evals
Create your eval

02 — Run · Round 1 · Detroit 1972: crime · gpt56terra-detroit-s1

GPT-5.6 Terra plays Detroit 1972

Detroit's crime is out of control in 1972. Bring average crime under 60 in 10 years, without emptying the city. Ended Jan 1982: lost, city score 597.

GOAL MISSED seed 1
GPT-5.6 Terra
openai/gpt-5.6-terra

LOST: Average crime 33 (goal < 60), population 6,120 (floor 25,000); engine verdict: won

Population (start → end)76,160 → 6,120 Peak population76,160City score (start → end)285 → 597 Funds at the end$23Crime / pollution / traffic33 / 48 / 0 Zones R / C / I142 / 64 / 85 Power plants (coal / nuclear)3 / 1Police / fire stations8 / 2 Turns (model / failed)40 (40 / 0) Actions applied / rejected189 / 13 Mean turn latency14.4sModel cost$1.065Wall clock9.8 min Jetty run ID175f2d53

The city over time

Population

People, after every turn

Funds

Treasury, after every turn

City score

0–1000, re-evaluated each January

Table view
TurnDatePopulationFundsCity scoreCrimePollution
0Jan 197276,160$20,0002859265
1Apr 197273,280$15,4225309265
2Jul 197267,960$14,2985307364
3Oct 197259,800$13,2785307062
4Jan 197355,860$10,0084277062
5Apr 197348,640$9,9154276359
6Jul 197343,960$8,7824275256
7Oct 197338,220$8,1974275256
8Jan 197433,980$7,4363504456
9Apr 197429,500$7,4363504149
10Jul 197426,960$7,4363504150
11Oct 197422,940$7,2803503650
12Jan 197520,360$6,0493263346
13Apr 197517,980$5,9483263343
14Jul 197516,420$5,8733265243
15Oct 197514,980$5,8573264844
16Jan 197614,420$4,3323464846
17Apr 197613,680$4,3323464546
18Jul 197612,860$3,9983464246
19Oct 197612,280$9883464245
20Jan 197711,520$3033833845
21Apr 197710,580$3033833745
22Jul 19779,500$3033833746
23Oct 19778,860$3033833643
24Jan 19788,560$1974203543
25Apr 19787,840$1974203544
26Jul 19787,700$1974203545
27Oct 19787,420$1974203444
28Jan 19796,380$1974013444
29Apr 19796,080$1974013448
30Jul 19795,620$1974013245
31Oct 19795,480$1974013245
32Jan 19806,220$1975153246
33Apr 19808,680$895153247
34Jul 19808,840$895153248
35Oct 19807,960$895153548
36Jan 19817,180$7615083445
37Apr 19816,520$7615083444
38Jul 19815,460$7615083344
39Oct 19815,320$235083346
40Jan 19826,960$235973348

Final map

Final city map: GPT-5.6 Terra on Detroit 1972: crime

Evaluation

failchecks 9/10 verdict: valid run (code checks and checklist) and task goal met (judge)

StatusKindCheckDetail
passstepenvironmentEnvironment check (snapshot, engine, key)
passsteplaunchLaunch run-city as a detached process
passstepsuperviseSupervise in the foreground until run-city exits
passsteprenderTimelapse video rendered
passstepnotesAppend Notes to summary.md
passcode_checkoutputs-existall present
passcode_checksim-reached-deadlineended Jan 1982 after 40 turns; scenario verdict True
passcode_checkdecisions-logged40 rows for 40 turns
passcode_checkllm-turns-valid0/40 model turns failed
passcode_checkvideo-decodesduration 36.9s
passcode_checklimits-respected9.8 wall-minutes (model budget 45.0 min)
failjudgescenario-goalLOST: Average crime 33 (goal < 60), population 6,120 (floor 25,000); engine verdict: won (score 0 / threshold 1)
passchecklistnotes-explain-outcomesummary.md Notes section cites turn 1 police strategy, turn 20 power crisis pivot, turns 28-30 bankruptcy, and final turn concession on population target
passchecklistplayer-matches-requestresult.json shows player=gpt56terra model=openai/gpt-5.6-terra task=detroit seed=1 matching the runbook parameters
passchecklistno-silent-outagerun.log shows 40 consecutive successful turns with 0 LLM failures; no gaps or error stretches observed

Decision timeline

What the mayor thought and did each turn. Click a turn to jump the timelapse there.

Ready to stop guessing if your outputs are good?

Create your eval Book a 15-minute walkthrough
Connect your agent to Jetty →