userbench/UserBench-train400
UserBench train400 twin: the same 620 held tasks plus 400 leak-safe prior turns per developer. Composer 2.5 multi-label acts score by set Jaccard/IoU.
Published 7/21/2026 by Kevin Xiang Li
harbor run -d userbench/UserBench-train400UserBench-train400
Twin of the public UserBench eval with leak-safe prior-session context.
Same 620 held tasks / 62 developers as userbench/UserBench@v2 (history, gold, verifier unchanged). Each task additionally ships earlier sessions for that developer under /sim/train/, totaling 400 human user-turns sampled with sqrt two-stage.
What the agent sees
| Path | Role |
|---|---|
/sim/history.md |
Held conversation so far (identical to baseline) |
/sim/train/_index.json |
Index of packed prior sessions (ids, timestamps, repos, turn takes, paths) |
/sim/train/<sid>.md |
Contiguous prefixes of those sessions (same > DEVELOPER / > AGENT format) |
/sim/answer.txt |
Where the agent writes the next developer message |
The instruction is count-agnostic: it points at /sim/train/ without hardcoding “400”. The turn budget is a publish-time sampler parameter.
Leakage invariant
For each developer let T* = min(held_sessions[].ts) (manifest session timestamp).
Every packed train session satisfies:
session.ts < T*max(turn.ts) < T*
This is stricter than the cohort’s session-start train/held split alone. Sessions that wall-clock-overlap the earliest held session are never included.
QC on this package: 0 / 620 tasks violate max(turn.ts) < T*.
Sampling
- Start from the developer’s full leak-safe train pool (all sessions/turns passing the T* rule).
- Allocate exactly 400 human-turns with
sqrt_two_stage(time-stratified session pick, then contiguous-prefix waterfill). - Emit prefixes through the k-th human turn (including intervening AGENT/TOOL/SYSTEM), with the same ~200-word-per-turn truncation as eval history.
The full pools (not just the 400-turn subsample) live in the GitHub repo under train_pools/ so future Hub variants (e.g. train1000) can resample from the same T*-filtered corpus without changing the leakage rule.
Scale
| Developers | 62 |
| Tasks | 620 (10 per developer; same selection as baseline) |
| Condition | train400 |
| Train budget | 400 human-turns / developer (shared across that developer’s 10 tasks) |
Task naming
Same short IDs as baseline (Harbor org/name):
userbench/<username>__<hash>
Versioning is via Harbor tags on this package (e.g. @v2). Digests differ from the noprofile baseline; consume via the dataset ref below so baseline UserBench@v2 stays pinned to zero-train digests.
Keywords
userbench · user-simulation · coding-agents · move-prediction · train400
How to reference
| What | Ref |
|---|---|
| This twin | userbench/UserBench-train400@v2 (also latest) |
| Zero-train baseline | userbench/UserBench@v2 |
| Hub | https://hub.harborframework.com/datasets/userbench/UserBench-train400 |
| Full train pools (GitHub) | https://github.com/AlienKevin/user-simulator/tree/main/train_pools |
harbor run -d userbench/UserBench-train400@v2 -a <agent> -m <model>
Links
- Site / dataset EDA: https://userbench.vercel.app
- Baseline package README: sibling
UserBench(noprofile, no/sim/train/)
| Task |
|---|
userbench/dc_010__fa91db13 |
userbench/4thwithme__56e34cff |
userbench/thieso2__3dbff2d6 |
userbench/jdsingh122918__bb859790 |
userbench/kungfusaini__76670d24 |
userbench/FSM1__123d9f43 |
userbench/wildlily1021__85caed49 |
userbench/TheurgicDuke771__01470379 |
userbench/Soph__1a8c6a0c |
userbench/4thwithme__1e9b9622 |
userbench/jdsingh122918__d44c4e2b |
userbench/hutusi__67ea1d83 |
userbench/barbogast__4aa039d9 |
userbench/cyyeh__61ced54e |
userbench/135yshr__488a7841 |
userbench/nathanbooth-konecta__8e4c4321 |
userbench/malkoG__f0e27113 |
userbench/jhoetter__eb3780f4 |
userbench/dc_004__74515f8d |
userbench/wildlily1021__c031a215 |
userbench/gabadi__ae3008ee |
userbench/cyyeh__a430a77c |
userbench/jdsingh122918__9d260baa |
userbench/kohaku500__cc54a542 |
userbench/scottdensmore__17cdc6a5 |
userbench/dc_004__667e2977 |
userbench/TheurgicDuke771__99a7df5d |
userbench/nathanbooth-konecta__33b00343 |
userbench/robouden__f12f7493 |
userbench/henryph24__f0fc860a |
userbench/fcamblor__10653b23 |
userbench/blackgirlbytes__df26765a |
userbench/dc_000__d5528693 |
userbench/cyyeh__5f330796 |
userbench/admarble__b47ec152 |
userbench/oddessentials__b4bcc213 |
userbench/4thwithme__a2549fc8 |
userbench/admarble__dbb54fbd |
userbench/kmiki0__672300e9 |
userbench/alishakawaguchi__3dc3da3f |
userbench/MohammedMqat__81504953 |
userbench/junaid-appointy__beeefe8f |
userbench/raman325__9730850f |
userbench/oddessentials__0558d1c7 |
userbench/achildrenmile__97a0d30b |
userbench/nathanbooth-konecta__d2090df8 |
userbench/jeevanpillay__e77b5133 |
userbench/kungfusaini__9f8077cf |
userbench/robouden__e896e954 |
userbench/PJensen__409f7f83 |
userbench/melagiri__af1394ab |
userbench/achildrenmile__7dafef0e |
userbench/marcus-sa__81996502 |
userbench/ta93abe__25e7c1b9 |
userbench/lyston11__182e418e |
userbench/wildlily1021__ff1c6f83 |
userbench/dcambria__3462644a |
userbench/kmiki0__a9bb3ed2 |
userbench/FSM1__e064c641 |
userbench/ababushkin__0dafbf18 |
userbench/4thwithme__e0b170b3 |
userbench/dcambria__8b203ca9 |
userbench/FSM1__9ca02217 |
userbench/dc_004__b1a1b133 |
userbench/melagiri__e0dad522 |
userbench/hutusi__2751eb8c |
userbench/robouden__c6e9e844 |
userbench/kungfusaini__6409200a |
userbench/alishakawaguchi__095c91a6 |
userbench/KeKs0r__7d6d7ccd |
userbench/christso__7616b4d3 |
userbench/marcus-sa__60346b6a |
userbench/johyunduk__4bab3ef7 |
userbench/dc_001__4b92ecaf |
userbench/khaong__8bd56ca2 |
userbench/blackgirlbytes__64f0503c |
userbench/dc_004__ce4ab540 |
userbench/fcamblor__4c9bb47a |
userbench/blackgirlbytes__116d4a16 |
userbench/135yshr__e4b9747c |
userbench/scottdensmore__42381bd0 |
userbench/KeKs0r__638dbf35 |
userbench/jhoetter__bcc3790b |
userbench/singampalliveerendra__13c48199 |
userbench/marcus-sa__56b8c4e7 |
userbench/TheurgicDuke771__b6a77404 |
userbench/jhoetter__2da05fce |
userbench/kungfusaini__f8d51bd4 |
userbench/robouden__36ed725c |
userbench/barbogast__5c442045 |
userbench/jakobtfaber__1395ade5 |
userbench/singampalliveerendra__de5a3f57 |
userbench/KeKs0r__dbbc1fb2 |
userbench/yyovil__a5d654fb |
userbench/nosman__7c59d97b |
userbench/junaid-appointy__311a12f4 |
userbench/mvanhorn__89ae7ff7 |
userbench/malkoG__a7ddaeaf |
userbench/henryph24__c072ea86 |
userbench/dc_004__1b611cb2 |
Displaying 100 of 620 tasks