the ledger?
- The largest write went to Akash Network (2 entries): AKT going live on Solana via Sunrise as it migrates from Cosmos, and an urgent governance parameter-change Proposal 349. Phala Network and OpenServ took 2 each; Chutes, Darkbloom, Dolphin Network, NEAR AI and Warden Protocol took 1 each. Lium and Gensyn candidates again found no tracker record file and stayed in the human queue.
- How does cycle 9 compare to cycle 8's attempt record?
- Cycle 8 set the absolute record (18 attempts of 38 logged) but its share was 47 percent. Cycle 9's 9 of 30 is a 30 percent share — the lowest of any cycle — because cycle 8's usable methods (direct probes, attestation checks, pricing fetches) simply had no candidates this week. The absolute count fell but the share fell faster, and the blind-pull-only mix means no check even reached a provider-controlled surface.
Sources
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.