Example experience — local review
26 October 2026 · Local source refresh, not a new live deployment. Try this checkout's browser application · Run locally and inspect the protocol · AAP examples hub.
What changed
Choose a human need: research for a shop, translation of a product page or review of a dataset. These inert presets fill editable fields; they never approve terms. Immediate draft-field restore is available. Signed versions and evidence remain unchanged when another example is selected; changed draft fields require a new proposal and fresh acceptance before delivery actions become available.
The page now states the next useful action and keeps fingerprint and public-JSON copy controls visible. Clipboard refusal selects text for manual copying and reports that limitation. It does not report a successful copy when permission was refused. Export disclosure remains explicit: terms and public signing material always travel, evidence text only when selected, private keys never.
Observed local checks
| Check | Result and scope |
|---|---|
| Protocol, CLI and actual SDK interoperability | 12 tests pass, zero skips. New synthetic presets produce valid exact agreements; unknown preset names refuse. Historical protocol.mjs is unchanged. |
| Loopback server | Three tests pass. Five application assets are explicitly allowlisted, including the new inert examples module. Private files and foreign Host headers refuse; no application network API or browser storage is added. |
| Browser product journey | 12 named checks pass in a fresh official Chromium headless profile: visible choices; draft-field restore; offer/acceptance; immutable versions and refused tampering; delivery/dispute claims and scoped evidence export; separate Node verification of real browser signatures; wrong-pin refusal and read-only import; copy controls and denied-clipboard fallback; actual 1280×900 and emulated 390×844 viewport checks; local help/source-download checks; and the implementation HTML guide. |
| Browser safety and rendering | No page errors or unexpected external requests. Included HTML-looking evidence remains inert text. Both tested widths have no document overflow. Screenshots were inspected; an emulated mobile viewport is not a physical-phone or multi-browser qualification. |
Browser receipt · Actual synthetic browser export · Independent Node result · Protocol tests · HTTP tests.
The local launcher now also serves specifically named public help, source-download and evidence files and exact /site/ aliases. Relative guide links work offline without enabling directory listing or arbitrary paths. The three HTTP tests cover these positive routes alongside foreign-Host, traversal, private-file and POST refusals.
Reproduce
python3 run_demo.py --port 0 node --test tests/protocol.test.mjs python3 -B -m unittest discover -s tests -p 'test_*.py' -v
For the two actual SDK interoperability cases, set PYTHONPATH to the reviewed implementation checkout and AGREEMENT_REFERENCE_PYTHON to its installed interpreter, as described in the earlier review. Without that runtime those two cases skip; a skipped run is not the 12-test result stated here.
Remaining boundaries
The public browser demo uses its existing printable-ASCII profile. Broader installed agreement paths can use separately versioned features; this small UI does not activate them. You play both synthetic roles. Signatures prove carried statements, not real identity, quality, legal effect, observed external payment or latest global state. No AI model, production blockchain, real funds, real arbitrator or independent operator is exercised.
The older live deployment record and original review remain dated historical evidence. New source/download bytes must be published and checked on their actual host before calling this refreshed UI live. This review does not repeat the private implementation's complete regression or security qualification.