Current decisions — 2026-11-08

- Retain the BlueWallet fork and accept reviewed upstream changes rather than
  extracting a new repository now. Reusing upstream fixes is the chosen
  maintenance tradeoff; every merge still requires conflict and product review.
- Include an opt-in, index-mode pruned on-device Core node. Server mode remains
  the default. Signing keys stay in the app; Core receives public descriptors.
- The owner-approved changes supersede the historical feature freeze and
  extraction plan below. Preserve wallet/backup compatibility. Prioritize
  physical-device evidence before further feature expansion; do not infer
  release-gate completion from source review, emulator tests or informal use.
- Current Android preview: 0.1.20-preview, versionCode 38. Historical build
  numbers below identify earlier artifacts, not competing current versions.
  The wallet has no release tag yet; do not invent one or relabel an APK.
  Local website sources still link to actual 0.1.19-preview artifacts. Update
  the download, source archive, checksums and displayed version together when
  publishing a new artifact; do not change only the website badge.
- Embedded Core remains pinned to the full reviewed source commit documented
  in FULLNODE-PROTOTYPE.txt. Prefer an audited release tag when one is available;
  verify source identity, compatibility and native build evidence before moving
  the pin. A tag alone is not a consensus or security certification.

Historical decisions (retained for context; superseded where stated above)
=========================================================================

Backend-first follow-up decisions — 2026-10-19

1. Freeze wallet features and recovery/storage/QR formats until real backend and
   physical-device tests are complete. Source code cleanup is not a reason to
   silently change existing phone addresses or discard backups.
2. Keep m/84'/19333'/0' and current change discovery bounds as the existing-wallet
   recovery contract. A coin-type-0 switch is not Core wallet interoperability.
   Windows pairing must use actual descriptor/key-origin/script/network data.
3. v2 is still the current optional-wallet-password storage path, not merely an
   obsolete preview format. v3-only requires a product/migration decision about
   mandatory encryption. Do not remove readers before phone/backup validation.
4. Prototype ElectrumX's documented coin-class/session extension points, pinned
   to a reviewed revision, instead of claiming an unmodified Bitcoin server works.
   It reads blocks via RPC, so P2P magic-byte changes are not its main integration
   point. Core validates consensus; the adapter must still handle expiry/index
   semantics. Query-layer expiry filtering and shortened-prefix rollback are
   experiments, not a public backend certification.
5. electrs source was inspected, not ported or compiled. No claim is made that
   electrs/Fulcrum need only a magic-byte patch. No new private indexer is built.
6. Retain existing signer path/ownership/fee checks. BC-UR transport and Core
   descriptors are future interoperability work with separate validation; they
   do not automatically provide network or derivation compatibility.
7. Fix the measured discovery bottleneck during the freeze: bounded read batches,
   per-account public discovery hints and normal cursor windows. Preserve full
   change recovery 0..1999 on initial synchronization and manual request only
   (owner's 2026-10-22 change removes daily/time/reorg triggers). Do not alter accepted
   signer paths or assume distant new change is visible in a normal refresh.
8. Existing upstream scaffolding remains outside the active app entry/native set.
   Its removal/extraction follows the verified baseline and device test, retaining
   original MIT/Apache/dependency notices. Do not relabel Apache-derived ECC as MIT.
9. Keep CI disabled as the owner configured. Do not edit pipeline files, query
   or alter CI settings, or request a pipeline. Local commands are manual checks.
10. A real public backend needs a chosen VPS/domain/trusted TLS, capacity/rate-limit
    and logging/privacy decisions. Physical tests require the user's devices.
    These dependencies are not fulfilled by the loopback laboratory.
