userbench/UserBench

UserBench eval: 620 next-message tasks across 62 developers. Composer 2.5 classifies gold and predicted messages into multi-label acts; reward is set Jaccard/IoU.

Published 7/21/2026 by

harbor run -d userbench/UserBench

UserBench

UserBench is a Harbor eval for how well an agent can stand in for a real software engineer mid coding-agent session.

At each held-out turn, the agent reads the conversation so far (/sim/history.md) and writes the single next developer message to /sim/answer.txt. A judge labels that message into a 4-way move taxonomy; reward is 1.0 iff the predicted move matches the gold move (move-match), else 0.0.

This package is the public noprofile cut: tasks ship history only. Developer style profiles are optional Harbor skills at job time (--skill / agents[].skills), not part of the task instruction.

Scale (this revision)

Developers 62 (everyone with ≥10 eval points; 6 with <10 dropped)
Tasks 620 (exactly 10 per developer)
Condition noprofile

Task selection rule

Same rule as build_agentic.py --per-dev N:

  1. Start from the held-out eval points in the clean UserBench v2 cohort.
  2. Keep developers with ≥10 points; drop the rest.
  3. For each remaining developer, sort points by environment/history.md byte size ascending (equivalent to sorting by len(context) in the builder) and take the first 10.
  4. Sessions may be shared across selected points for a developer. Ties break by legacy task directory name ascending.

Task naming

Harbor requires org/name package refs. Tasks are published as:

userbench/<username>__<hash>

Examples:

  • GitHub: userbench/winksaville__aad4a4f9
  • DataClaw (anonymized): userbench/dc_004__05fe03e7

Versioning is via Harbor tags only (e.g. @v2), not embedded in the task name.

Keywords

userbench · user-simulation · coding-agents · move-prediction · noprofile

Taxonomy / judging

Moves are one of: approve · critical · directive · inquiry (fault-first decision rule).

The task verifier classifies the agent’s predicted message (and gold real when gold_move is null) with a configurable judge (SIMBENCH_JUDGE, default Gemini). Published agentic numbers on userbench.vercel.app/results used Composer 2.5 as the job-time judge on a 226-point / 10-developer subset of the older package — those runs are not full-620 results.

Train / held-out split

Each developer has a deep train history and a strictly later held-out set. Eval tasks are prediction points carved from held-out sessions (noprofile: history for the held conversation only).

How to reference

What Ref
Dataset userbench/UserBench@v2 (also latest)
Task userbench/<username>__<hash>@v2
Hub https://hub.harborframework.com/datasets/userbench/UserBench
harbor run -d userbench/UserBench@v2 -a <agent> -m <model>
harbor run -p userbench/winksaville__aad4a4f9@v2 -a <agent> -m <model>

Links

Task
userbench/kohaku500__31ef4143
userbench/KeKs0r__90c80d6d
userbench/PJensen__d78917b2
userbench/mvanhorn__a4b76ce8
userbench/marcus-sa__288f068e
userbench/4thwithme__404e13d3
userbench/jeevanpillay__2d2d580e
userbench/admarble__b47ec152
userbench/singampalliveerendra__bfe1ab76
userbench/christso__d8ca1042
userbench/blackgirlbytes__eb3010a3
userbench/PJensen__1a77191b
userbench/dcambria__311fbdc2
userbench/mvanhorn__c15f2981
userbench/thieso2__491f5eda
userbench/kohaku500__829d75fd
userbench/robouden__158fa32c
userbench/manderson240__5c697026
userbench/robouden__cfbea6a0
userbench/jskswamy__81abacf0
userbench/dc_004__74515f8d
userbench/achildrenmile__71a8a165
userbench/jobinlawrance__8d3fd0f5
userbench/admarble__bacf1d41
userbench/melagiri__af1394ab
userbench/singampalliveerendra__91bc17ba
userbench/ta93abe__3b70eafa
userbench/junaid-appointy__cac807fd
userbench/heddendorp__8a98faf8
userbench/kungfusaini__9f8077cf
userbench/ababushkin__7f121b4c
userbench/nosman__458811d2
userbench/jhoetter__2da05fce
userbench/jhoetter__eb3780f4
userbench/alishakawaguchi__e546a69e
userbench/jdsingh122918__44679c06
userbench/ET-NoahDolev__cf912278
userbench/fcamblor__962f5e71
userbench/135yshr__446e7a75
userbench/FSM1__d3bad63b
userbench/TheurgicDuke771__dd18f7dc
userbench/oddessentials__90f1206d
userbench/dcambria__3462644a
userbench/dcambria__cff3e93f
userbench/135yshr__e4b9747c
userbench/hutusi__67ea1d83
userbench/hutusi__d53dbb64
userbench/kungfusaini__0dcc9805
userbench/admarble__dbb54fbd
userbench/manderson240__c62b91cf
userbench/dc_010__b9fc11b2
userbench/armelhbobdad__8d99ddd6
userbench/johyunduk__b15e8506
userbench/KeKs0r__ac28eed7
userbench/jobinlawrance__ba8c72c7
userbench/scottdensmore__6d44807a
userbench/dc_001__536cd200
userbench/TheurgicDuke771__99a7df5d
userbench/nathanbooth-konecta__30525504
userbench/yutakobayashidev__52dbe4cd
userbench/armelhbobdad__2856056c
userbench/jakobtfaber__fbdba2d7
userbench/wildlily1021__2e8bcca3
userbench/hutusi__aa2cfed0
userbench/fcamblor__d55d0502
userbench/cyyeh__8615d833
userbench/ababushkin__55f888fa
userbench/gabadi__de9f94c6
userbench/heddendorp__8c0f454e
userbench/kmiki0__a9bb3ed2
userbench/lyston11__5fac2a4d
userbench/Soph__ec880138
userbench/ta93abe__25e7c1b9
userbench/ta93abe__8cd6cb7b
userbench/MohammedMqat__09e14a28
userbench/johyunduk__75a332c5
userbench/jskswamy__264ee35c
userbench/kmiki0__da54e58e
userbench/junaid-appointy__915b9090
userbench/oddessentials__0558d1c7
userbench/ababushkin__3c505e52
userbench/fcamblor__4c9bb47a
userbench/mvanhorn__8c97d4de
userbench/achildrenmile__5d92c930
userbench/MohammedMqat__81504953
userbench/dc_000__e5a61347
userbench/wildlily1021__e4c4eb11
userbench/manderson240__4d6f4d55
userbench/kungfusaini__e97ae043
userbench/nathanbooth-konecta__8fa7119c
userbench/dc_004__667e2977
userbench/lyston11__c3e8ff4c
userbench/jdsingh122918__d44c4e2b
userbench/alishakawaguchi__fcb29da8
userbench/jeevanpillay__335616a6
userbench/jdsingh122918__170092ed
userbench/FSM1__271c699b
userbench/nosman__d7d30e52
userbench/blackgirlbytes__116d4a16
userbench/alishakawaguchi__04c32a20

Displaying 100 of 620 tasks