Skip to main content
{A}

AlgoArena Blog

Research

Repairing a Problem Bank at Catalog Scale

How we judged every stored solution against every stored test across the full catalog, arbitrated whether the tests or the solutions were wrong, regenerated expected outputs by consensus, and shipped checkers for problems with many valid answers.

6 min
Three assessment agents work in one shared workspace whose output is validation evidence a reviewer can trust, not ceremony.
Assessments

Shared Agent Workspaces Need Evidence, Not Ceremony

Assessment agents got more useful when they stopped living in a separate file universe. A field note on shared workspaces, attribution, validation, and reviewer trust.

8 min
Performance versus pressure as an inverted U: grinding alone sits on the flat left, panic collapses the right, interview-real stress is the peak.
Interview Prep

Stress Is the Skill: Why 5,000 Solved Problems Won't Get You Hired

You can grind LeetCode forever and still freeze in a live interview. Performance under pressure is its own skill, and you can only train it under pressure.

4 min

Subscribe to release notes, research posts, and original field notes.

RSS feed Latest changelog
Jun 23, 2026 PracticeArena 90 and the Ghost LadderJun 23, 2026 ResearchAuditing the Arena 90 Campaign PathJun 22, 2026 AssessmentsAI-Native Assessments Need Evidence, Not TheaterJun 22, 2026 ResearchThe Five-Competency AI-Native Scoring ModelJun 21, 2026 ResearchGhost Replay Eligibility Without Publishing Raw LabelsJun 17, 2026 ResearchKeeping AI Features Reliable on Cloud CreditsJun 15, 2026 ClassroomBounded AI Practice in the ClassroomJun 6, 2026 VibecodingA Good Idea Shouldn't Have to Be in EnglishJun 5, 2026 ProductFrom Demos to SignalJun 4, 2026 Interview PrepThe Problem Was Never VibecodingMay 31, 2026 AI FluencyAI Fluency, Assessments, and RenaMay 31, 2026 ResearchBrowser-Grounded Assessment Validation

Field stories

Arena 90 and the Ghost Ladder

The Arena 90 campaign map: a ninety-node route winding through nine themed worlds.

AI-Native Assessments Need Evidence, Not Theater

A reviewer-facing assessment report with competency cards and evidence links.

Bounded AI Practice in the Classroom

A classroom coding workspace with Rena giving a hint-only nudge.

A Good Idea Shouldn't Have to Be in English

An orange on-ramp rising from a plain-English prompt into a working-code panel, with a faint road continuing to a distant destination still ahead.
View all field notes

Research and evidence

Repairing a Problem Bank at Catalog Scale

How we judged every stored solution against every stored test across the full catalog, arbitrated whether the tests or the solutions were wrong, regenerated expected outputs by consensus, and shipped checkers for problems with many valid answers.

Read

Auditing the Arena 90 Campaign Path

How the 90-node campaign was pinned to existing problems, test-count coverage, topic balance, and a stable curriculum contract.

Read

The Five-Competency AI-Native Scoring Model

How assessment scoring moved from a raw score toward public competencies for problem solving, planning, AI direction, verification, and workflow autonomy.

Read

Ghost Replay Eligibility Without Publishing Raw Labels

The shipped gates that decide whether a completed run can become a self ghost, join the human ghost pool, or enter the solution bank.

Read
View all research

Changelog

Assessments

Recruiter fit controls and Discord identity

Recruiters can review fit scores, AI-suggested re-weighting, and clearer attention cards before acting on candidate results.

Classroom

Course catalog, quizzes, and ranked stability

The course catalog now covers 30 CS courses, with database, operating systems, and data-structure quiz authoring filled out in tracked batches.

Assessments

Shared agent workspace and attribution

Assessment agents now work from the same shared workspace as editor chats, with no separate promote ceremony or private file universe.

Assessments

Verification proofs and lifecycle emails

Assessment sessions gained richer verification proof cards, walkthrough artifacts, replay evidence, and read-first review metrics.

See what changed

Build skill people can inspect.

Start freeMeet with the team