data/opus-10.jsonl

{"prompt": "please put together the design for our model serving layer. we have 6 models, two need gpu, and the traffic is spiky — 10x for 20 minutes twice a day. i want to see the options for autoscaling and what the cold start cost is for each", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "triton inference server config for the two gpu models, dynamic batching with a 20ms queue delay, and a health endpoint the mesh understands", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "model registry ui: versions per model, which is serving in which env, and a promote action with a confirm that shows the eval delta", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "the inference timeout is 2s and the p99 is 2.4s, raise it to 5", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.1, "slice": "core", "lang": "en"}
{"prompt": "our preprocessing runs in the request handler for one model and in a sidecar for another. pick the sidecar approach for both, outputs must be numerically identical", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "gpu memory fragments over a day of serving and eventually a request OOMs even though total usage is under 40%", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "could you explain how the batching interacts with our per-request timeout? i want to know whether a slow batch can blow the deadline for everyone in it", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "write the api docs for the inference endpoint — input schema per model, the batching behavior clients should expect, and the error codes", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "boundary", "lang": "en"}
{"prompt": "kotlin: paging 3 for the transactions list, with a room-backed remote mediator so it works offline", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "the transaction row layout: amount right-aligned and monospaced, merchant name truncating with the category chip never squeezed out", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "core", "lang": "en"}
{"prompt": "negative amounts show a minus AND red parentheses, pick one", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.2, "slice": "boundary", "lang": "en"}
{"prompt": "would you mind planning the modularization of the android app? one big :app module today, 90k lines, and build times are 6 minutes incremental", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "our fragments and composables both exist for the same three screens because of a half-finished migration. finish it — delete the fragments, no behavior change", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "app takes 4.5s to first frame on a mid-range device, was 1.8s three releases ago. no obvious single culprit in the startup trace", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "is the datastore migration from sharedprefs actually idempotent if the app is killed mid-migration? tell me what the code does", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "boundary", "lang": "en"}
{"prompt": "contribution guide for the android repo — module boundaries, the di conventions, and how to add a new screen", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "core", "lang": "en"}
{"prompt": "nix flake for the dev environment so people stop having version drift, with a devshell that includes the toolchains for all three services", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "our CI takes 34 minutes. i want a plan to get it under 10 — caching, splitting, what we move to a nightly job. include what we give up", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "the docker build ignores .dockerignore because it's in the wrong directory", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.2, "slice": "core", "lang": "en"}
{"prompt": "consolidate our four dockerfiles into a multi-stage one with targets, same final images by digest-equivalent content", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "builds are cache-missing on every run even though nothing in the lockfile changed. buildkit with a registry cache", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "take a look at our github actions permissions — i think several jobs have write access they don't need", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "document the release process: tags, what CI does automatically, the manual steps, and how to cut a hotfix", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "core", "lang": "en"}
{"prompt": "build status widget for our internal dashboard, per-branch, with the duration trend as a sparkline", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "just keep going", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "vague-eval", "lang": "en"}
{"prompt": "would you take a look at this and tell me what's happening? i've been at it since this morning:\n\n$ ./gradlew :app:assembleRelease\n\n> Task :app:minifyReleaseWithR8 FAILED\nAGPBI: {\"kind\":\"error\",\"text\":\"Missing class com.acme.sync.proto.SyncEnvelope$Builder (referenced from: void com.acme.sync.SyncClient.push(java.util.List))\",\"sources\":[{}],\"tool\":\"R8\"}\nAGPBI: {\"kind\":\"error\",\"text\":\"Missing class io.grpc.okhttp.OkHttpChannelBuilder (referenced from: io.grpc.ManagedChannelProvider)\",\"sources\":[{}],\"tool\":\"R8\"}\n\nFAILURE: Build failed with an exception.\n* What went wrong:\nExecution failed for task ':app:minifyReleaseWithR8'.\n> A failure occurred while executing com.android.build.gradle.internal.tasks.R8Task$R8Runnable\n   > Compilation failed to complete, position: null, origin: null\n\ndebug builds are fine. we added the grpc dependency last week", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "pasted-context", "lang": "en"}
{"prompt": "grpc client for the sync service in the android app, with a channel that survives network changes and doesn't leak on config change", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "sync status indicator in the toolbar: idle, syncing with a spinner, error with a tap-to-retry, and 'last synced 4m ago' text", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "the sync interval is 15 minutes minimum per workmanager but we set 5, so it silently becomes 15. say so in the settings copy", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "boundary", "lang": "en"}
{"prompt": "plan the conflict resolution ui for sync conflicts, because right now last-write-wins silently and users lose edits", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "our repositories all implement the same cache-then-network pattern with hand-rolled variations. one pattern, and the observable emission order must stay the same", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "pull-to-refresh spins forever if the request fails without an exception, like on a captive portal", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "explain the ordering guarantees of our outbox in the android client, specifically across app restarts", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "write the protocol doc for the sync api — envelope format, the cursor semantics, and how a client recovers from a rejected batch", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "boundary", "lang": "en"}
{"prompt": "sketch the strategy for supporting tablets and foldables properly, not just stretched phone layouts", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "two-pane layout on wide screens with the list on the left, and the selection should survive a fold/unfold", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "landscape on phones shows a stretched empty state illustration at 4x", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.2, "slice": "core", "lang": "en"}
{"prompt": "our dimens are hardcoded dp in composables. move to a spacing scale in the theme, identical rendering at 1x", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "boundary", "lang": "en"}
{"prompt": "on a fold, opening the detail pane then folding loses the item and shows the empty state", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "check the app against talkback on the three main screens and tell me what's broken, don't fix anything yet", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "accessibility statement for the play store listing and the website, honest about what's not there yet", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "core", "lang": "en"}
{"prompt": "in-app update prompt using the play core api, flexible for minor and immediate for a security release", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "design the offline map tile cache — size cap, eviction, and letting the user pin a region deliberately", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "downloaded regions list with sizes, a delete swipe, and a total storage used bar at the top", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "tile cache dir isn't in the app's cache directory so android never clears it", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "core", "lang": "en"}
{"prompt": "figure out the plan for the maps sdk migration and then port the tile layer as a first step", "purpose": "planning", "secondary": "frontendImpl", "mixed": true, "difficulty": 0.7, "slice": "mixed", "lang": "en"}
{"prompt": "diagnose the tile download stalls then note the workaround in the support doc", "purpose": "debugging", "secondary": "writing", "mixed": true, "difficulty": 0.7, "slice": "mixed", "lang": "en"}
{"prompt": "let's do the next one", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "vague-eval", "lang": "en"}
{"prompt": "i'd like the design written down for how we shard the write path. our single postgres primary is at 70% cpu at peak and we've exhausted the easy wins. options as i see them: split by tenant to separate clusters, citus, or move the two hottest tables out. compare and recommend, with the failure modes of each", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "tenant-to-shard routing layer with a lookup table, connection pools per shard, and a way to move a tenant with a short write pause", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "shard admin view: tenants per shard, size, write rate, and a move action that shows the estimated pause", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "the shard id is an int in the code and a string in the config", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.2, "slice": "core", "lang": "en"}
{"prompt": "every query in the repo assumes one connection. thread the shard context through properly — no global state, and every existing query hits the same data it does now", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "one tenant's writes land on the wrong shard occasionally and we end up with rows in two places", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "read the routing code and tell me what happens to an in-flight request during a tenant move", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "the sharding doc for the team — how routing works, what's shard-local vs global, and the rules for writing a new query", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "boundary", "lang": "en"}
{"prompt": "please add read-replica routing for the reporting queries, with a staleness guard so we don't serve a write-then-read from a lagging replica", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "reporting page shows a spinner then old data then new data, three states in two seconds. settle it down", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "the replica lag threshold is 30s, way too lax, make it 2", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.2, "slice": "core", "lang": "en"}
{"prompt": "outline how we'd test shard failover without a staging environment that has multiple shards", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "our connection handling opens a transaction for read-only queries in half the endpoints. clean that up, same results, fewer write locks", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "does the replica routing respect an explicit transaction? i don't want a read inside a write tx going to a replica", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "boundary", "lang": "en"}
{"prompt": "after a failover last night some rows are missing from one table and present in the audit log. i need to understand what we lost and how", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "write the failover runbook, and be explicit about the steps that are NOT automated", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "お願いします。バッチ推論のパイプラインを設計してください。1日3億レコード、GPU は 8 枚、失敗したシャードだけ再実行できるように", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "ja"}
{"prompt": "feature drift monitoring — compute the training/serving distribution distance per feature daily and alert on the ones that move", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "drift dashboard: one small multiple per feature, sorted by how much it moved, click through to the distributions", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "the drift threshold is hardcoded 0.1 for every feature, make it configurable per feature", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "core", "lang": "en"}
{"prompt": "our three notebooks that compute drift each define the metric slightly differently. one implementation in the library, all three call it", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "drift alerts fire for a feature that is provably constant. i've checked the input data, it's the same value every row", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "does our monitoring compare against the training distribution or the previous day? tell me which, because the alerts only make sense for one of those", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "boundary", "lang": "en"}
{"prompt": "docs for the monitoring service's api — the metrics endpoints, the query params, and what the response fields mean", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "boundary", "lang": "en"}
{"prompt": "cool cool", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "vague-eval", "lang": "en"}
{"prompt": "here's the TODO list i left myself, please turn it into a proper sequenced plan with what depends on what — i suspect it's not doable in this order:\n\n- switch the frontend to the new auth endpoints\n- delete the old session table\n- move refresh token storage to httpOnly cookies\n- add device list + revoke-per-device UI\n- backfill device records for existing sessions (we don't record device today)\n- rotate the signing key\n- turn off the legacy /login endpoint (the ios app 3.x still uses it, ~6% of users)\n- add rate limiting to /login (currently none)", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "pasted-context", "lang": "en"}
{"prompt": "device list endpoint with last-seen, ip and user agent, plus a revoke-one and revoke-all-others", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "active sessions screen — device icon by platform, current device labeled, revoke button per row with a confirm", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "user agent parsing puts 'Other' for every ios app request", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "core", "lang": "en"}
{"prompt": "rate limit /login per account and per ip, with a lockout that doesn't let an attacker lock out a victim", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "our token verification is implemented twice, once in the middleware and once in the websocket handler, with different clock skew tolerance. one implementation", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "logging out on one device logs you out everywhere, which is not what we want and not what i thought the code did", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "boundary", "lang": "en"}
{"prompt": "help center article about active sessions and what revoking does, plus what it doesn't do", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "core", "lang": "en"}
{"prompt": "map out the plan for our mobile app's token storage. keychain/keystore, refresh rotation, and what happens on a device restore from backup", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "biometric gate on app open, with a passcode fallback and a setting to require it every time vs after 5 minutes", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "the biometric prompt title says 'Authenticate' which is meaningless, write proper copy for it", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.2, "slice": "boundary", "lang": "en"}
{"prompt": "lock screen overlay that blurs the content behind it, and it must appear before the app snapshot is taken", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "after a phone restore users get logged out but the app shows their cached data for a second first", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "review our token refresh for the case where two requests both get a 401 and both try to refresh", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "the keychain accessibility is kSecAttrAccessibleAlways, should be AfterFirstUnlockThisDeviceOnly", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "core", "lang": "en"}
{"prompt": "propose the approach for supporting multiple accounts in the mobile apps, including how we keep their caches and push tokens separate", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "background upload queue that survives app kill, resumes on next launch, and reports progress to a notification", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "upload queue screen with per-item progress, retry, cancel, and a clear-completed action", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "progress notification doesn't clear when the queue finishes", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.2, "slice": "core", "lang": "en"}
{"prompt": "the upload code exists in the camera flow and the document flow separately. unify, same retry behavior and the same notification updates", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "uploads restart from zero after a network blip instead of resuming from the last part", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "walk me through what our uploader does when the file is deleted from the device mid-upload", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "internal doc on the upload protocol, part sizes, the resume handshake, and the server-side assembly step", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "boundary", "lang": "en"}
{"prompt": "podemos continuar", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "vague-eval", "lang": "pt"}
{"prompt": "propose the architecture for our new realtime collaboration service. requirements: 50 concurrent editors per doc, sub-100ms local echo, server authoritative ordering, and the doc has to be reconstructible from the log for audit. i'd like the tradeoffs of crdt vs ot spelled out for OUR constraints, not in general", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "operation log storage with per-doc ordering, compaction into snapshots every 500 ops, and a replay endpoint", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "presence cursors with names, colors assigned stably per user, and they should fade after 5s of inactivity", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "cursor color palette has two nearly identical greens", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.1, "slice": "boundary", "lang": "en"}
{"prompt": "the editor has its own document model separate from the sync layer's, and they're kept in step by hand. make one the source of truth", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "two people typing in the same paragraph occasionally produces interleaved garbage, roughly once an hour of active use", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "explain the ordering our server assigns to concurrent ops, and whether a client can observe two different orders across a reconnect", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "write up the sync protocol for the team, including the reconnect handshake and what the client must do with ops it already applied", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "boundary", "lang": "en"}
{"prompt": "comments anchored to text ranges that survive edits around them, including edits that delete the anchored text entirely", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "comment sidebar aligned to its anchor's vertical position, stacking when two are close, and highlighting the range on hover", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "resolved comments still show in the count", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.2, "slice": "core", "lang": "en"}
{"prompt": "plan how we handle a doc that grows to 50MB of ops. compaction, cold storage, and what the ui does while a big doc loads", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "comments on deleted text end up anchored to position 0 and cluster at the top of the doc", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "look at the presence implementation and tell me how many messages per second 50 editors generate", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "the version history uses timestamps for ordering and the ops log uses a lamport counter. make them consistent, history must show the same order the ops applied in", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "changelog and a short blog post about shipping collaborative editing, technical audience", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "core", "lang": "en"}
{"prompt": "design the offline editing story for the collaborative doc, then implement the local op queue", "purpose": "planning", "secondary": "backendImpl", "mixed": true, "difficulty": 0.9, "slice": "mixed", "lang": "en"}
{"prompt": "explain how anchoring works and then document it in the architecture folder", "purpose": "review", "secondary": "writing", "mixed": true, "difficulty": 0.7, "slice": "mixed", "lang": "en"}
{"prompt": "ok", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "vague-eval", "lang": "en"}
{"prompt": "need a plan for our search infrastructure. postgres full text is at its limit — 40M docs, faceting is slow, and no typo tolerance. opensearch, typesense, or a managed thing. include the reindexing story and how we keep it in sync with writes", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "indexing pipeline that tails the change log and updates the search index within a second, with a full reindex mode", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "search results page with facets in a left rail, counts per facet, and applied filters as removable chips at the top", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "the search input doesn't submit on enter in safari", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "core", "lang": "en"}
{"prompt": "query building for search is duplicated between the api and the admin tool with different escaping. one query builder, and check the escaping is right in both cases", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "documents updated in the last few minutes sometimes don't appear in search and sometimes do. the indexer log shows them all as indexed", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "is our relevance tuning actually applied at query time or index time? and which fields are boosted", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "search api docs — the query syntax we support, facet params, pagination limits, and the relevance caveats", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "boundary", "lang": "en"}
{"prompt": "typeahead endpoint with prefix matching, 50ms budget, and per-tenant filtering that can't be bypassed", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "typeahead dropdown with keyboard nav, grouped results by type, and a 'see all results' footer row", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "min query length for typeahead is 1, make it 2", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.1, "slice": "core", "lang": "en"}
{"prompt": "plan how we'd measure search quality — click-through, zero-result rate, and an offline judgement set. i want to know if a relevance change helped", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "search analytics logging: query, result count, position clicked, latency, with the query text hashed for anything that looks like an email", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "zero results page: suggest spelling corrections, drop the least selective filter, and offer a saved-search signup", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "queries with a trailing space return nothing", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.2, "slice": "core", "lang": "en"}
{"prompt": "check the analytics logging for anything that would record pii, we log the raw query string in one path i think", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "we compute 'zero result rate' in the dashboard sql and in the python job differently. one definition, and say which numbers change", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "search latency p99 tripled at 2pm yesterday and stayed there. index size is the same, query mix looks the same", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "write the internal doc on our relevance model and the levers we have, so the next person doesn't rediscover it", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "boundary", "lang": "en"}
{"prompt": "ugh, fine, plan it properly then start on the indexer", "purpose": "planning", "secondary": "backendImpl", "mixed": true, "difficulty": 0.8, "slice": "mixed", "lang": "en"}
{"prompt": "review the reindex script and fix the obvious bugs", "purpose": "review", "secondary": "quickFix", "mixed": true, "difficulty": 0.5, "slice": "mixed", "lang": "en"}
{"prompt": "one sec, actually the other file", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "vague-eval", "lang": "en"}
{"prompt": "give me the migration plan from our custom auth to an off-the-shelf idp. 400k users with bcrypt hashes, social logins, and enterprise sso for 30 orgs. what moves first and how we avoid a mass password reset", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "lazy password migration — verify against bcrypt, rehash into the new format on successful login, and track the percentage migrated", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "the login page needs a 'continue with' section that only shows the providers configured for that email's domain", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "bcrypt cost is 8, that's low. take it to 12 for new hashes", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.2, "slice": "core", "lang": "en"}
{"prompt": "our user lookup is by email in some places and by a normalized email in others, which is how we ended up with two accounts for one person. one normalization, applied everywhere", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "social login creates a duplicate account when the email matches an existing password account, but only for google not github", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "look at the account linking logic and tell me if an attacker could link their oauth identity to someone else's account", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "write the migration comms — the email to users if anything changes for them, and the internal faq for support", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "core", "lang": "en"}
{"prompt": "account merge flow: detect the duplicate, show what will be combined, and do it transactionally with an undo window", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "merge accounts screen showing both accounts side by side with what's kept from each", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "the 'account already exists' error suggests signing in, but with the wrong provider name", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.2, "slice": "core", "lang": "en"}
{"prompt": "figure out the plan for supporting org-enforced sso, where a member with a password account gets converted", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "users in an sso-enforced org can still log in with a password if they use the old bookmarked url", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "explain what our idp discovery does with an email whose domain is claimed by two orgs", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "admin docs for domain verification and sso enforcement, including the escape hatch for locked-out admins", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "consolidate the four places we build a login redirect url — they disagree on trailing slashes and one of them drops the `next` param", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "boundary", "lang": "en"}
{"prompt": "wrap up whatever's half done", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "vague-eval", "lang": "en"}