claudecodex

Explore

Every result on the site. Arena entries open the live side-by-side builds; bench entries open the graded trial with its diffs.

Prompts
11
Trials
33
Models
11
Results
44
44 / 44 results
Arena · live build

Beam Deflection Studio

Build an inspectable beam deflection studio with load cases, reactions, shear, moment, and deflection views.

Opus 4.8 · Fable 5 · Opus 5 · Codex 5.5 · Codex 5.6 Sol · Codex 5.6 Terra · Pro tier7 runs
Arena · live build

Building Drift Dashboard

Create a building drift dashboard for comparing story drift, demand ratios, and limit-state status across floors.

Opus 4.8 · Fable 5 · Opus 5 · Codex 5.6 Sol · Codex 5.6 Terra · Pro tier6 runs
Arena · live build

Response Spectrum Explorer

Create a response-spectrum explorer that compares seismic inputs, damping assumptions, and structural response views.

Opus 4.8 · Fable 5 · Opus 5 · Codex 5.5 · Codex 5.6 Sol · Codex 5.6 Terra · Pro tier7 runs
Arena · live build

Structural Vibration Lab

Build a structural vibration lab for inspecting modal behavior, forcing input, and response over time.

Opus 4.8 · Fable 5 · Opus 5 · Codex 5.6 Sol · Codex 5.6 Terra · Pro tier6 runs
Arena · live build

Rotating Machinery Vibration Analyzer

Build a rotating machinery vibration analyzer with speed, spectrum, fault signatures, and diagnostic cues.

Opus 4.8 · Fable 5 · Opus 5 · Codex 5.5 · Codex 5.6 Sol · Codex 5.6 Terra · Pro tier7 runs
Arena · live build

Executive Analytics

Build an executive analytics dashboard with KPI scorecards, trends, operating signals, and decision-ready summaries.

Fable 5 · Opus 5 · Codex 5.5 · Codex 5.6 Sol · Codex 5.6 Terra · Pro tier6 runs
Arena · live build

Fatigue Life Estimator

Create a fatigue life estimator that presents stress ranges, S-N assumptions, damage accumulation, and life projections.

Opus 4.8 · Fable 5 · Opus 5 · Codex 5.5 · Codex 5.6 Sol · Codex 5.6 Terra · Pro tier7 runs
Arena · live build

Groundwater Flow Lab

Build a groundwater flow lab for inspecting aquifer parameters, hydraulic gradient, drawdown, and flow-field behavior.

Opus 4.8 · Opus 5 · Codex 5.5 · Codex 5.6 Sol · Codex 5.6 Terra · Pro tier6 runs
Arena · live build

Museum Heist

Create a playable museum heist browser game with stealth, guarded rooms, collectibles, and clear win/loss feedback.

Fable 5 · Codex 5.5 · Codex 5.6 Sol · Codex 5.6 Terra · Pro tier5 runs
Arena · live build

NovaOS Dashboard

Design a NovaOS dashboard with system widgets, live status panels, productivity controls, and polished app-shell behavior.

Opus 4.8 · Fable 5 · Codex 5.5 · Codex 5.6 Sol · Codex 5.6 Terra · Pro tier6 runs
Arena · live build

Tuned Mass Damper Playground

Build an interactive tuned mass damper playground: a swaying tower with an adjustable TMD, showing how mass ratio and tuning tame the response.

Fable 5 · Codex 5.6 Sol2 runs
Bench · graded trial

01 — Interval scheduler bug hunt

A maintenance-window scheduler with three planted regressions - a missed adjacency merge, a subtract() that applies only the first hole, and a gap finder that ignores t=0 and crashes on an empty calendar. Docstrings are the contract; 6 of 19 tests fail at the start.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.519 tests
Bench · graded trial

02 — Sliding-window rate limiter

Implement SlidingWindowLimiter from a written contract: weighted requests, per-key isolation, exact expiry boundaries, retry_after arithmetic - all against an injected clock. Every one of the 14 tests starts red.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.514 tests
Bench · graded trial

03 — Calendar recurrence expansion

An RRULE-style expand(): DAILY/WEEKLY/MONTHLY with intervals, Monday-based week buckets, BYDAY and BYMONTHDAY (-1 = last day), leap Februaries and skipped short months. No stdlib shortcut exists. 17 tests from zero.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.517 tests
Bench · graded trial

04 — Async task pool

Implement createTaskPool from a written contract: a concurrency limit, priority scheduling with FIFO ties, three distinct AbortSignal behaviours (pre-aborted, aborted-in-queue, aborted-in-flight), live size/pending counters and onIdle(). Deterministic - no timers, only settled promises.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.517 tests
Bench · graded trial

05 — Delimited-text codec repair

Classic 1-based Delphi string code with three planted regressions: a trim that eats tabs it must preserve, a quote-escape that consumes one character too few, and a dropped trailing empty field. Compiled with dcc32 and run as a native Win32 console suite.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.519 tests
Bench · graded trial

06 — Deterministic topo-sort

Kahn's algorithm with ordinal tie-breaking, plus shortest-cycle extraction with normalized reporting: start at the smallest node, follow edge direction, break length ties lexicographically. The discipline is entirely in the tie-break and normalization rules.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.516 tests
Bench · graded trial

07 — Checkout package repair

Five modules, six interacting regressions: banker's rounding where half-up is specified, a cart that duplicates lines instead of merging and leaks its internals, discounts applied in listed order without the zero floor, and tax computed before discounts. The receipt is only right when every module is.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.523 tests
Bench · graded trial

08 — Undo/redo event store

createStore with undo/redo, branching history, an eviction cap, and transactions that batch into one undo step, roll back on throw, and flatten when nested. Object.is no-ops must leave no trace, and listeners get snapshot semantics plus a reentrancy guard.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.519 tests
Bench · graded trial

09 — Log query engine

A query language with NOT > AND > OR, implicit AND, parentheses, quoted phrases with escapes, prefix wildcards, and a full FormatException lattice for malformed input. Parser discipline, not parsing tricks.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.519 tests
Bench · graded trial

10 — Pipeline rescue

The package won't even import - and the error it raises ('data source missing') is a lie, planted by an over-broad exception wrapper that swallows a circular import. Diagnose past the misleading symptom, untangle the modules, then fix two logic bugs the crash was hiding.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.515 tests
Bench · graded trial

11 — Schema validator

A mini schema engine where the product is the error report: exact paths ($.users[2].id), exact codes, exact sorting, no descent past a wrong type, bool-is-not-int, open vs closed objects. Validation is easy; reporting exactly right is the task.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.518 tests
Bench · graded trial

12 — URL router

Static, :param and *wildcard segments with static > param > wildcard precedence decided left-to-right, percent-decoding for captures only, trailing-slash tolerance, and the 404-versus-405 distinction with a sorted Allow list.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.517 tests
Bench · graded trial

13 — Fixed-width sales report

Byte-exact golden strings: three aligned columns, first-seen region grouping, subtotals and a grand total, hand-rolled thousands separators, product truncation with '..'. Every test compares the whole report string character by character.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.511 tests
Bench · graded trial

14 — Template engine

{{name}} with dotted paths, {{#each}} with scope chains and {{.}}, {{#if}}/{{else}} with a truthiness table, escaping, and a structural-error lattice - missing keys throw in substitutions but are simply false in conditionals.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.518 tests
Bench · graded trial

15 — Formula evaluator

An Int64 expression parser: precedence with repeatable unary minus, division truncating toward zero, remainder with the dividend's sign, overflow detection on literals and every operation, and distinct zero/overflow/syntax error classes.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.528 tests
Bench · graded trial

16 — Search under a probe budget

The starter is already correct: linear scans that pass every behaviour test. But the suite meters every probe - hard logarithmic call budgets, no duplicate reads, no out-of-range probes - so the task is making the code efficient without breaking a single edge case.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.520 tests
Bench · graded trial

17 — Event-time window aggregator

Hopping event-time windows driven only by pushed timestamps: watermark advancement, allowed lateness, exclusive window ends, multi-window membership, gap events that belong nowhere, ascending emission, and a flush that finalizes the stream. Every count lands in stats.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.518 tests
Bench · graded trial

18 — Binary frame codec

A framed wire format over raw bytes: big-endian lengths, a Fletcher-16 checksum, and a resynchronizing stream decoder that must skip garbage, reject false headers, and count bad checksums and truncated tails - byte-exact against golden frames.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.517 tests
Bench · graded trial

19 — Unified-diff patch applier

A strict unified-diff subset: exact hunk headers, marker and count validation, ascending-hunk enforcement, context matching that names the failing hunk, pure insertions anchored at line zero, and reverse application that un-applies a patch.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.519 tests
Bench · graded trial

20 — Version constraint resolver

Semver-lite ranges (caret pins the leftmost non-zero component) over a text registry, resolved to the highest admissible versions by a normative backtracking order - including a diamond conflict that forces the solver to abandon its first greedy choice.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.519 tests
Bench · graded trial

21 — Approval workflow repair

A five-state approval workflow with three planted regressions: an inclusive tier boundary turned exclusive, illegal transitions mislabelled DENY in the audit log, and a withdrawal hole in the Approved state. The audit transcript is compared line by line.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.520 tests
Bench · graded trial

22 — Virtual-DOM reconciler

A normative diff: text/replace/set/unset/insert/remove/move ops compared literally, keyed reconciliation with a working-copy walk, alphabetical prop ordering, sequential path semantics - plus a pure apply() that must reproduce the new tree without mutating anything.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.520 tests
Bench · graded trial

23 — CSS cascade resolver

Selector parsing (compounds, > and descendant combinators, attribute tests, comma lists), specificity as (ids, classes+attrs, types), !important, source order, and matching that must backtrack across candidate ancestors. The computed style is compared key by key.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.517 tests
Bench · graded trial

24 — Contrast toolkit repair

A WCAG contrast toolkit with three planted regressions: shorthand hex expanded by sixteens instead of duplication, a contrast ratio that forgets to put the lighter luminance on top, and AAA large-text grading that ignores the large flag. Real formulas, real constants.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.518 tests
Bench · graded trial

25 — HTML sanitizer

A whitelist sanitizer over a strict tag grammar: script and style swallowed with their content, URL scheme policy on href/src, attribute whitelists with fixed output order, entity-blind escaping, and stack-based repair of mis-nested and unclosed markup.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.521 tests
Bench · graded trial

26 — Flexbox-lite layout

Greedy line packing with gap arithmetic, integer growth distributed by largest remainder with leftmost tie-breaks, per-line heights and top-aligned stacking - every X, Y, W, H asserted exactly, including a golden two-line scenario.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.516 tests
Bench · graded trial

27 — Markdown renderer

A Markdown subset rendered byte-exactly: single-line headings, all-dash lists, fenced code with language classes and trailing newlines, a normative inline scan (code spans, bold containing italic, links with parsed labels), and HTML that never passes through unescaped.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.523 tests
Bench · graded trial

28 — Chunk-proof stream scanner

A feed()/finish() scanner whose event stream must be identical for every way of splitting the same document - the suite replays adversarial cuts (mid-escape, between CR and LF) plus thirty fixed-seed fuzz chunkings, and a normative emission law pins exactly which feed() call returns each event.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.524 tests
Bench · graded trial

29 — Report formatter refactor

The starter already works: a legacy formatReport whose bytes are locked by regression tests that pass on day one. Grow it into a composable createFormatter/with() API with module-level defaults the legacy entry point must honor too - strict where new, lenient where old, identical bytes throughout.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.525 tests
Bench · graded trial

30 — Bytecode stack machine

A fifteen-opcode stack VM whose product is its execution trace: step numbers, instruction pointers, post-instruction stacks, one exactly-formatted line each. Int64-checked overflow, toward-zero division, call/ret with depth caps, and eight distinct fault codes that must end the trace on the right line.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.520 tests
Bench · graded trial

31 — Three-way text merge

diff3-lite from a normative LCS: suffix DP with a pinned tie-break, two alignments walked into stable and changed regions, exact <<<<<<< ours conflict blocks. Adjacent edits must collide, separated ones must merge cleanly, identical insertions must collapse to one copy.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.520 tests
Bench · graded trial

32 — Job-shop scheduler

A deterministic scheduler graded on its event log - with the specification deliberately scattered: the model in TASK.md, tie-breaks and validation in a code comment block, the log grammar defined only by an annotated example file. Assembling the spec is the task.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.514 tests
Bench · graded trial

33 — Scoped DI container

Delegate-factory dependency injection with three lifetimes, cycle errors that spell the full path (Alpha -> Beta -> Alpha), captive-dependency detection through transient chains, and a create/dispose log whose order - dependencies first, reverse on disposal - is asserted exactly.

Fable 5 · Opus 4.8 · Sonnet 5 · 5.6 Terra · 5.6 Sol · 5.523 tests