Skip to content

Protected backup still fails at local capture after 2.6.5 update #7066

Open

Description

@antuz123

Owner post-release evidence and priority — 31 October 2026

The affected iOS installation is 2.6.5 (621). The owner supplied diagnostics from both actions:

Action Recorded diagnostic HTTP status Elapsed
Check Again check / local-capture / non-transport — 318 ms
Back Up Now backup / local-capture / non-transport — 311 ms

The second screenshot is the requested Back Up Now result, captured before another Check Again could replace its details. The original incident remains unresolved; a successful protected backup/readback has not been established. Both failures are recorded during local capture, before the protected server snapshot request. This is not evidence of an HTTP/server failure, a missing backup, or loss of recovery data. The exact exception and affected local record remain unconfirmed.

Current priority: investigate the backup failure and the still-failing hide-wallet flow #7075 ahead of Version History. Inspect their shared local recovery capture/guard path as a possible link; a common cause is not yet established. Use focused source/fixture diagnosis to identify the local blocker and retain safe diagnostics. Further repeated phone checks are not requested at this checkpoint.

Version History is explicitly deferred to Later, unscheduled. Its installed-version header correctly shows 2.6.5 (621), while the newest displayed upgrade row is 2.6.3 (619); 2.6.4 and 2.6.5 rows are absent. Retain that observation under this issue without expanding the immediate backup investigation into a separate history fix.

Keep this issue open until the affected behavior is corrected and verified. Hide-wallet retains its existing R6 / 2.6.6 placement; this update does not assign the backup fix to a new release or start implementation.


Earlier checkpoints and evidence retained

Current checkpoint — 22 October 2026

Separate follow-up #7067 at c1c4b803039b61b940373c4dc9d08473cf25e042 implements the focused backup/history/UI correction and is included in combined R5 #7068. Keep this issue OPEN: passing source regressions do not establish the original phone's cause or corrected behavior. Exact native/disposable-upgrade and non-destructive original-phone acceptance remain pending. Preserve the owner's reported installed iPhone2.6.4; no repeated phone version check is requested. Do not duplicate implementation, interrupt running R5 jobs or reopen owner-deferred R4 Delete/recovery/C8/navigation campaigns. Earlier environment-blocked kickoff comments are historical; the implementation PR exists. The original scheduling and acceptance below remain in force.

Problem and goal

An existing iOS installation still cannot verify its protected server backup after the owner reports updating to 2.6.4. Attempting Delete protected backup also returns an unconfirmed-deletion error. The affected screens have visible action/text layout defects.

Resolve the actual failure and provide an understandable path to a current, verified protected backup. The affected phone's root cause is not yet established. Neither a green toggle nor a successful synthetic backend test establishes that this account is backed up.

Scheduling — separate follow-up after current R5 implementation

Owner clarification, 22 October 2026: keep the R5 work already underway in #7058 unchanged. This supersedes the earlier immediate R5 scope addition.

Queue this issue as the first focused follow-up patch after the current R5 implementation reaches its checkpoint, targeting inclusion before 2.6.5 is publicly released. Use a separate branch/PR from the verified R5 baseline available at that time. Do not add this implementation to the running #7058 scope or interrupt, cancel or duplicate active work. Check live GitHub state before starting; preserve any relevant work already completed elsewhere.

Sequence: finish current R5 implementation → diagnose/fix #7066 in a separate patch → perform the targeted native and affected-phone acceptance below → owner reviews the combined release candidate. The issue remains open until its actual failure and phone confirmation are resolved. This scheduling update launches no Codex session and grants no merge, deployment or publication approval.

Patch estimate: provisional 5k–10k changed code/test tokens. Keeping it separate bounds the implementation task; it does not erase the earlier overall 18k–28k estimate for R5 plus this correction. Re-estimate after diagnosis and suggest a smaller split if the repair grows substantially. Keep all-copy legacy purge and unrelated parked investigations outside this patch.

Observed evidence

Owner-supplied iPhone screenshots show:

  1. The More screen has Protected server backup enabled.
  2. Delete confirmation is available. The secondary action Turn off automatic backup is clipped off the left; only the ending is readable.
  3. The delete attempt reports: Could not confirm protected-backup deletion. Your local wallets and Recovery Key are unchanged. Check the backup status before trying again. The switch remains on.
  4. Protected-backup status reports Protected backup not verified, followed by a generic list of possible authorization, connection, size or changed-data causes.
  5. The backup row description is truncated, the lengthy toast crowds/overlaps bottom navigation, and the retained-copy notice has insufficient space above Check Again.

These are screenshot observations and source inspection, not a native reproduction in this investigation. Version 2.6.4 is owner-reported; the installed build number and the failing request's safe diagnostic have not yet been captured.

Established scope and current baseline

Codex Cloud task

Required environment — owner clarification: use the existing Codex Cloud environment named Bitcoin Keeper. Do not implement this task in base. Verify and report the actual selected environment before changing code. If the GitHub kickoff selected base or the environment cannot be verified, stop before implementation and report the task link/environment so it can be started in Bitcoin Keeper. Preserve the separate-patch scope and existing R5 work.

Start this queued patch after the current R5 implementation checkpoint. Read current AGENTS.md and current PRs/comments first. Use one separate task branch and a focused PR based on the verified R5 source, preserving ongoing work. Prefer sprint as PR base per AGENTS.md, but establish the actual R4/R5 source ancestry before branching so completed backup and R5 work is neither lost nor duplicated. Do not add this patch to the already-running R5 implementation task.

Provisional changed code/test estimate: approximately 5k–10k tokens for one backup defect investigation/correction and the adjacent screen fixes. Cause is unknown, so revise after diagnosis. Warn and propose a smaller split if the remaining delta approaches 20k tokens or more than two substantial independent features; do not turn this into a backup redesign.

1. Establish and diagnose the failure

  • Bind the affected app version/build to source and check compatibility with the deployed backend protocol. Use existing records first; request only the minimum missing device evidence.
  • Trace current local capture, authority derivation, snapshot request/validation/decryption, comparison, enrollment/repair and protected deletion paths.
  • Prefer the existing privacy-safe diagnostic in src/services/backup/repair.ts (fixed stage/category). If the released app cannot expose the needed result, add the smallest appropriate user-initiated diagnostic detail needed to distinguish the failure; no broad logging or telemetry project.
  • Capture only bounded operation/stage, safe category/error enum, HTTP status where available, app build and relevant timing. Never collect or publish Recovery Key words, private keys, tokens, stable account/namespace identifiers, wallet contents, ciphertext, raw bodies or headers.
  • Reproduce using disposable synthetic data following an existing-account upgrade path, including relevant older backup and saved consent state. A passing fresh account is a control, not this reproduction.
  • If actual-phone evidence remains necessary, document the exact non-destructive collection step and continue independent UI/source work. Do not guess a cause or report the phone fixed.

2. Apply a narrow correction

  • Fix the established failure and preserve a failing-before/passing-after regression.
  • Keep Check Again read-only. It currently checks freshness, rather than repairing, in the screenshot's unverified state. Provide a valid, understandable route to completing backup using established authorization/consent and conflict-preserving repair rules.
  • Preserve current-account ownership, complete readback verification, prior recovery records and honest uncertain/error results. Do not turn authentication/read errors into absent/deleted backups or force overwrite to make a check pass.
  • Distinguish automatic backup being enabled from the last verified backup state.
  • Correct the affected delete dialog, explanatory copy, toast placement/readability and status-screen spacing/action guidance. Both stop-automatic-backup and delete-protected-copy choices must be fully readable. Keep changes local where possible; avoid a global component redesign.

Acceptance and completion

On disposable upgraded accounts using the actual native app and compatible backend:

  • Reproduce the relevant failing path, then show protected backup upload/update, authenticated complete readback and matching recovery data.
  • Reopen/cold launch and recheck the same account; make an ordinary relevant metadata change and verify it reaches the protected copy.
  • Recover the synthetic backup into a separate disposable installation and verify complete expected recovery contents; preserve restored consent semantics.
  • Separately verify turn off automatic backup retains copies, while delete protected backup confirms the intended server deletion, turns automatic backup off and preserves local data. Verify the deleted state through a fresh read and explicit recreation without resurrecting deleted protected data.
  • Verify the failure case relevant to the correction: interrupted/uncertain requests never claim Verified/Deleted, and account/current-data binding remains intact. Do not reopen unrelated owner-parked campaigns.
  • Validate the affected screens on iOS and Android, including small width and larger text. Labels remain readable/tappable and messages do not obscure navigation.

Follow AGENTS.md: include an exact-head Codex local handoff, actual Cloud outcomes and targeted native steps. Required local checks: IOS, ANDROID, HARDWARE (the affected physical-phone confirmation). Cloud tests do not imply native or phone acceptance.

The affected owner's phone receives only a non-destructive final normal backup/verification check, followed by reopen/recheck, once the corrected approved build/backend is available. Do not reset, uninstall, restore over, delete backups from or export private recovery data from the working phone as a diagnostic.

Report separately:

  1. Root cause demonstrated and correction verified on disposable upgrade fixtures.
  2. Native acceptance for the exact corrected app/backend pair.
  3. Owner confirms the original phone now completes and retains successful backup verification.

Do not close this report merely because CI or synthetic service checks pass. Any release/build distribution, backend production change, merge or publication remains subject to the owner's existing approval boundaries. This issue is a concrete engineering handoff; it is not evidence that a Codex session has been launched.

Activity

  1. changed the title [-]2.6.4 iOS: protected backup remains unverified, deletion unconfirmed, and backup UI clips actions[/-] [+]R5: fix 2.6.4 iOS protected-backup verification, deletion and affected UI[/+] on Oct 22, 2026
  2. changed the title [-]R5: fix 2.6.4 iOS protected-backup verification, deletion and affected UI[/-] [+]Queued follow-up before 2.6.5: fix protected-backup verification, deletion and affected UI[/+] on Oct 22, 2026
  3. antuz123 commented on Oct 22, 2026

    @antuz123
    CollaboratorAuthor

    @codex Please start the implementation of this issue as the first separate follow-up patch after the recorded R5 implementation checkpoint, targeting inclusion before 2.6.5 is publicly released.

    Required Cloud environment: Bitcoin Keeper

    Owner clarification: run this task only in the existing Codex Cloud environment named Bitcoin Keeper. The base environment is not authorized for this work. Verify the actual selected environment before implementation and report its name with the task link. If the GitHub trigger selected base, or the environment cannot be verified, stop before code changes and report that blocker so the task can be started in Bitcoin Keeper. Do not create a substitute environment or alter environment settings.

    The owner requested this kickoff on 22 October 2026. Keep the existing R5 work and scope in #7058 intact. Its current published handoff is NEEDS_LOCAL at 77b72da29ac5f2cdf3f51c9a0e9f3fd1c2af7238 on codex/r5-key-deletion-startup; native acceptance is still pending. This is an implementation checkpoint, not release approval.

    Start and isolate the patch

    1. Read current AGENTS.md, this issue in full, and the latest R5: safe key deletion and startup fixes on submitted 2.6.4 source (draft) #7058 description/comments. Check live branches/PRs for another implementation of Protected backup still fails at local capture after 2.6.5 update #7066 before creating work; preserve any existing relevant changes.
    2. Use a new isolated task branch from the verified R5 checkpoint. Recheck the baseline SHA and source ancestry before branching; record the chosen SHA. Do not push this patch to the R5: safe key deletion and startup fixes on submitted 2.6.4 source (draft) #7058 branch, expand its scope, cancel its ongoing work, or duplicate another active patch.
    3. Open a separate focused draft PR. Follow AGENTS.md for the PR base and provide a patch-only comparison against the recorded R5 baseline so inherited R4/R5 changes are not presented as this patch.

    Implementation and evidence

    Follow this issue's diagnosis, narrow correction and acceptance sections. Establish the existing-account upgrade failure after 2.6.4, preserve a failing-before/passing-after regression, and fix the affected backup verification/deletion path and adjacent UI defects.

    Use the latest backend evidence: production Relay checks have passed for a fresh synthetic namespace, which does not establish the affected upgraded phone's behavior. Do not assume the backend is undeployed or infer that the phone is fixed from fresh-account tests.

    Preserve complete authenticated readback, account/session binding, prior recovery records, consent rules, read-only Check Again behavior and honest uncertain/error results. Keep stop-automatic-backup distinct from delete-protected-copy. Limit UI changes to the affected dialog labels, explanatory copy, toast placement/readability and status spacing/action guidance.

    If phone evidence is essential, specify the minimum non-destructive diagnostic step and continue independent work. Never reset, uninstall, restore over or delete backups from the owner's working phone, and never request or publish recovery words, keys, tokens, stable account identifiers, wallet contents or raw request/response data. Use disposable synthetic accounts for deletion and recovery testing.

    Provisional patch estimate is 5k–10k changed code/test tokens. Re-estimate after diagnosis and propose a smaller split if the delta approaches 20k or exceeds two substantial independent features. Keep legacy all-copy purge, broad UI redesign and parked #7056/#7057 outside this patch.

    Deliver

    Produce the draft PR with the demonstrated cause, exact baseline/head, actual Cloud check results, remaining gaps, and the AGENTS.md exact-head Codex local handoff for IOS, ANDROID and HARDWARE. Keep implementation verification, native acceptance and confirmation on the original phone as separate claims; this issue remains open until its acceptance is actually complete.

    This kickoff authorizes investigation, implementation, tests and a draft PR. It does not authorize merging, signing, distributing a release, production deployment or publication.

  4. chatgpt-codex-connector commented on Oct 22, 2026

    @chatgpt-codex-connector

    Status: Blocked before implementation

    I could not safely start this follow-up patch because the required R5 checkpoint is absent from the checkout and GitHub access is unavailable:

    • The checkout is currently at def0d6a90c80287edb4f58ae272cbfc340f00d73, not the specified R5 checkpoint 77b72da29ac5f2cdf3f51c9a0e9f3fd1c2af7238.
    • The required checkpoint is not present in the local Git object database.
    • Fetching sprint, codex/r5-key-deletion-startup, and the checkpoint SHA failed because the environment’s network proxy returned HTTP 403.
    • GitHub CLI is unauthenticated, so I could not inspect the live issue, latest R5: safe key deletion and startup fixes on submitted 2.6.4 source (draft) #7058 comments, or competing Protected backup still fails at local capture after 2.6.5 update #7066 branches/PRs.
    • This stale checkout predates the protected-backup implementation: src/services/backup/repair.ts and even src/services/backup/ are absent. Implementing here would therefore risk replacing, duplicating, or silently omitting the R4/R5 work the request explicitly says to preserve.
    • The repository requires work to use a task-specific branch, preserve existing work, and use sprint as the PR base.
    • It also requires the handoff to identify the exact final head SHA and truthfully report checks and remaining risks.

    I made no tracked changes, created no commit, and did not call make_pr; creating a PR from this unverifiable baseline would have violated the requested isolation and ancestry requirements. The pre-existing untracked .yarnrc.yml was left untouched.

    Checks

    • ✅ cat AGENTS.md
    • ✅ git status --short --branch
    • ✅ git rev-parse HEAD — returned def0d6a90c80287edb4f58ae272cbfc340f00d73.
    • ❌ git cat-file -e 77b72da29ac5f2cdf3f51c9a0e9f3fd1c2af7238^{commit} — required R5 checkpoint is not available locally.
    • ⚠️ git fetch --no-tags origin sprint codex/r5-key-deletion-startup — blocked by the environment proxy (CONNECT tunnel failed, response 403).
    • ⚠️ git fetch --no-tags origin 77b72da29ac5f2cdf3f51c9a0e9f3fd1c2af7238 — blocked by the same network limitation.
    • ⚠️ gh auth status — no authenticated GitHub host was configured.
    • ⚠️ gh pr view 7058 --json number,title,state,isDraft,headRefName,headRefOid,baseRefName,body,comments,reviews,url — unavailable without GitHub authentication.
    • ⚠️ gh issue view 7066 --json number,title,state,body,comments,url — unavailable without GitHub authentication.
    • ✅ test -d src/services/backup — correctly failed, confirming the requested backup implementation is absent from this stale source tree.

    To resume safely, the workspace needs the repository objects through exact commit 77b72da29ac5f2cdf3f51c9a0e9f3fd1c2af7238 and authenticated/readable GitHub state for #7058, #7066, and live branches/PRs.

    View task →

  5. changed the title [-]Queued follow-up before 2.6.5: fix protected-backup verification, deletion and affected UI[/-] [+]Protected backup still fails at local capture after 2.6.5 update[/+] on Oct 31, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions