nucleic-purpose-classifier/data/round2-14.jsonl

{"prompt": "our checkpoint restart produces wrong results but only for jobs restarted after more than 48 hours:\n\ncheckpoint written at step 412000, size 8.1 TB\nrestart at step 412000, continues to step 500000\nvalidation against a continuous run: max relative difference 4.1e-3 (tolerance 1e-9)\n\nthe checkpoint stores the state arrays and the RNG seed but not the RNG stream position, and the thermostat's random forces resume from the seed rather than from where they were\n\nthe relevant part of the checkpoint code:\n\n  ! checkpoint.f90\n  write(unit) step, n_atoms\n  write(unit) positions, velocities\n  write(unit) rng_seed\n  ! rng_pos declared, never written  <- compiler warns about it, we've ignored that for years\n\n  ! thermostat.f90, on restart\n  call random_seed(put=rng_seed)     ! stream restarts from the beginning\n\nvalidation runs so far:\n  restart after 6h   max rel diff 2.1e-11   (looks fine)\n  restart after 24h  max rel diff 8.8e-6\n  restart after 48h  max rel diff 4.1e-3\n\nthe two papers under discussion both used runs restarted at least twice", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "pasted-context", "lang": "en"}
{"prompt": "the checkpoint writer and reader have drifted apart across three format versions:\n\n! checkpoint.f90 — writer\nwrite(unit) step, n_atoms, positions, velocities, seed\n\n! checkpoint.f90 — reader, with version sniffing\nread(unit) step, n_atoms\nif (n_atoms > 0 .and. n_atoms < HUGE_N) then\n   read(unit) positions, velocities, seed          ! v2\nelse\n   rewind(unit); read(unit) step, positions, velocities   ! v1, no seed\nend if\n\nversion detection by whether a field looks plausible, no magic number, no version field, and v3 adds the RNG position we need for correct restarts", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "pasted-context", "lang": "en"}
{"prompt": "HRIS special cases into configuration", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "a real checkpoint format with a version field", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "boundary", "lang": "en"}
{"prompt": "job adverts stopped syncing to one job board and their support says they see nothing:\n\nPOST https://partner.jobboard.example/v2/postings\n  → 200 OK  {\"accepted\": true, \"id\": \"jb_88412\"}\n\nour sync log:\n  posted 412 adverts, 412 accepted, 0 errors\n\ntheir side:\n  \"we have received 0 postings from your account since 21 July\"\n\nwe changed the partner API key on 21 July; the old key still returns 200 with an accepted body but their system discards the postings, which their docs describe in a footnote as \"deprecated key behaviour\"", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "pasted-context", "lang": "en"}
{"prompt": "our receipt rendering path, which is why the cheap tills fall over:\n\nfun renderToBitmap(receipt: Receipt): Bitmap {\n    val height = receipt.lines.size * LINE_HEIGHT_PX      // 203dpi, ~34px per line\n    val bmp = Bitmap.createBitmap(WIDTH_PX, height, Bitmap.Config.ARGB_8888)\n    val canvas = Canvas(bmp)\n    receipt.lines.forEachIndexed { i, line -> canvas.drawText(line, 0f, (i * LINE_HEIGHT_PX).toFloat(), paint) }\n    return bmp\n}\n\nARGB_8888 at 576 pixels wide, a 60-line receipt is about 4.7MB, a 500-line stocktake report is 40MB, and the printer takes a monochrome bitmap anyway", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "pasted-context", "lang": "en"}
{"prompt": "the HRIS field mapping lives in YAML with per-customer special cases in code:\n\n# mappings/acme.yaml\nfirst_name: givenName\nstart_date: startDate       # format: YYYY-MM-DD\n\n# mappings/globex.yaml\nfirst_name: fname\nstart_date: start           # format: DD/MM/YYYY  <- handled by an if in the pusher\n\n// pusher.ts\nif (customer === 'globex') payload.start = formatUk(offer.startDate)\nif (customer === 'initech') payload.employee_id = padLeft(offer.id, 8)\nif (customer === 'umbrella') delete payload.middle_name\n\nnine customers, four with code special cases, and adding a tenth means a deploy", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "pasted-context", "lang": "en"}
{"prompt": "a migration archived forty thousand applications at one customer because their talent pool records had never been touched since creation, and unarchiving them is a click each in the product. i want the recovery planned properly — how we identify exactly which records we archived, how we put them back without a script that could touch anyone else's data, and what we change so a migration can't do this again", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "core", "lang": "en"}
{"prompt": "HRIS integration guide was written for one customer and is now used for nine, four of whom have special cases living in if-statements. write the guide that reflects reality, and separately list every special case that should be configuration so we can stop deploying to onboard a customer keep it concrete — file names and line numbers are more use than principles here.", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "offline banner with today's total", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.35, "slice": "core", "lang": "en"}
{"prompt": "what makes an application \"active\"?", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "simulation is one four-thousand-line module and only one person can work on it, which is now the binding constraint on everything else in the group's list. before splitting it i'd like agreement on the seams — I/O, decomposition, physics, output — and on how we prove the split changed nothing, given our only test is a stored trajectory", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "boundary", "lang": "en"}
{"prompt": "job adverts stopped reaching a job board three weeks ago while their API kept returning 200 with an accepted body, which our sync treats as success. work out what we should have been checking, and whether any other partner integration has the same shape of silent failure nobody has trusted this code for about a year, which is part of the problem.", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "interview scheduling emails go out with the wrong timezone for about one candidate in twenty:\n\ninterview i_4471\n  scheduled_at (db, timestamptz): 2026-08-03 14:00:00+00\n  interviewer timezone: Europe/Berlin\n  candidate timezone: America/New_York (from their profile)\n  email to candidate said: 3 August, 14:00 (your local time)\n  actual time in New York: 10:00\n\nthe template formats with the interviewer's timezone but labels it as the candidate's, and only candidates whose profile timezone differs from the interviewer's notice", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "pasted-context", "lang": "en"}
{"prompt": "an MPI job hangs at the same point every time on more than 512 ranks:\n\n[rank 0] entering collective: MPI_Allreduce (comm=MPI_COMM_WORLD, count=4194304, MPI_DOUBLE)\n[rank 1] entering collective: MPI_Allreduce\n...\n[rank 511] entering collective: MPI_Allreduce\n[rank 512] entering collective: MPI_Bcast (comm=MPI_COMM_WORLD, count=1, MPI_INT)\n[rank 513] entering collective: MPI_Bcast\n...\n<no further output, job killed by the scheduler at the 24h wall clock>\n\nranks 512 and above take a different branch that broadcasts a convergence flag before the reduction, and the branch depends on a value computed from the local domain size", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "pasted-context", "lang": "en"}
{"prompt": "stock counts drift in stores that use the handheld scanners, about 0.3% of lines a week:\n\nscan event  sku=SKU-4471  qty=1  device=hh_881  at=11:02:14.114  synced_at=11:02:19\nscan event  sku=SKU-4471  qty=1  device=hh_881  at=11:02:14.118  synced_at=11:02:19\nstock movement  sku=SKU-4471  delta=-1  source=scan  at=11:02:19\nstock movement  sku=SKU-4471  delta=-1  source=scan  at=11:02:19\n\nthe scanner debounces a double trigger in firmware at 200ms, these are 4ms apart, and each scan gets a device-local id that the backend uses for idempotency — but the id is a counter that resets when the handheld is docked", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "pasted-context", "lang": "en"}
{"prompt": "three places decide whether an application is \"active\" and they disagree:\n\n// api/pipeline.ts\nconst active = (a: Application) => !['rejected', 'withdrawn', 'archived'].includes(a.stage)\n\n// jobs/nurture.ts\nconst active = (a: Application) => a.stage === 'pool' || a.stage === 'applied'\n\n-- reports/active_candidates.sql\nWHERE a.stage NOT IN ('rejected', 'withdrawn') AND a.archived_at IS NULL\n\nthe report includes archived-by-migration records that have archived_at set but stage still 'pool' for some rows, because the migration set both and an earlier one set only stage", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.65, "slice": "pasted-context", "lang": "en"}
{"prompt": "careers page says \"Sorry, no jobs founds\"", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.1, "slice": "core", "lang": "en"}
{"prompt": "payment buttons need to be 72dp", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.25, "slice": "core", "lang": "en"}
{"prompt": "pipeline filters should live in the URL", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.35, "slice": "core", "lang": "en"}
{"prompt": "is the scan idempotency key unique enough?", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "boundary", "lang": "en"}
{"prompt": "checkpoint restarts are silently wrong for long jobs because the RNG stream position isn't saved, and two published papers used runs that were restarted. i need the plan: how we determine which results are affected, what we tell the group and the journals if any are, and how the format changes so this class of error is impossible rather than merely fixed", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.95, "slice": "core", "lang": "en"}
{"prompt": "offline card payments are now contractual with limits, automatic retransmission and same-day manager reporting, none of which we have. work through what that means for the till, the backend and the store manager's tooling, and be explicit about which parts need the payment vendor's cooperation rather than just ours we've been burned by guessing at this before, so evidence over instinct please.", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "core", "lang": "en"}
{"prompt": "simulation module split needs the seams agreed before anyone moves a line, and then the I/O layer extracted first because it's the least entangled. do both, and keep the stored-trajectory test passing at every step", "purpose": "planning", "secondary": "refactor", "mixed": true, "difficulty": 0.85, "slice": "mixed", "lang": "en"}
{"prompt": "what does the till do with an offline voucher that is declined the next morning", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "por favor, revisa la consulta que alimenta el panel de reclutamiento antes de que la subamos:\n\nSELECT j.id, j.title,\n       count(a.*) FILTER (WHERE a.stage = 'applied') AS applied,\n       count(a.*) FILTER (WHERE a.stage = 'interview') AS interviewing,\n       (SELECT count(*) FROM interviews i WHERE i.job_id = j.id AND i.starts_at > now()) AS upcoming,\n       (SELECT avg(EXTRACT(epoch FROM (a2.hired_at - a2.created_at))/86400)\n          FROM applications a2 WHERE a2.job_id = j.id AND a2.hired_at IS NOT NULL) AS avg_days_to_hire\nFROM jobs j\nLEFT JOIN applications a ON a.job_id = j.id\nWHERE j.tenant_id = $1 AND j.status = 'open'\nGROUP BY j.id, j.title;\n\nel inquilino más grande tiene 4.100 vacantes abiertas y 8,2 millones de candidaturas", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "pasted-context", "lang": "es"}
{"prompt": "changelog for the POS release, which stores will read on their tills:\n\n41c9e0b fix(print): receipts render in bands rather than one bitmap, fixing crashes on 1GB tills\n88f21c0 fix(payments): terminal client retries once on a read timeout before falling back to offline\nc0aa774 feat(offline): offline vouchers now carry a reference the till can reconcile automatically\n2e91b45 fix(scan): scan idempotency now uses a device-persistent id rather than a resetting counter\naa30f19 feat(stock): stocktake reports print in landscape on the wide printers\n9c1d004 chore: minimum android is now 11\n4410bb7 fix(ui): the numeric keypad no longer accepts a leading zero on quantities\nb77e910 feat(returns): returns without a receipt require a manager PIN\n\nour readers are store managers, not engineers; two of these change what staff do at the till", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.55, "slice": "pasted-context", "lang": "en"}
{"prompt": "unsere Mandantenprüfung ist an vierzehn Stellen unterschiedlich implementiert:\n\n// api/jobs.ts\nwhere: { tenantId: session.tenantId }\n\n// api/applications.ts\nwhere: { job: { tenantId: session.tenantId } }\n\n// api/reports.ts\nconst rows = await prisma.$queryRaw`SELECT ... WHERE tenant_id = ${session.tenantId}`\n\n// jobs/nurture.ts\n// keine Prüfung – der Job läuft für alle Mandanten und filtert nach Stage\n\nvierzehn Stellen, drei Muster, und eine Stelle ohne Prüfung; ich hätte gern eine Durchsetzung, die man nicht vergessen kann", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "pasted-context", "lang": "de"}
{"prompt": "design spec for the recruiter pipeline board, which recruiters have open all day:\n\nPipeline board\n- Columns per stage, cards per candidate, virtualised — the largest tenant has 4,000 candidates in one job.\n- Card shows name, current stage age, source, and a flag if an interview is scheduled within 24 hours.\n- Drag between columns moves the stage optimistically, with a toast that can undo for ten seconds.\n- Bulk select with shift-click, then a bulk action bar; every bulk action states how many candidates it affects.\n- Filters: source, stage age, interviewer, tag. Filters live in the URL so recruiters can share a view.\n- Keyboard: arrow keys move between cards, enter opens, m moves stage via a menu — recruiters asked for this.\n- An archived candidate must be visibly distinct rather than absent, because \"where did they go\" is our top support question.\n- Everything must stay responsive with 4,000 cards and a websocket delivering updates from other recruiters.", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "pasted-context", "lang": "en"}
{"prompt": "terminal read timeout to 10s with one retry", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.35, "slice": "boundary", "lang": "en"}
{"prompt": "stage needs an indicator besides colour", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "core", "lang": "en"}
{"prompt": "apply modal traps focus with no escape", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.35, "slice": "boundary", "lang": "en"}
{"prompt": "one active-application predicate", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.55, "slice": "core", "lang": "en"}
{"prompt": "doc comments on the public jobs API", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "core", "lang": "en"}
{"prompt": "why does the 512-rank job hang?", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "boundary", "lang": "en"}
{"prompt": "applications lost at the morning spike", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "core", "lang": "en"}
{"prompt": "till screen", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "vague-eval", "lang": "en"}
{"prompt": "tenant scoping is done by hand in fourteen places with three patterns and one omission, and we have no test that would catch a fifteenth. i want a position on enforcement — row-level security, a repository layer, or something else — that survives our transaction-scoped pooler and the reports that use raw SQL", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "boundary", "lang": "en"}
{"prompt": "nobody can tell me whether a restarted run can differ from a continuous one in a way that matters, and two papers depend on the answer. read the checkpoint writer, the reader and the thermostat's use of the RNG, and tell me exactly what diverges and by how much on a realistic run whatever you find, write it somewhere the next person will actually look.", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "core", "lang": "en"}
{"prompt": "scan idempotency key is device id plus a device-local counter that resets when the handheld is docked, held in a one-hour cache, across four hundred devices. work out how often two different scans can collide on that key, and whether the stock drift we see is consistent with that rate i'm not attached to the current approach if there's an obviously better one.", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "boundary", "lang": "en"}
{"prompt": "before we touch the checkpoint format i want the failure modes written down — restart on a different rank count, a changed thermostat, a truncated file — and then the versioned header implemented against that analysis", "purpose": "planning", "secondary": "backendImpl", "mixed": true, "difficulty": 0.85, "slice": "mixed", "lang": "en"}
{"prompt": "i'd like an honest read of whether our OpenMP reduction can produce different totals for the same input on the same machine, and if it can, the change that makes it deterministic without costing more than a few percent", "purpose": "review", "secondary": "refactor", "mixed": true, "difficulty": 0.85, "slice": "mixed", "lang": "en"}
{"prompt": "recruiter app ignores the tokens and uses stage colour as the only stage indicator. bring it onto the tokens with a non-colour indicator, and tell me whether the card height changes enough to affect how many fit on a laptop screen", "purpose": "frontendImpl", "secondary": "review", "mixed": true, "difficulty": 0.55, "slice": "mixed", "lang": "en"}
{"prompt": "ATS background jobs each construct their own prisma client, which is a meaningful share of the connections at 09:00. share one, and tell me which jobs were relying on their own transaction isolation before i approve it", "purpose": "quickFix", "secondary": "review", "mixed": true, "difficulty": 0.5, "slice": "mixed", "lang": "en"}
{"prompt": "payment vendors behind one interface", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.65, "slice": "core", "lang": "en"}
{"prompt": "stock drifts on handheld scans", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "core", "lang": "en"}
{"prompt": "an HPC customer's security team asked six questions about credentials, shared filesystems, telemetry and job isolation, and our honest answers include a couple of \"not currently\". write the document from the code, mark those clearly, and give each one a remediation note rather than leaving it hanging this is the third time it's bitten us and i'd like it to be the last.", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "boundary", "lang": "en"}
{"prompt": "checkpoint design note assumes restarts happen on the same rank count and that the RNG is reseeded from the step, neither of which is true today. read it against the code and tell me which of its other assumptions have quietly stopped holding as our jobs got longer context if it helps: this has been open since before i joined the team.", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "boundary", "lang": "en"}
{"prompt": "three payment vendors have three client shapes, three error taxonomies and three copies of the offline voucher logic with slightly different conditions. put them behind one interface with one offline policy, keeping each vendor's on-the-wire behaviour exactly as it is, and make the differences visible in configuration rather than in branches", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "core", "lang": "en"}
{"prompt": "pipeline board is what recruiters have open all day and it currently renders four thousand cards eagerly and loses their filters on refresh. rebuild it to the spec — virtualised, filters in the URL, keyboard navigation, archived candidates visible rather than absent — without changing the drag interaction people are used to assume whoever picks it up next has no context beyond what you write.", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "our careers-site API, which twelve customers embed in their own sites from a README:\n\nGET /v1/public/jobs?tenant=&location=&department=&remote=\n  returns open jobs only; a job closed while a candidate is on the page 404s on apply\n  ordering is by posted_at desc with no tie-break, so equal timestamps reorder between requests\n  the description field is HTML we sanitise on write, not on read, and old records predate the sanitiser\nPOST /v1/public/applications\n  multipart with a CV up to 10MB; larger uploads fail with a generic 400\n  duplicate applications to the same job are accepted silently and appear twice to the recruiter\n  the response is 201 with an application id that is not shown to the candidate anywhere\n\nwrite the reference for customers embedding this, and be explicit about the duplicate behaviour", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.65, "slice": "pasted-context", "lang": "en"}
{"prompt": "the simulation group's list, with a paper deadline in eight weeks:\n\n- fix the checkpoint RNG position, which may invalidate results in the last two papers\n- decide on -ffast-math, which costs 18% if we remove it\n- fixed-order reduction, which costs 4% and a kernel rewrite\n- the 512-rank hang, which blocks the largest runs entirely\n- split the 4,000-line module so more than one person can work on it\n- job scripts that don't waste 88% of each node\n\none Fortran engineer, and the deadline is real but the correctness questions are older", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "pasted-context", "lang": "en"}
{"prompt": "HRIS field mapping is YAML plus four customer names hardcoded in the pusher, so onboarding a tenth customer means a deploy. move the special cases into the mapping format itself, keep every current customer's payload byte-identical, and add the test that proves it for each tell me if this is the wrong shape entirely, i won't be offended.", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "boundary", "lang": "en"}
{"prompt": "a 512-rank job hangs at the same collective every time and the branch that broadcasts a convergence flag looks like the culprit. confirm it, then restructure the collectives so all ranks agree on what they're calling", "purpose": "debugging", "secondary": "refactor", "mixed": true, "difficulty": 0.9, "slice": "mixed", "lang": "en"}
{"prompt": "job list widget should degrade to plain server-rendered HTML when scripts are blocked", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.55, "slice": "boundary", "lang": "en"}
{"prompt": "our runbook for the 09:00 application spike is \"watch the dashboard\". what the team actually does:\n\n- job adverts go out by email at 09:00 and the apply endpoint takes ten times its normal load for about twenty minutes\n- the first symptom is connection pool exhaustion, which shows as 500s on apply and nothing else\n- scaling the app up makes it worse, because each instance takes ten more connections\n- the actual mitigation is scaling down to 20 instances, which nobody believes until they try it\n- pgbouncer exists in staging and has never been enabled in production\n- if applications are lost rather than delayed, they are recoverable from the email bounce log, painfully\n\nwrite the runbook, and mark clearly that scaling up is the wrong instinct", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "pasted-context", "lang": "en"}
{"prompt": "the retailer's requirements for offline payments, which are now contractual:\n\n\"The point of sale shall continue to accept card payments when the payment terminal is temporarily unreachable, subject to a configurable per-store limit on the total value of offline transactions. Offline transactions shall be transmitted for authorisation within 15 minutes of connectivity being restored, without staff intervention. Declined offline transactions shall be reported to the store manager the same day. The system shall not permit offline acceptance for transactions above a configurable value.\"\n\nwe have no limits, no automatic retransmission except on restart, and no manager-facing report. i want the plan by contractual exposure", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "pasted-context", "lang": "en"}
{"prompt": "a walkthrough of how a scan becomes a stock movement would help before i touch idempotency", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "boundary", "lang": "en"}
{"prompt": "HRIS integration guide needs writing and i expect it will surface behaviours we should change rather than document — the silent failure after three retries especially. write the guide, and give me that list", "purpose": "writing", "secondary": "review", "mixed": true, "difficulty": 0.65, "slice": "mixed", "lang": "en"}
{"prompt": "careers widget documentation has to exist for the customers embedding it, and while writing it please confirm whether duplicate applications really are accepted silently, because support believes they're rejected", "purpose": "writing", "secondary": "review", "mixed": true, "difficulty": 0.6, "slice": "mixed", "lang": "en"}
{"prompt": "offline payments runbook for store managers should exist, and the till should stop printing vouchers past a sensible daily limit. write the runbook, then add the limit with a configurable default", "purpose": "writing", "secondary": "quickFix", "mixed": true, "difficulty": 0.6, "slice": "mixed", "lang": "en"}
{"prompt": "interview emails show the wrong local time for candidates in other timezones, which looks like the formatter using the interviewer's zone. confirm that, then fix it and work out how many past emails were wrong so we can decide whether to apologise i've already spent an afternoon on it and got nowhere useful.", "purpose": "debugging", "secondary": "writing", "mixed": true, "difficulty": 0.6, "slice": "mixed", "lang": "en"}
{"prompt": "is our job efficiency really 12%, or is the accounting measuring something else", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "what guarantees does the apply endpoint make once it has returned a 201 to the candidate", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.65, "slice": "boundary", "lang": "en"}
{"prompt": "simulation writes trajectory output through three routines depending on format, and two of them buffer differently so a killed job loses a different amount of data. unify them behind one writer, and tell me which format's flush behaviour we should adopt given jobs are killed by the scheduler routinely this has come up in three separate reviews now and never gets done.", "purpose": "refactor", "secondary": "planning", "mixed": true, "difficulty": 0.7, "slice": "mixed", "lang": "en"}
{"prompt": "job adverts should be verified by reading the posting back rather than trusting a 200, and we should decide first what we do when the read-back disagrees — retry, alert, or mark the advert failed. settle that, then implement flag anything you'd want to change before doing it rather than after.", "purpose": "backendImpl", "secondary": "planning", "mixed": true, "difficulty": 0.6, "slice": "mixed", "lang": "en"}
{"prompt": "candidate applications vanish between the form and the database, maybe five a day:\n\nPrismaClientUnknownRequestError: \nInvalid `prisma.application.create()` invocation:\nError occurred during query execution:\nConnectorError(ConnectorError { user_facing_error: None, kind: QueryError(PostgresError { code: \"53300\", message: \"remaining connection slots are reserved for non-replication superuser connections\" }) })\n    at Ai.handleRequestError (/app/node_modules/@prisma/client/runtime/library.js:121:6412)\n    at async POST (/app/app/api/apply/route.ts:88:22)\n\nconnection pool: 10 per instance, 40 instances behind the autoscaler, postgres max_connections 100\nthe apply endpoint is the only one that spikes, because job ads go out by email at 09:00", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "pasted-context", "lang": "en"}
{"prompt": "so our simulation gives different answers on the new cluster and i can't tell which is right:\n\n  old cluster (gcc 11, -O2, AVX2):  total energy = -1042.884112 Ha\n  new cluster (gcc 14, -O3, AVX512): total energy = -1042.884109 Ha\n  reference (published):             total energy = -1042.884110 Ha\n\ncompiler flags on the new cluster: -O3 -march=native -ffast-math -funroll-loops\ncompiler flags on the old cluster: -O2 -march=haswell\n\nthe kernel sums a 4-million-element array of pair interactions in a plain loop, and the reduction order differs between the two builds", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "pasted-context", "lang": "en"}
{"prompt": "tills stop taking card payments in one store every afternoon, always around the same time:\n\n14:02:11.114 [pos-04] payment start amount=1240 currency=GBP\n14:02:11.882 [pos-04] terminal connect 192.168.10.44:9100\n14:02:41.902 [pos-04] terminal read timeout after 30s\n14:02:41.905 [pos-04] payment failed, offline voucher printed\n14:02:44.001 [pos-04] terminal connect 192.168.10.44:9100\n14:03:14.020 [pos-04] terminal read timeout after 30s\n\nstore wifi shows a channel change at 14:00 daily, which is when the bakery ovens come on, and the terminals are on wifi while the tills are wired", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "pasted-context", "lang": "en"}
{"prompt": "seit dem letzten Deploy verschwinden Bewerbungen aus dem Talent-Pool, aber nur bei einem Kunden:\n\nSELECT count(*) FROM applications WHERE tenant_id = 'acme' AND stage = 'pool';\n  vor dem Deploy:  41.882\n  nach dem Deploy: 1.204\n\nMigration aus dem Deploy:\n  UPDATE applications SET stage = 'archived'\n   WHERE stage = 'pool' AND updated_at < now() - interval '180 days';\n\nbei diesem Kunden werden Bewerbungen im Pool nicht angefasst, solange niemand sie öffnet, also ist updated_at das Erstellungsdatum; bei allen anderen Kunden läuft ein Nurturing-Job, der die Datensätze regelmäßig berührt", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "pasted-context", "lang": "de"}
{"prompt": "ok so the POS app crashes on a specific till model when printing a receipt with more than about forty lines:\n\nFATAL EXCEPTION: main\nProcess: io.lumenpos.till, PID: 4412\njava.lang.OutOfMemoryError: Failed to allocate a 41943040 byte allocation with 12582912 free bytes\n\tat android.graphics.Bitmap.nativeCreate(Native Method)\n\tat io.lumenpos.print.ReceiptRenderer.renderToBitmap(ReceiptRenderer.kt:141)\n\tat io.lumenpos.print.PrintJob.execute(PrintJob.kt:88)\n\nwe render the whole receipt into one bitmap at 203dpi before sending it to the printer, and these tills have 1GB of RAM", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.65, "slice": "pasted-context", "lang": "en"}
{"prompt": "could you look at the pooling before i sign this off? the 09:00 spike is what worries me:\n\nexport const prisma = new PrismaClient({\n  datasources: { db: { url: process.env.DATABASE_URL } },\n})\n\n// route.ts\nexport async function POST(req: Request) {\n  const body = await req.json()\n  const application = await prisma.application.create({ data: { ...body } })\n  await sendConfirmationEmail(application)      // awaited, 400-800ms\n  await notifyRecruiters(application)           // awaited, fans out to up to 12 recipients\n  return Response.json({ id: application.id })\n}\n\nforty instances, ten connections each, postgres allows a hundred, and the two awaits hold the request open for over a second", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "pasted-context", "lang": "en"}
{"prompt": "honestly the reduction kernel, which is where our numbers differ between clusters:\n\nsubroutine pair_energy(n, r, q, energy)\n  integer, intent(in) :: n\n  real(8), intent(in) :: r(3,n), q(n)\n  real(8), intent(out) :: energy\n  integer :: i, j\n  energy = 0.0d0\n  !$omp parallel do reduction(+:energy) private(j)\n  do i = 1, n\n    do j = i+1, n\n      energy = energy + q(i)*q(j)/norm2(r(:,i) - r(:,j))\n    end do\n  end do\nend subroutine\n\nbuilt with -ffast-math on the new cluster, thread count differs between runs, and the result is quoted to ten significant figures in our papers", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "pasted-context", "lang": "en"}
{"prompt": "right, the idempotency handling for handheld scans, which i think is behind the stock drift:\n\n@PostMapping(\"/v1/scans\")\npublic ResponseEntity<Void> scan(@RequestBody ScanBatch batch) {\n    for (Scan s : batch.scans()) {\n        String key = s.deviceId() + \":\" + s.localId();\n        if (seen.putIfAbsent(key, Boolean.TRUE) == null) {\n            stock.apply(s.sku(), -s.qty(), \"scan\");\n        }\n    }\n    return ResponseEntity.accepted().build();\n}\n\n`seen` is a Caffeine cache with a one hour expiry, localId is a device counter that resets on docking, and there are 400 handhelds across the estate", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "pasted-context", "lang": "en"}
{"prompt": "quick one — the checkpoint format's design note, written when jobs ran for eight hours rather than three weeks:\n\n## Checkpoint contents\nState arrays, the current step, the input parameters and the RNG seed. Written every 10,000 steps to the parallel filesystem.\n\n## Assumptions\n- A restart continues a run that would otherwise have been identical.\n- The RNG is reseeded deterministically from the seed and the step.\n- Checkpoints are read back on the same number of ranks.\n\n## Not covered\nRestarting on a different rank count. Changing the thermostat between restarts. Reproducibility across compilers.\n\nwe now restart on different rank counts routinely, and the RNG is not reseeded from the step at all", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "pasted-context", "lang": "en"}
{"prompt": "fyi the migration that archived a customer's talent pool, which i'd like reviewed before we write the recovery:\n\n-- 20260722_archive_stale_pool.sql\nUPDATE applications\n   SET stage = 'archived', archived_at = now()\n WHERE stage = 'pool'\n   AND updated_at < now() - interval '180 days';\n\n-- rollback (as written in the migration file)\n-- none\n\napplications has no history table; archived_at is the only trace; and updated_at means \"last touched by anything\", which for one tenant is the creation date because they don't run the nurturing job", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "pasted-context", "lang": "en"}
{"prompt": "heads up: the job scheduler script we hand to researchers, which i suspect is why jobs sit in the queue:\n\n#!/bin/bash\n#SBATCH --nodes=64\n#SBATCH --ntasks-per-node=8\n#SBATCH --cpus-per-task=16\n#SBATCH --time=24:00:00\n#SBATCH --exclusive\n#SBATCH --mem=0\n\nexport OMP_NUM_THREADS=16\nexport OMP_PLACES=cores\nsrun ./simulate input.dat\n\nthe cluster has 128 cores per node, our job asks for 8 tasks × 16 threads = 128, exclusive plus mem=0 takes the whole node, and the queue's backfill window is four hours", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "pasted-context", "lang": "en"}
{"prompt": "support's notes on the archived applications, which we need to turn into a customer explanation:\n\n- a migration archived applications in the talent pool that hadn't been updated in 180 days\n- for most customers a nurturing job touches these records, so almost nothing was archived\n- for one customer the nurturing job is disabled, so updated_at was the creation date and 40,000 records were archived\n- the records are not deleted; they're in the archived stage and can be moved back\n- moving them back individually is a click each; there is no bulk unarchive in the product\n- we can do it with a script, but archived_at is the only evidence of which records we touched\n\nwrite the customer explanation and the internal note about what we're doing to restore them", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "pasted-context", "lang": "en"}
{"prompt": "incident notes from the till outage, and the retailer wants a written explanation:\n\n14:02 store 44 reports card payments failing on all six tills\n14:20 other stores unaffected; store 44's terminals are on wifi, tills are wired\n14:35 store wifi logs show a channel change at 14:00, coinciding with the bakery ovens starting\n15:10 terminals moved to the wired network as a workaround, payments resume\n16:00 root cause: the terminals lose association during the channel change and our client has a 30 second read timeout with no retry\n17:30 during the outage, staff used offline vouchers for 412 transactions, of which 8 were later declined\n\nthe retailer's question is why a wifi hiccup takes down payments at all when the tills are wired", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.65, "slice": "pasted-context", "lang": "en"}
{"prompt": "as notas da reunião sobre reprodutibilidade, para transformar em documento de decisão:\n\n- os resultados diferem entre o cluster antigo e o novo na décima casa decimal\n- a causa provável é -ffast-math combinada com uma redução OpenMP cuja ordem varia com o número de threads\n- publicámos valores com dez algarismos significativos em dois artigos\n- opções: fixar a ordem da redução, remover -ffast-math, ou passar a publicar com incerteza declarada\n- remover -ffast-math custa cerca de 18% de desempenho segundo o teste que fizemos\n- a ordem fixa custa cerca de 4% e obriga a reescrever o kernel\n\nescreve a nota de decisão com as opções, o custo de cada uma e uma recomendação para o grupo", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "pasted-context", "lang": "pt"}
{"prompt": "small thing but the questions our new HPC customer's IT security team sent, which need answering as a document:\n\n\"What data does the job submission tool send to your servers, and can it operate fully offline? Where are user credentials stored on shared login nodes? Does your software write anything to a world-readable location? Can a user's job read another user's checkpoint data on the shared filesystem? How do you handle a job that exceeds its allocation — is it killed, and by whom? What telemetry do you collect from academic sites and can it be disabled?\"\n\nour honest answers are mixed at best; write the document from the code, and mark clearly which answers are \"not currently\" rather than presenting them as design decisions", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "pasted-context", "lang": "en"}
{"prompt": "社内向けに、レジ端末のオフライン運用手順が文書化されていません。現状は次の通りです:\n\n- カード端末が 30 秒応答しない場合、レジはオフラインバウチャーを印刷する（リトライはしない）\n- オフラインバウチャーは後で一括送信されるが、与信が通らない場合は店舗の損失になる\n- 一日あたりのオフライン上限は設定されておらず、先週は 412 件が積み上がった\n- レジ側のログはローカルにしか残らず、7 日で上書きされる\n- 店舗マネージャーはオフライン件数を確認する画面を持っていない\n- 復旧後の再送はレジの再起動時にのみ行われる\n\n設定例:\n\nterminal:\n  read_timeout_ms: 30000\n  retries: 0\n  offline_voucher_limit: null\n\n店舗マネージャー向けの手順書として、また社内の開発者向けの注意点として、それぞれまとめてください\n\n先週のオフライン件数（店舗別、上位）:\n  store 44   412 件  合計 £18,204   後日否認 8 件\n  store 12    88 件  合計  £3,110   後日否認 1 件\n  store 07    41 件  合計  £1,204   後日否認 0 件\n\n関連する設定とログ:\n\n  terminal:\n    read_timeout_ms: 30000\n    retries: 0\n    offline_voucher_limit: null\n    resubmit_on: [\"app_start\"]\n\n  14:02:41 [pos-04] terminal read timeout after 30s\n  14:02:41 [pos-04] payment failed, offline voucher printed\n\n店舗マネージャーはこの数字をどこからも見られません", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "pasted-context", "lang": "ja"}
{"prompt": "not urgent, but the ATS's integration guide for HR systems, written for one customer and now used by nine:\n\nwe push hires to the customer's HRIS on offer acceptance:\n  POST {customer_endpoint}/employees   with our own field names, mapped per customer in a YAML file\n  retries: 3, then a support ticket is created and someone re-runs it by hand\n  the mapping file lives in our repo, so a new field for one customer is a deploy\n  four customers require a custom date format and two require a custom ID scheme\n  nobody has documented which customer needs what; the YAML comments are the only record\n  a failed push leaves the offer accepted in our system with no indication that the HRIS is out of sync\n\nwrite the integration guide, and separately list what should be configuration rather than code", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "pasted-context", "lang": "en"}
{"prompt": "eslint and tsc on the ATS, gate goes on next sprint:\n\napp/api/apply/route.ts:88:22 - error TS2345: Argument of type 'unknown' is not assignable to parameter of type 'ApplicationCreateInput'\napp/lib/mail.ts:41:9 - warning: Promise returned in function argument where a void return was expected  @typescript-eslint/no-misused-promises\napp/lib/tenant.ts:141:3 - error TS2532: Object is possibly 'undefined'\napp/components/PipelineBoard.tsx:212:5 - warning: React Hook useEffect has a missing dependency: 'tenantId'\napp/lib/dates.ts:22:11 - error TS2769: No overload matches this call (formatInTimeZone)\n\n3 errors, 2 warnings, and dates.ts is the file behind the timezone emails", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.35, "slice": "pasted-context", "lang": "en"}
{"prompt": "compiler warnings from the simulation build, which we've been ignoring for years:\n\nsrc/pair.f90:141:22: warning: 'energy' may be used uninitialized [-Wmaybe-uninitialized]\nsrc/io.f90:88:5: warning: array 'buf' is used uninitialized [-Wuninitialized]\nsrc/thermostat.f90:41:9: warning: comparison of real values with == [-Wfloat-equal]\nsrc/checkpoint.f90:212:13: warning: unused variable 'rng_pos' [-Wunused-variable]\nsrc/main.f90:22:1: warning: obsolescent feature: COMMON block\n\n41 warnings total; checkpoint.f90's unused rng_pos is exactly the variable our restart bug is about", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "pasted-context", "lang": "en"}
{"prompt": "genuinely puzzled by this: the till app's config across store types, and only the small stores crash:\n\n# flagship stores\nprint:\n  render_mode: banded\n  band_height_px: 512\n  bitmap_config: ALPHA_8\ndevice: { ram_mb: 4096, android: 13 }\n\n# small stores\nprint:\n  render_mode: single\n  band_height_px: null\n  bitmap_config: ARGB_8888\ndevice: { ram_mb: 1024, android: 11 }\n\nbanded rendering was added for the flagship rollout and never enabled anywhere else\n\nand the crash rate per store type over the last month:\n\n  flagship (4GB, android 13, banded)   0 crashes / 41,882 receipts\n  standard (2GB, android 12, single)   14 crashes / 88,214 receipts\n  small    (1GB, android 11, single)   412 crashes / 21,004 receipts\n\nevery crash is the same OutOfMemoryError in renderToBitmap, and the receipts that trigger it are stocktake reports and long promotional receipts", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.35, "slice": "pasted-context", "lang": "en"}
{"prompt": "slurm accounting for our group this month, and we're being asked why utilisation is low:\n\nJobID    Nodes  Elapsed    CPUTime   TotalCPU   Efficiency\n881204   64     23:58:12   1536:00   184:12     11.9%\n881207   64     23:59:02   1536:00   201:44     13.1%\n881209   32     04:12:44   134:48    128:02     94.9%\n881211   64     23:58:44   1536:00   190:22     12.4%\n\nthe 64-node jobs request 8 tasks per node with 16 threads each on 128-core nodes, and the three long ones all hit the wall clock rather than converging", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.45, "slice": "pasted-context", "lang": "en"}
{"prompt": "dependabot on the ATS, four open and one blocks the store submission:\n\nnext 15.1.0 → 15.4.2          (patch series; middleware behaviour changed for rewrites)\n@prisma/client 5.19 → 6.2     (major; connection pool defaults changed, `connection_limit` now per-process)\ndate-fns-tz 2.0.0 → 3.2.0     (major; formatInTimeZone signature changed)\nnodemailer 6.9.13 → 6.10.1    (patch)\n\nthe date-fns-tz major is the library our timezone bug is in, and the prisma major changes exactly the pooling behaviour we're fighting", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.45, "slice": "pasted-context", "lang": "en"}
{"prompt": "les seuils d'alerte de la plateforme de recrutement, on nous réveille pour rien :\n\n- alert: ApplyErrorRate\n  expr: rate(http_requests_total{route=\"/api/apply\",code=~\"5..\"}[5m]) > 0\n  for: 1m\n  labels: { severity: page }\n\n- alert: DbConnections\n  expr: pg_stat_activity_count > 90\n  for: 5m\n  labels: { severity: ticket }\n\n- alert: SyncFailures\n  expr: increase(partner_sync_failed_total[1h]) > 0\n  for: 0m\n  labels: { severity: ticket }\n\nen pratique : le taux d'erreur dépasse zéro tous les matins à 9h ; la saturation des connexions n'est qu'un ticket alors que c'est la cause ; et la synchronisation vers un job board renvoyait 200 tout en jetant nos annonces, donc aucune alerte n'a été déclenchée pendant trois semaines", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.45, "slice": "pasted-context", "lang": "fr"}
{"prompt": "this module does I/O, domain decomposition and the physics in one file, and every change touches all three:\n\nmodule simulate\n  ! reads input.dat, including three formats we've supported over the years\n  ! decides the domain decomposition from the rank count\n  ! allocates the state arrays\n  ! runs the integrator, calling the pair kernel and the thermostat\n  ! writes checkpoints and trajectory output\n  ! handles MPI setup and teardown\nend module\n\n4,000 lines of Fortran, one test that runs a tiny system and compares against a stored trajectory", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "pasted-context", "lang": "en"}
{"prompt": "quarter planning input, needs sequencing:\n\n- 40,000 applications were archived by a migration at one customer and unarchiving them is manual\n- the 09:00 spike takes the apply endpoint down most mornings and scaling up makes it worse\n- our simulation results differ between clusters in the tenth significant figure, and we publish ten\n- the POS crash on small tills affects 300 stores and the workaround is telling staff to print shorter receipts\n- checkpoint restarts are silently wrong for long jobs, which may affect published work\n- one Fortran engineer, two platform engineers, and the POS team is three people\n- there's an HPC customer procurement in october with a security questionnaire we can't currently pass", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "pasted-context", "lang": "en"}
{"prompt": "architecture ticket, i want the thinking before code:\n\nATS-410 — Tenant isolation enforcement\nTenant scoping is applied by hand in fourteen places with three different patterns and at least one omission. The proposal is row-level security in postgres with the tenant set per connection, so a missing filter fails closed. Concerns: our connection pooler is transaction-scoped and the tenant would have to be set per transaction; some background jobs legitimately run across tenants; raw queries in reports bypass the ORM entirely; and we have no test that would catch a missing filter today, so we cannot prove the change is safe or that it was needed.", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "pasted-context", "lang": "en"}
{"prompt": "spec for the till's payment screen, which cashiers use hundreds of times a day:\n\nPayment screen\n- Amount due in the largest type on the screen, unmissable from a metre away.\n- Payment method buttons at least 72dp tall, arranged so the most common method is under the thumb.\n- While waiting on the terminal: a progress state that names what is happening (\"waiting for card\", \"authorising\") rather than a generic spinner, with the elapsed seconds visible.\n- On timeout: an explicit choice — retry, or take the payment offline — never an automatic silent fallback.\n- Offline mode shows a persistent banner with the count and total value of offline transactions taken today.\n- Errors from the terminal are shown in the words the cashier needs (\"ask for another card\"), with the vendor code available on a long press for support.\n- The screen must be operable with gloves and must never require a precise tap.", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "pasted-context", "lang": "en"}
{"prompt": "accessibility findings for the careers site widget, which our customers embed and are legally responsible for:\n\n1. The job list is a div soup with no list semantics and no headings, so screen reader users cannot navigate it.\n2. The filter controls are custom dropdowns with no keyboard support and no announced state.\n3. The apply form's file input is hidden behind a styled button with no label association.\n4. Required field errors appear only as red borders.\n5. The widget sets its own font size in px, overriding the host page's user scaling.\n6. Focus is trapped in the apply modal with no escape key handling.\n7. Colour contrast on the \"apply now\" button fails at 3.2:1 against our default background, which most customers keep.", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "pasted-context", "lang": "en"}
{"prompt": "design tokens versus what the recruiter app actually uses:\n\ntokens:\n  color.surface      #FFFFFF / #0F1216\n  color.text         #0F1216 / #E8EDF3\n  color.muted        #626C76\n  color.stage.new    #0B62D6\n  color.stage.offer  #1A7F37\n  color.stage.reject #B42318\n  space 4/8/12/16/24/32, radius 6/10, focus 2px offset 2\n  type: title 18/24, body 14/20, caption 12/16\n\nthe recruiter app: eight hardcoded colours including three blues, stage colour used as the only stage indicator on cards, focus removed on the card press target, paddings of 5/9/13/17, and a card title at 15px which is in no scale\n\nbring it onto the tokens and give stage a non-colour indicator on the card", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.55, "slice": "pasted-context", "lang": "en"}
{"prompt": "banded rendering on the small tills", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "core", "lang": "en"}
{"prompt": "date-fns-tz major, before the timezone fix", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "core", "lang": "en"}
{"prompt": "enable pgbouncer in production", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "core", "lang": "en"}
{"prompt": "rotate the job board API key", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.25, "slice": "core", "lang": "en"}
{"prompt": "el correo de entrevista sale sin zona horaria", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.25, "slice": "core", "lang": "es"}
{"prompt": "our database connection settings versus what postgres allows:\n\napp instances (autoscaler): min 10, max 60\nprisma connection_limit: 10 (default, not set explicitly)\npostgres max_connections: 100\npostgres superuser_reserved_connections: 3\npgbouncer: deployed in staging only, transaction pooling\n\nat 09:00 the autoscaler goes to 40+ instances within two minutes because the apply endpoint's latency rises, which is itself caused by connection contention\n\nand what 09:00 looks like from the database side:\n\n  08:58  connections 41   apply p95 180ms   instances 12\n  09:00  connections 96   apply p95 2.4s    instances 18\n  09:02  connections 100  apply p95 timeout instances 41\n  09:06  connections 100  apply 500s 38%    instances 60 (autoscaler ceiling)\n  09:22  connections 44   apply p95 210ms   instances 14\n\npg_stat_activity during the spike is almost entirely idle-in-transaction, which is the two awaited emails inside the request", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "pasted-context", "lang": "en"}
{"prompt": "the HRIS integration spec from our largest customer, which we implement:\n\nPOST /api/v2/employees\n  auth: OAuth2 client credentials, token valid 15 minutes, their token endpoint rate limits to 1/minute\n  body: their field names, dates as YYYY-MM-DD, employee id assigned by them and returned in the response\n  a duplicate push for the same offer must be idempotent on our `external_ref`, which they store but do not index\n  they respond 202 and process asynchronously; failures arrive by email to a shared mailbox, not by callback\n  a rejected employee record leaves the offer accepted on our side with no automatic reconciliation\n  they have a nightly maintenance window during which everything returns 503\n  we push about 400 hires a month, in bursts on the first of the month", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "pasted-context", "lang": "en"}
{"prompt": "HRIS pusher should reconcile nightly rather than leaving an offer accepted and unsynced", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.65, "slice": "boundary", "lang": "en"}
{"prompt": "a short note on why the apply endpoint is moving to a queue, for the decision log", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "core", "lang": "en"}
{"prompt": "drop -ffast-math from the release build", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "boundary", "lang": "en"}
{"prompt": "DbConnections should page, not ticket", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "boundary", "lang": "en"}
{"prompt": "cap prisma connections per instance", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.35, "slice": "boundary", "lang": "en"}
{"prompt": "OMP_NUM_THREADS passt nicht zur Knotengröße", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.35, "slice": "boundary", "lang": "de"}
{"prompt": "uninitialised `energy` in pair.f90", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.35, "slice": "boundary", "lang": "en"}
{"prompt": "archived candidates should be visible, not absent", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.35, "slice": "core", "lang": "en"}
{"prompt": "求人ウィジェットの文字サイズが固定です", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "core", "lang": "ja"}
{"prompt": "required errors are red borders only", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "boundary", "lang": "en"}
{"prompt": "terminal errors show vendor codes to cashiers", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "boundary", "lang": "en"}
{"prompt": "board drops frames past 2000 cards", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.45, "slice": "boundary", "lang": "en"}
{"prompt": "split the 4000-line simulate module", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "un seul contrôle de locataire, partout", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "fr"}
{"prompt": "`stage` naming across the API", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "boundary", "lang": "en"}
{"prompt": "extract the offline voucher logic", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "boundary", "lang": "en"}
{"prompt": "inline `padLeft`, one caller", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.2, "slice": "boundary", "lang": "en"}
{"prompt": "changelog for the POS release", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "core", "lang": "en"}
{"prompt": "nota aos clientes sobre as candidaturas arquivadas", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.45, "slice": "core", "lang": "pt"}
{"prompt": "document the duplicate application behaviour", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "boundary", "lang": "en"}
{"prompt": "summarise the tenant isolation proposal", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "boundary", "lang": "en"}
{"prompt": "PR body for the checkpoint fix", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.35, "slice": "boundary", "lang": "en"}
{"prompt": "can a restart change published results?", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "core", "lang": "en"}
{"prompt": "¿por qué el 09:00 tumba el endpoint de solicitudes?", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "es"}
{"prompt": "walk me through the offer-to-HRIS push", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "boundary", "lang": "en"}
{"prompt": "card payments fail every afternoon in one store", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "boundary", "lang": "en"}
{"prompt": "warum unterscheiden sich die Ergebnisse je Cluster?", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "boundary", "lang": "de"}
{"prompt": "endpoint for a store's offline transactions", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "carry it on", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "vague-eval", "lang": "en"}
{"prompt": "less flaky", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.55, "slice": "vague-eval", "lang": "en"}
{"prompt": "pick the order yourself", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.55, "slice": "vague-eval", "lang": "en"}
{"prompt": "lo del cluster, continúa", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "vague-eval", "lang": "es"}
{"prompt": "scanners again", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "vague-eval", "lang": "en"}
{"prompt": "clean it as you go", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.45, "slice": "vague-eval", "lang": "en"}
{"prompt": "same shape as last time", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "vague-eval", "lang": "en"}
{"prompt": "security answers for procurement", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "vague-eval", "lang": "en"}
{"prompt": "nothing that needs a deploy", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "vague-eval", "lang": "en"}
{"prompt": "give it a read", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.45, "slice": "vague-eval", "lang": "en"}
{"prompt": "どれから着手するかはお任せします", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "vague-eval", "lang": "ja"}
{"prompt": "next on the list", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "vague-eval", "lang": "en"}
{"prompt": "09:00 application spike takes down the apply endpoint most mornings and every instinct the team has — scaling up, raising the pool size — makes it worse. i'd like the actual capacity model worked out, including whether pgbouncer solves it or just moves the queue, and what the endpoint should do when it genuinely can't accept an application", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "core", "lang": "en"}
{"prompt": "twelve customers embed our careers widget in their own sites and are legally responsible for its accessibility, which is currently poor enough to be a problem for them. write the integration documentation, including the accessibility characteristics they're inheriting and what they can configure, without pretending the widget is in better shape than it is if the answer is that it's fine as it is, that's a useful answer too.", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "retailer wants a written explanation of why a wifi channel change in one store stopped card payments on wired tills, which is a fair question. write the incident report for their operations team — the terminal association, our timeout, the offline vouchers and the eight that were later declined — without hiding behind the word \"connectivity\"", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "runbook for the morning spike should say plainly that scaling up is the wrong instinct, because three separate people have made it worse that way. write it in the order someone would need at 09:05, with the connection maths spelled out so the advice is believable rather than folklore a rough ordering matters more to me than a complete answer right now.", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.65, "slice": "boundary", "lang": "en"}
{"prompt": "checkpoint format is detected by whether a field looks plausible, with no magic number and no version, and we now need to add the RNG position. give it a proper header with a version, keep reading both existing formats, and make an unrecognised file a clear error rather than a plausible misinterpretation i'd rather have the reasoning written down than a quick answer.", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "payment screen currently falls back to offline silently on a timeout, which is how a store took four hundred vouchers without noticing. rebuild it to the spec — named progress states, an explicit choice on timeout, a persistent offline banner — and keep every tap target usable with gloves the sooner we know roughly how big this is, the better for planning.", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "boundary", "lang": "en"}
{"prompt": "tenant isolation needs an approach decided and one concrete step taken. work through the options with me, then implement the repository-level scoping for the reports module, which is where the raw queries live", "purpose": "planning", "secondary": "backendImpl", "mixed": true, "difficulty": 0.85, "slice": "mixed", "lang": "en"}
{"prompt": "morning spike needs a capacity model and it needs the apply endpoint to stop dropping applications tomorrow. give me the model, then make the endpoint enqueue rather than write synchronously so a full pool delays rather than loses", "purpose": "planning", "secondary": "backendImpl", "mixed": true, "difficulty": 0.85, "slice": "mixed", "lang": "en"}
{"prompt": "our till's payment flow has three implementations depending on terminal vendor:\n\n// VendorA.kt — synchronous socket, 30s timeout, no retry\n// VendorB.kt — vendor SDK with callbacks, retries internally 3 times, reports only final failure\n// VendorC.kt — HTTP to a local bridge, 10s timeout, retries once, returns a different error taxonomy\n\nthe till's payment screen handles all three with a when(vendor) block in the view model, and offline voucher logic is duplicated in each branch with slightly different conditions", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "pasted-context", "lang": "en"}
{"prompt": "the schema we agreed for offline transactions, now it needs building on the till and the backend:\n\nCREATE TABLE offline_transactions (\n  id            uuid PRIMARY KEY,\n  store_id      uuid NOT NULL,\n  till_id       text NOT NULL,\n  amount_cents  bigint NOT NULL,\n  currency      char(3) NOT NULL,\n  taken_at      timestamptz NOT NULL,\n  submitted_at  timestamptz,\n  outcome       text CHECK (outcome IN ('approved','declined','expired')),\n  card_ref      text NOT NULL\n);\n\nthe till holds these locally until connectivity returns, submits within 15 minutes without staff action, enforces a per-store value limit and a per-transaction cap, and the store manager needs a same-day view of anything declined", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "pasted-context", "lang": "en"}
{"prompt": "unsere Offline-Zahlungen brauchen erst ein Konzept mit Limits und automatischer Nachübertragung und danach die Umsetzung der Limits auf der Kasse, weil das der vertraglich heikelste Teil ist", "purpose": "planning", "secondary": "backendImpl", "mixed": true, "difficulty": 0.8, "slice": "mixed", "lang": "de"}
{"prompt": "escribe la guía del operador para el pico de las 09:00 y comprueba en el código si escalar instancias realmente empeora la situación, porque quiero poder demostrarlo y no solo afirmarlo", "purpose": "writing", "secondary": "review", "mixed": true, "difficulty": 0.65, "slice": "mixed", "lang": "es"}
{"prompt": "active-application predicate exists three times and the report includes records the product hides. unify them, then document what each stage means, because \"archived\" apparently means two different things depending on which migration set it", "purpose": "refactor", "secondary": "writing", "mixed": true, "difficulty": 0.65, "slice": "mixed", "lang": "en"}
{"prompt": "receipt rendering should be banded everywhere rather than only on the flagship tills. make the change, and confirm from the store inventory how many devices are actually on the old configuration before we call it fixed", "purpose": "refactor", "secondary": "review", "mixed": true, "difficulty": 0.55, "slice": "mixed", "lang": "en"}
{"prompt": "fourteen tenant checks should become one mechanism, and i'd like to know first which of the fourteen is actually wrong rather than merely different. audit them, then unify", "purpose": "refactor", "secondary": "review", "mixed": true, "difficulty": 0.75, "slice": "mixed", "lang": "en"}
{"prompt": "applications are lost rather than delayed during the morning spike, which the team assumes is the connection pool. verify that assumption end to end, then make the failure mode a delay rather than a loss", "purpose": "debugging", "secondary": "backendImpl", "mixed": true, "difficulty": 0.8, "slice": "mixed", "lang": "en"}
{"prompt": "careers widget fails seven accessibility items and our customers carry the legal risk. fix them, and write the statement customers can put on their own accessibility pages", "purpose": "frontendImpl", "secondary": "writing", "mixed": true, "difficulty": 0.75, "slice": "mixed", "lang": "en"}
{"prompt": "offline transaction limits need the semantics agreed — per store, per till, per day, and what happens at the boundary mid-transaction. decide that with me, then implement on both sides", "purpose": "backendImpl", "secondary": "planning", "mixed": true, "difficulty": 0.75, "slice": "mixed", "lang": "en"}
{"prompt": "i'd like to understand how the autoscaler decides to add instances during the spike", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.65, "slice": "core", "lang": "en"}
{"prompt": "why does the nurturing job touch every pool record, and what does it actually change", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "boundary", "lang": "en"}
{"prompt": "is it expected that a job advert can be accepted by the board and then silently discarded", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "boundary", "lang": "en"}
{"prompt": "pouvez-vous m'expliquer comment le point de restauration gère un redémarrage sur un nombre de rangs différent ?", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "boundary", "lang": "fr"}
{"prompt": "docs/checkpoints.md claims restarts are bit-for-bit reproducible, which they are not", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "core", "lang": "en"}
{"prompt": "partner changelog needs an entry for the job board key rotation and what it broke", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "boundary", "lang": "en"}
{"prompt": "doc comment on renderToBitmap claims it streams, which it very much does not", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.45, "slice": "boundary", "lang": "en"}
{"prompt": "write the note to affected candidates whose interview email showed the wrong time", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.45, "slice": "boundary", "lang": "en"}
{"prompt": "health endpoint reports the sync healthy while the job board has discarded everything for three weeks", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "core", "lang": "en"}
{"prompt": "apply endpoint awaits two emails before responding, which holds a connection for over a second", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.4, "slice": "core", "lang": "en"}
{"prompt": "staging runs one app instance and prod runs forty, with the same per-instance connection limit", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "boundary", "lang": "en"}
{"prompt": "how should we handle a customer who wants their ATS data in their own region", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "core", "lang": "en"}
{"prompt": "what's the right way to validate a simulation change when the reference is a published number", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.85, "slice": "core", "lang": "en"}
{"prompt": "i want a position on whether the till should hold stock state locally or always ask the backend", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "core", "lang": "en"}
{"prompt": "three customers want single sign-on into the recruiter app with different providers", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "boundary", "lang": "en"}
{"prompt": "we need a plan for running the simulation on GPUs, which the next cluster procurement assumes", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.9, "slice": "boundary", "lang": "en"}
{"prompt": "stores run four android versions and we've never dropped one", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "boundary", "lang": "en"}
{"prompt": "what should happen to a candidate's data when a customer leaves us", "purpose": "planning", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "boundary", "lang": "en"}
{"prompt": "an endpoint that returns an application's full stage history with who moved it and when", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.55, "slice": "core", "lang": "en"}
{"prompt": "pipeline board needs a bulk action bar that states how many candidates it will affect", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.45, "slice": "core", "lang": "en"}
{"prompt": "simulation's progress output should be parseable rather than a wall of formatted text", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.45, "slice": "boundary", "lang": "en"}
{"prompt": "whatever we can ship before the audit", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "vague-eval", "lang": "en"}
{"prompt": "rest of the till work", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "vague-eval", "lang": "en"}
{"prompt": "email templates format dates with the server's locale and timezone rather than the recipient's, which is the same class of bug as the interview emails but affects every template we send. centralise the formatting with the recipient's zone as the input, and tell me which templates change appearance for which recipients before we ship it", "purpose": "refactor", "secondary": "review", "mixed": true, "difficulty": 0.65, "slice": "mixed", "lang": "en"}
{"prompt": "till keeps two receipt models, one for the printer and one for the customer copy, synchronised by hand and already divergent on discount lines. merge them, and confirm from the last month's receipts whether any customer copy has actually shown a different total from the printed one it doesn't have to be elegant, it has to be defensible in a review.", "purpose": "refactor", "secondary": "review", "mixed": true, "difficulty": 0.65, "slice": "mixed", "lang": "en"}
{"prompt": "internal page on the apply pipeline stops at \"the form posts to the API\", and everything after — the two awaited emails, the recruiter fan-out, the connection cost of each — is why nobody understands the morning spike. write the page properly, following one application from the form to the recruiter's inbox", "purpose": "writing", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "boundary", "lang": "en"}
{"prompt": "job scripts we hand researchers request eight tasks of sixteen threads on 128-core nodes and then take the whole node exclusively, which is most of our reported 12% efficiency. fix the templates, and add a comment explaining the arithmetic so the next person doesn't quietly restore the old numbers there's no rush on this week specifically, but it keeps costing us time.", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.5, "slice": "boundary", "lang": "en"}
{"prompt": "nobody has been able to tell me what happens to an offer when the HRIS push fails all three attempts — whether the ticket that gets created is actually worked, whether the offer shows any sign of being unsynced, and how the customer finds out. trace it end to end and tell me how many offers are currently in that state", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "recruiter reports module builds its queries with raw SQL and a tenant id interpolated by hand, and one of them takes the tenant from a query parameter rather than the session. read the module and tell me whether a recruiter can reach another tenant's data today, with the specific request that would do it if so", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.8, "slice": "core", "lang": "en"}
{"prompt": "simulation reads input files three different ways and two of them treat comment characters differently, which means the same input file can produce two different runs depending on which path parses it. unify the reader, and check whether any of our stored regression inputs relies on the lenient behaviour the last person who touched this left, so there's nobody to ask.", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "`Application.stage` is a pipeline stage in the API, a lifecycle state in the database and something in between in the reports, which is why archived means two different things. give the concepts separate names and migrate the data, keeping the API's field names unchanged for the customers embedding our widget i'd like enough detail that i can hand it to someone else to finish.", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "scan idempotency needs a device-persistent identifier rather than a counter that resets when a handheld is docked, which means the handhelds need a stable id we can trust across firmware updates and battery swaps. design that with the constraint that four hundred devices are already in stores, then implement the backend side we've been burned by guessing at this before, so evidence over instinct please.", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "a manager screen showing today's offline transactions and their outcomes is now contractual, and it has to work on the till itself because that's the only device a store manager has. build it against the offline transaction table, including the declined ones from previous days that nobody has acknowledged if the answer is that it's fine as it is, that's a useful answer too.", "purpose": "frontendImpl", "secondary": null, "mixed": false, "difficulty": 0.6, "slice": "core", "lang": "en"}
{"prompt": "stock levels", "purpose": "debugging", "secondary": null, "mixed": false, "difficulty": 0.55, "slice": "vague-eval", "lang": "en"}
{"prompt": "one receipt model, not two", "purpose": "quickFix", "secondary": null, "mixed": false, "difficulty": 0.3, "slice": "core", "lang": "en"}
{"prompt": "ATS has three ways of resolving the current tenant — a session field, a header the reports module trusts, and a query parameter one legacy endpoint still accepts — and new code copies whichever it finds first. reduce it to one resolution path with the others explicitly rejected, and keep every current client working while we migrate them", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "core", "lang": "en"}
{"prompt": "till's stock cache, the backend's stock table and the handheld's local queue are three views of the same number that reconcile only overnight, which is why staff trust none of them. bring them onto one model with explicit staleness, keeping the till usable while offline, and tell me what changes about the numbers staff see during the day", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.75, "slice": "boundary", "lang": "en"}
{"prompt": "interview scheduling logic exists in the API and again in the calendar sync worker, and they disagree about what happens when an interviewer declines after the candidate has confirmed. consolidate them, with the API's behaviour as the reference, and list every scheduled interview whose state would differ under the unified logic", "purpose": "refactor", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "boundary", "lang": "en"}
{"prompt": "before i approve the offline limits work, i'd like to understand what the till currently does when a card is declined for an offline voucher taken three days ago — whether the customer is contacted, whether the store absorbs it, and whether anything in the product records that it happened at all", "purpose": "review", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "boundary", "lang": "en"}
{"prompt": "HRIS pusher should reconcile nightly rather than leaving an offer accepted on our side and absent from the customer's system with nothing to indicate the divergence. build the reconciliation against their employee list endpoint, handle their nightly maintenance window, and make a persistent divergence visible to the recruiter rather than only to us", "purpose": "backendImpl", "secondary": null, "mixed": false, "difficulty": 0.7, "slice": "boundary", "lang": "en"}