# 0617 — GRADE = CONFIRMED

- N sampled: 577 (target >= 500)
- Generated: 2026-09-15T20:14:31Z
- Chains: ethereum(1), base(8453), bnb(56), optimism(10), arbitrum(42161)
- Age range in final sample: 2026-01-29T15:26:44.475758+00:00 -> 2026-09-15T20:00:07+00:00 (span 229.19d)

## Source-verification correction (added 2026-09-16 by task 0624)

The pre-pinned claims and grade below are preserved verbatim as recorded on
2026-09-15. Task 0624 subsequently source-verified C2 by fetching 43 rows of
`dataset.csv` at the exact metadata URL 8004scan had captured in each row's
`services[0]`. **4 of the 18 rows that resolved to a valid JSON agent-card
carry a non-empty `capabilities` field**, falsifying the simple reading of C2.

- **What C2 correctly says**: `capabilities` was null on 100 % of the N=577
  sample **as captured by 8004scan on its 2026-06-06 bulk parse** (established
  by task 0623 as the last observed re-parse; the index has not re-parsed
  since).
- **What C2 does NOT say**: that 100 % of the underlying ERC-8004 agents
  publish empty `capabilities`. Four counter-examples (Otto AI on Ethereum,
  Normie #2613 and #1444 on Ethereum, Arca on Optimism) publish non-empty
  `capabilities` at the URL 8004scan itself already knows about. The 4-of-18
  observation is a **falsifier**, not a population rate — it disproves "100 %"
  under C2's own pre-registered rule; it does not measure how many agents
  publish. Sample accounting: 43 sampled, 18 resolvable, 25 unresolvable (20
  not_json, 3 http_404, 1 http_429, 1 network error).

Full audit trail at `docs/research/0624-c2-provenance/README.md` and
`.cc-agent/.0624-c2-provenance-status.json`.

## Pre-pinned claims

- **C1** (threshold): endpoint_less_pct >= 60 AND median_total_score <= 15  
  Held: **False** — pct_endpoint_less=43.85% (bar >=60), median_total_score=12.02 (bar <=15)

- **C2** (categorical): capabilities AND agent_url null on 100% of the sample  
  Held: **True** — capabilities_null_pct=100.0% AND agent_url_null_pct=100.0% (bar 100% both)  
  ⚠️ Source-verified 2026-09-16 by task 0624: the "sample" here is the
  8004scan snapshot, not the underlying agents; the simple reading of C2 is
  falsified by 4 of 18 resolvable live cards. See "Source-verification
  correction" above.

- **C3** (categorical): is_endpoint_verified true on 0 records  
  Held: **False** — is_endpoint_verified_true=1 (bar =0)

## Grade

**CONFIRMED** — C1=FAILED [pct_endpoint_less=43.85% (bar >=60), median_total_score=12.02 (bar <=15)] | C2=HELD [capabilities_null_pct=100.0% AND agent_url_null_pct=100.0% (bar 100% both)] | C3=FAILED [is_endpoint_verified_true=1 (bar =0)]

The grade is preserved as recorded. Downstream users of the dataset should
consult the "Source-verification correction" section above before citing C2
as a property of ERC-8004 agents; it is a property of the 8004scan index
snapshot on 2026-09-15.