[Programcode]

Offline reading for the iOS app. Debates, Library sources, saved results, flowcharts, chats and judged Sparring rounds stay readable without a connection. AI and saving wait for the network, as the plan says.

How it works

• The cache. offline-cache.ts keeps a copy of each read in AsyncStorage under the signed-in user's id. These modules read through it: debates.ts, sources.ts, results.ts, flowcharts.ts, chats.ts and sparring.ts.
• Network first. A read asks the network first and saves the answer. When the read fails, or expo-network says the device is offline, the saved copy answers instead.
• Stale answers. A read only replaces a copy when it started after the read that wrote it, so a slow answer never overwrites a newer one. A generation guard drops work started for an earlier account.
• Account changes. Signing out or switching account wipes the copies through the same setSession path that clears session memory. A copy is checked against its owner again on every read.
• Size limits. Each kind keeps a fixed number of copies (for example 40 per result kind and 20 Library lists), dropping the oldest first. A copy over 512k characters isn't kept.
• Offline notice. It looks like OneOffNotice and says what's happening: "Offline: showing saved copies. AI and saving wait for the connection." The string is in en, es and hu.
• Writes. There's no offline write queue. AI and saving fail with their existing translated network errors.

Checks (local; GitHub Actions is out of minutes)

At this head, run under the shared memory lock:

• npm run verify: exit 0. Vitest 320/320 (including the new cache tests for account isolation and stale answers), collaboration 30/30, edge 333, build OK, smoke 30/30.
• npm -w @debatelab/mobile run typecheck and npx eslint apps/mobile: exit 0.
• Offline proof on staging (ONLY=offline, iPhone light en and dark hu, 15 screenshots):
  • Device offline: the Debate tab, Library and a source, compared with the online read.
  • Staging unreachable after a reload: the Debate tab, Library, a saved case and a flowchart.
  • Another account: copies planted for another user are removed and never shown.
  • Sign-out: 6 copies before it, 0 after.

Not provable here

• real airplane mode on a phone
• expo-network's native reachability (the web build uses navigator.onLine)
• AsyncStorage size limits on a device
• a real second account, since staging has one demo code; unit tests cover it

Codex review rounds follow as comments.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
debate-navigator	Ready	Preview	Oct 14, 2026 10:00pm UTC
debatelab-staging	Ready	Preview	Oct 14, 2026 10:00pm UTC

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

• Configuration used: Repository YAML (base), Organization UI (inherited)
• Review profile: CHILL
• Plan: Free
• Run ID: 383f5660-2b87-496f-bc11-c48f02e276aa

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Note

🎁 Summarized by CodeRabbit Free

Your organization is on the Free plan. CodeRabbit will generate a high-level summary and a walkthrough for each pull request. For a comprehensive line-by-line review, please upgrade your subscription to CodeRabbit Essentials by visiting https://app.coderabbit.ai/settings/billing.

_{Comment @coderabbitai help to get the list of available commands.}

[Programcode]

Codex review r1 at ad4ee4d (gpt-6.1-sol, high): 5 findings, all checked against the code and accepted; fixes in progress.

1. Medium: switching or creating a debate leaves the cached Current Debate stale, so an offline restart restores the old one.
2. Medium: saving over a map doesn't refresh its cached copy or supersede earlier reads.
3. Medium: deleting a source (and results, flowcharts, chats, Sparring rounds) leaves cached lists with the deleted item.
4. Medium: the fallback serves a cached copy for any error, including 401/permission refusals; it must only fall back on connectivity failures.
5. Low: any successful read clears the offline notice while another screen still shows a cached copy.
   Codex found no cross-account leak: first-call ownership, same-account restart, different-account wipe, sign-out and read ordering all held.

[Programcode]

Codex review r2 at a587dae (gpt-6.1-sol, high): r1's five findings verified fixed (switch/create updates, map-save sequencing, HTTP-error rejection, owner isolation). Three new findings, accepted; fix in progress:

1. Medium (results.ts:283): a delete that returns no row (already deleted elsewhere) leaves cached lists and counts with the item. Same in sources, flowcharts, Sparring.
2. Medium (debates.ts:101): in parallel reads, a connectivity failure in one query masks a 401/refusal in another, so the cached list is served instead of the auth error.
3. Low (offline-cache.ts:197): a closed screen's key stays in the 'shown' set, so the offline notice can persist after reconnecting.

[Programcode]

Round-2 fixes: f7af0a0 (a delete that returns no row drops every list it may touch), 55433c9 (a server refusal wins over a missing answer in parallel reads, incl. loadChat), 7a45e57 (a saved copy stops counting once its screen closes or moves on). Tests for findings 1 and 2 fail without the fixes. Merged feat/native-ios (Sparring, dev-build prep, entitlements). Local at a846afc: npm run verify green (335 vitest, 30 smoke), mobile typecheck, eslint, knip pass. Codex r3 next.

[Programcode]

Codex review r3 at a846afc (gpt-6.1-sol, high): r2's three fixes hold (delete invalidation and refusal preference passed probes; both merges clean). Two findings:

1. Medium (offline-cache.ts:91): readAll held a known refusal until a stalled sibling query settled, leaving the screen loading. Fixed in 95313b1: a refusal rejects at once; only unanswered queries wait for the rest. New test fails without the fix.
2. Low (use-result-document.ts:68, flowchart/index.tsx:103): reopening a document held in session memory after a fallback read doesn't restore the 'saved copies' notice. Dismissed: that path shows the app's own in-memory copy of a document opened this session, not a fallback read, and when the device is offline the notice shows from the network state alone; the only gap is the server-unreachable-while-device-online case on a reopen, which is informational.
   Local at 95313b1: npm run verify green (336 vitest, 30 smoke), mobile typecheck, eslint, knip pass. Codex r4 next.

[Programcode]

Codex review r4 at 95313b1 (gpt-6.1-sol, high): no findings. readAll ordering, empty input, multiple refusals, mixed failures, later rejections and typing checked; no caller depends on waiting after a refusal. Merging into feat/native-ios. Not provable here: real airplane mode on a phone, expo-network's native reachability and AsyncStorage's native limits.