Three valid responses. An 81-minute-old quote.
In the recorded local browser run, all three bounded app requests returned valid evidence with a 5 ms median response time. Repeated text and geometry rendering stayed stable, and a known 4 × 4 pixel pattern matched its readback. Yet the quote embedded in that evidence was 81 minutes old.
Transport success, data freshness and economic demand are different observations. One green status cannot establish all three.
These are dated local measurements, not current performance numbers. The public demo preserves the observation timestamps and labels the evidence as a snapshot.
Give each claim its own evidence.
| Signal | What is checked | What it does not prove |
|---|---|---|
| App health | A bounded response arrives and matches the expected schema. | A fresh quote or healthy upstream RPC. |
| Freshness | The observation timestamp, age and refresh state. | A future price or realized return. |
| Outside demand | Covered token holdings and attributed on-chain changes. | One browser visit equals one buyer. |
| Rendering | Repeated local scenes and a known pixel readback. | A unique identity or evidence of tampering. |
| Return accounting | Integer lamports, fees, recovered deposits and retained balances. | Recovered capital is trading profit. |
Bound the work. Keep the last good evidence honest.
The browser diagnostics start only after a click. Each run makes at most three fixed same-origin requests, uses a three-second timeout and caps each response at 32 KiB. Credentials are omitted. Early rejection cancels the response body; a hidden tab cancels the run. A generation counter prevents a cancelled run from overwriting a newer one.
Canvas checks keep pixel buffers in memory only and discard them after comparison. The resulting labels describe repeatability and readback. Privacy features can modify pixels, so variation is not automatically called an attack.
The operator application uses a separate read-only refresh service. Public quote, holder and trade observations have independent timestamps and refresh intervals. Failed refreshes retain explicitly aged prior evidence; missing values remain unknown. The portfolio demo does not connect to that private operator service.
start → render locally → request known evidence
response arrived → health result
observation timestamp → freshness result
covered chain snapshot → demand result
hidden or cancelled → abort → discard late completion
unknown value → unknown, never zeroMake the distinction visible.
- Open Return Lab and change the fee or retained-balance fields. The calculator uses integer lamports and rejects malformed amounts.
- Run Browser Insight. Compare the response-health result with the independently displayed evidence age.
- Read the timestamped holder snapshot. Commonly controlled wallets are disclosed; wallet count is not presented as community size.
- Inspect the JavaScript. The public demo contains no signer, credential, private RPC endpoint or transaction control.
A clearer signal, not a larger claim.
The browser diagnostics and calculator are runnable public work. The chain evidence is a dated snapshot from a real experiment. Neither the public page nor a successful health probe promises a financial outcome. The local operator's signing and settlement components are deliberately outside this public release.
The transferable engineering is the boundary: validate provenance and freshness separately, cancel stale work, preserve unknown states, and make the reason for a displayed result inspectable.
PRIMARY REFERENCES
- W3C Beacon — queue acceptance is not an acknowledged application response.
- Page Visibility API — hidden-page lifecycle and throttling.
- fingerprint.garden canvas collection — captured rendering and readback examples.
- FingerprintJS canvas source — separate text and geometry behavior.