Sweep Cycle 3: Record Claim Intake, Still Zero Promotions

The cycle-two report ended with a question: would the third cycle convert a growing audit trail into a moving grade column? The answer arrived on October 5, and it has two halves again. The intake side set a record — 38 candidates surfaced across 14 projects, 15 of them appended as dated history entries to 10 of the 17 tracked providers — the largest claim flow the tracker has logged. The verification side did not move: 45 check results were recorded, 16 checks actually attempted an observation, and not one produced evidence strong enough to promote a grade. The export generated 2026-10-05 still reads 0 verified, 2 corroborated. This cycle also carries a wrinkle the first two did not: a recorded two-week gap, because no sweep completed for the week of 2026-09-28.

Key facts

• 3 of 3 cycles completed, over fourteen days this time. All 17 provider records now carry last_checked: 2026-10-05 (commit 82f3546). No sweep completed for 2026-09-28, so this cycle covers a two-week claim window — 33 of its 38 candidates were posted on 2026-09-29 or later.
• Record intake: 38 candidates, 15 entries appended. By provider: Chutes, io.net, NEAR AI, Phala, and SolRouter 2 each; Dolphin, Engy, OpenServ, Targon, and Venice AI 1 each. Cumulative history: 17 → 29 → 39 → 54 entries across the four dated runs.
• A 2026-09-28 run never happened. The digest chain jumps from 2026-09-21 to 2026-10-05. The series records the gap rather than averaging across it, the same treatment the Price Index gave its own missing 2026-09-28 snapshot.
• 45 checks logged, 16 attempted, 0 promotions. Attempted: 1 direct probe (Morpheus, HTTP 400), 2 gated probes (Venice, HTTP 402), 4 attestation checks (Phala), 9 OpenRouter lookups (no cross-check surface). The other 29 were blocked before starting: 12 API probes and 4 OpenRouter pulls on unset configuration, 13 on-chain queries on the unbuilt chain adapter.
• The grade census is unchanged: 62 claimed, 2 corroborated (both Chutes API documentation), 130 not-yet-checked, 0 verified across 17 providers.
• 1 Morpheus item hit the neutrality gate and was never written. The project's own node v7.14.0 release post routed to the backer-neutrality gate — the third consecutive cycle exercising that code path.

What cycle 3 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 is claimed):

Five of the fifteen entries are numeric scale or cost claims; one is a privacy-mechanism claim; the rest are launch and record items. All fifteen carry their X source URL in the record.

The chart: intake sets a record, the check funnel splits

The intake doubling is real but partly calendar: fourteen days of claim flow produced 38 candidates where single weeks had produced 19 and 19. The more honest comparison is the check funnel. Attempted checks rose from 6 to 10 to 16, but the character of the attempts changed — cycle 2's probes hit live endpoints and got answers (429s, a 402, a 400); cycle 3's attempts were mostly lookups that found no cross-check surface: nine of the sixteen were OpenRouter queries confirming the claimed provider is not listed there, and four were attestation checks finding nothing published to validate.

Why nothing moved — and what the funnel says about the harness

Every attempted check resolved to claimed, each with a named reason recorded in the resolved file:

• No validatable attestation, Phala ×4. Phala's privacy claims are the most specific in the cycle — hardware attestation, TLS key pinning, fail-closed forwarding — and the attestation checker still found nothing published it could validate. The standing rule holds: an absence-claim is never verified from a policy statement.
• Not on OpenRouter, ×9. io.net, NEAR AI, Engy, SolRouter, Lium, and Akash volume claims have no public cross-check surface. Lium, a Tier-3 watchlist item, still has no tracker record file at all — its claimed SN51 revenue report stays in the human queue.
• Gated (HTTP 402), Venice ×2. Unchanged from cycle 2: the endpoint requires an access key the harness does not hold.
• Probe rejected (HTTP 400), Morpheus ×1. The marketplace answered but rejected the probe's model/config — an answered-but-invalid response is not a liveness contradiction, so the record notes the raw outcome and the claim stays claimed.

The un-attempted 29 are the real headline for anyone watching the zero column. Twelve API probes and four OpenRouter pulls never ran because api_base remains unset for NEAR AI, OpenServ, io.net, and Targon; thirteen on-chain queries wait on a chain adapter that has been "pending" since cycle 1. The blockers are configuration and plumbing, not contradictions — which is exactly why none of them could produce a disputed either. A claim that was never checked keeps its grade; it does not lose one.

The harness fix, on the record

Cycle 3 also shipped a rule change with retroactive significance, recorded in the run digest: HTTP 429/5xx and 400/404/422 responses from direct probes were previously being graded disputed — the contradiction grade. Those codes mean the server answered (rate-limited, transient, or rejecting our probe request), which is not a liveness contradiction. They now resolve claimed with an explicit probe note, and only connection failure or timeout across spaced attempts yields disputed. The fix prevented two false contradiction entries (Chutes 429, Morpheus 400) from entering the append-only history. A verification system's error modes deserve the same audit trail as its results: this one is dated, and the two affected claims are named.

The gate that never writes

One candidate reached the tracker's edge and stopped there, by design. Morpheus' own node v7.14.0 stability release — a status post from the project itself, the kind of item that would have been logged for any other provider — was classified favorable-by-subject and routed to the backer-neutrality gate, which never writes it into the history. Cycle two exercised the same path on a third-party setup post; cycle one logged zero Morpheus items. The rule is not that Morpheus is excluded — it sits in the tracker under the identical five-grade rubric — it is that no favorable item about the publication's funder gets an automated write, from anyone, including the funder. Three cycles in, the gate has fired three times and written zero times.

What has to happen before the zero column moves

The requirements are unchanged from cycle two, and the funnel now quantifies them precisely. A second key in the environment store unblocks the two Venice 402 probes overnight. Four api_base values unlock twelve API probes. The chain adapter would put thirteen on-chain queries to work — including Lium's and Akash's claims, which currently die by arithmetic (no roster record, no cross-check surface) rather than judgment. The OpenRouter puller remains the highest-leverage build: it is what would let Phala's token counters and Chutes' volume figures reach corroborated the way the Price Index grows — by dated, reproducible snapshots. Until then the chart holds: four dated runs, 54 source-backed entries, 32 checks attempted across three cycles, zero promotions — each zero now carrying a named, recorded reason.