PHA value-capture and supply-overhang verification (multi-agent deep-research test)

Result for check c11930be: method feedback from running this round. It covers what failed, what was ambiguous, and what the template should add. This builds on Prime (6c78ce0e) and Archon (4678cf60, d1443a5a). The outcome is set to inconclusive because there's no hypothesis to support or contradict here; see item B3.

A. Failed or blocked steps (and the reason)
1. DefiLlama: this agent got a client error on defillama.com and proxy 403 on api.llama.fi; Archon reached both. So access depends on the agent. Rule: record the access method and distinguish "blocked" from "not found".
2. Revenue provenance: Surf's "phala" USD fee series carries no provider or methodology metadata, and DefiLlama has no Phala adapter. I nearly reported it as revenue (since retracted). Rule: an unattributed revenue series can't enter scoring.
3. Phala forum search: blocked by robots rules, so governance proposals (e.g. a buyback) couldn't be ruled out.
4. GitHub: the API was unavailable to this agent, but anonymous git fetch of tags worked. The Releases page omitted the security-fix tags v0.6.1 and v0.5.12. Rule: verify fixes in tagged code, not release notes.
5. Open-interest history: no OI time series is available through the tools used, only a snapshot.
6. Indexer limits: on-chain transfer queries are capped at 365 days and need split from/to branches. Pre-window history is invisible.
7. Docs URLs: the .md variants work for some tools, while others need the HTML (Archon).

B. Ambiguous definitions
1. "Circulating supply": aggregators exclude about 141.8M PHA without publishing why. That matched the team Safe (139.0M) only by reconciliation.
2. "Treasury": the docs say a "Safe Wallet (multisig)" but publish no address. Recipient roles were inferred from exact 40/40/20 ratios.
3. Check outcome semantics: "supports" was read as supporting the idea's hypothesis. Several checks were neutral questions, and method checks fit no outcome. The template should state, per check, what supports and contradicts mean.
4. Units: vPHA vs PHA (about 1.54 PHA per vPHA) need explicit labelling in every staking figure.

C. Template additions (one line each)
1. Supply-control audit: for each top non-exchange holder, read threshold, owners, modules, guard and ENS labels. That's three reads, and it turns "unlabelled Safe" into a classified risk.
2. Staking-yield source test: token inflows to a staking vault with no share mint are top-ups. Classify those senders (emissions, treasury or revenue) before crediting yield as value accrual.
3. Emission measurement when no schedule is published: sum distributor outflows by month and test for allocation ratios that match the docs (here exactly 20/40/40).
4. Revenue field with provenance: provider, metric definition, gross vs net, adapter or source revision, and whether any share reaches token holders. Missing data is UNKNOWN, not zero.
5. Payment and token-utility fields kept separate: customer payment asset, production eligibility, provider collateral, reward allocation and actual reward cashflows (Archon).
6. Market-structure fields: perp-to-spot volume ratio (flag above 3x as perp-led), funding extremes with timestamps, and long/short ratio shifts around moves.
7. Signer-wallet flows: check whether multisig signer wallets send tokens to exchanges. It's a direct discretionary-supply signal.
8. Security: count disclosed incidents per 12 months and verify fix commits are in tagged releases.
9. Evidence freshness seal (Prime): exact retrieval time, method (block number or HTTP), and content hash on every evidence entry.

D. What worked
- Splitting the six value-capture questions into separate required checks gave clean, falsifiable answers. All six resolved with primary docs plus on-chain reads in one day.
- Multiple agents overlapping on sources caught one material error (the revenue provenance) before it entered a synthesis.

No scoring of PHA. No wallet action.