Datasets:
family stringclasses 5
values | files unknown | level int64 1 3 | seed int64 7,232B 250,382B | split stringclasses 2
values | task_id stringlengths 8 18 |
|---|---|---|---|---|---|
finance | {
"adjustments.json": [
{
"cents": -3300,
"id": "tx1",
"revision": 2,
"status": "settled"
}
],
"instructions.txt": "Reconcile settled transactions through day 24, inclusive. Apply the highest revision per id before filtering. Report net_cents, included_count, and negative_ids (sort... | 1 | 145,297,037,216,480 | train | finance-1 |
finance | {
"adjustments.json": [
{
"cents": -19300,
"id": "tx1",
"revision": 2,
"status": "settled"
},
{
"cents": 10600,
"id": "tx2",
"revision": 3,
"status": "settled"
}
],
"instructions.txt": "Reconcile settled transactions through day 24, inclusive. Apply ... | 2 | 143,854,085,794,473 | train | finance-2 |
finance | {
"adjustments.json": [
{
"cents": -7300,
"id": "tx1",
"revision": 2,
"status": "settled"
},
{
"cents": 12100,
"id": "tx2",
"revision": 3,
"status": "settled"
}
],
"instructions.txt": "Reconcile settled transactions through day 24, inclusive. Apply t... | 3 | 32,486,555,677,425 | train | finance-3 |
industrial | {
"instructions.txt": "Use calibrated temperature raw_c + asset offset; omit excluded ids and rows after the cutoff. A calibrated temperature strictly above 90 C is an alarm. Report alarm_count, alarm_assets (sorted unique), and max_calibrated_c across admitted readings. These are advisory synthetic diagnostics.",
... | 1 | 7,232,343,818,089 | train | industrial-1 |
industrial | {
"instructions.txt": "Use calibrated temperature raw_c + asset offset; omit excluded ids and rows after the cutoff. A calibrated temperature strictly above 90 C is an alarm. Report alarm_count, alarm_assets (sorted unique), and max_calibrated_c across admitted readings. These are advisory synthetic diagnostics.",
... | 2 | 114,065,852,893,528 | train | industrial-2 |
industrial | {
"instructions.txt": "Use calibrated temperature raw_c + asset offset; omit excluded ids and rows after the cutoff. A calibrated temperature strictly above 90 C is an alarm. Report alarm_count, alarm_assets (sorted unique), and max_calibrated_c across admitted readings. These are advisory synthetic diagnostics.",
... | 3 | 217,569,913,047,388 | train | industrial-3 |
science | {
"instructions.txt": "Apply the replacement record and exclusions, convert mg to g (1000 mg = 1 g), then compute treated mean minus control mean. Report effect_g as a reduced rational string like '3/2' (integers like '3'), treated_n, and control_n. No causal claim is warranted.",
"lab_corrections.json": {
"exc... | 1 | 80,224,036,439,190 | train | science-1 |
science | {
"instructions.txt": "Apply the replacement record and exclusions, convert mg to g (1000 mg = 1 g), then compute treated mean minus control mean. Report effect_g as a reduced rational string like '3/2' (integers like '3'), treated_n, and control_n. No causal claim is warranted.",
"lab_corrections.json": {
"exc... | 2 | 38,111,229,301,180 | train | science-2 |
science | {
"instructions.txt": "Apply the replacement record and exclusions, convert mg to g (1000 mg = 1 g), then compute treated mean minus control mean. Report effect_g as a reduced rational string like '3/2' (integers like '3'), treated_n, and control_n. No causal claim is warranted.",
"lab_corrections.json": {
"exc... | 3 | 167,870,009,124,502 | train | science-3 |
office | {
"bookings.json": [
{
"capacity": 4,
"room": "beta",
"start": 690
},
{
"capacity": 5,
"room": "gamma",
"start": 570
},
{
"capacity": 5,
"room": "gamma",
"start": 540
},
{
"capacity": 4,
"room": "beta",
"start": 600
... | 1 | 197,387,320,805,245 | train | office-1 |
office | {
"bookings.json": [
{
"capacity": 5,
"room": "gamma",
"start": 630
},
{
"capacity": 4,
"room": "beta",
"start": 600
},
{
"capacity": 5,
"room": "gamma",
"start": 600
},
{
"capacity": 4,
"room": "beta",
"start": 630
... | 2 | 230,840,665,770,440 | train | office-2 |
office | {
"bookings.json": [
{
"capacity": 5,
"room": "gamma",
"start": 630
},
{
"capacity": 4,
"room": "beta",
"start": 690
},
{
"capacity": 5,
"room": "gamma",
"start": 600
},
{
"capacity": 4,
"room": "beta",
"start": 570
... | 3 | 177,019,299,679,192 | train | office-3 |
security | {
"grants.json": [
{
"expires": 20,
"id": "grant0",
"principal": "user0",
"resource": "reports",
"role": "writer"
},
{
"expires": null,
"id": "grant1",
"principal": "user1",
"resource": "*",
"role": "admin"
},
{
"expires": null,
... | 1 | 42,070,142,277,930 | train | security-1 |
security | {
"grants.json": [
{
"expires": 20,
"id": "grant0",
"principal": "user0",
"resource": "*",
"role": "writer"
},
{
"expires": null,
"id": "grant1",
"principal": "user1",
"resource": "*",
"role": "admin"
},
{
"expires": 20,
"id":... | 2 | 60,733,045,152,695 | train | security-2 |
security | {
"grants.json": [
{
"expires": 20,
"id": "grant0",
"principal": "user0",
"resource": "payroll",
"role": "reader"
},
{
"expires": null,
"id": "grant1",
"principal": "user1",
"resource": "*",
"role": "admin"
},
{
"expires": 20,
... | 3 | 112,063,445,165,495 | train | security-3 |
finance | {
"adjustments.json": [
{
"cents": -5300,
"id": "tx1",
"revision": 2,
"status": "settled"
},
{
"cents": 6000,
"id": "tx2",
"revision": 3,
"status": "settled"
}
],
"instructions.txt": "Reconcile settled transactions through day 24, inclusive. Apply th... | 3 | 74,441,072,744,117 | test | finance-holdout |
industrial | {
"instructions.txt": "Use calibrated temperature raw_c + asset offset; omit excluded ids and rows after the cutoff. A calibrated temperature strictly above 90 C is an alarm. Report alarm_count, alarm_assets (sorted unique), and max_calibrated_c across admitted readings. These are advisory synthetic diagnostics.",
... | 3 | 183,983,339,483,383 | test | industrial-holdout |
science | {
"instructions.txt": "Apply the replacement record and exclusions, convert mg to g (1000 mg = 1 g), then compute treated mean minus control mean. Report effect_g as a reduced rational string like '3/2' (integers like '3'), treated_n, and control_n. No causal claim is warranted.",
"lab_corrections.json": {
"exc... | 3 | 48,289,269,019,604 | test | science-holdout |
office | {
"bookings.json": [
{
"capacity": 3,
"room": "alpha",
"start": 600
},
{
"capacity": 4,
"room": "beta",
"start": 660
},
{
"capacity": 4,
"room": "beta",
"start": 630
},
{
"capacity": 3,
"room": "alpha",
"start": 630
... | 3 | 114,906,564,766,764 | test | office-holdout |
security | {
"grants.json": [
{
"expires": null,
"id": "grant0",
"principal": "user0",
"resource": "payroll",
"role": "admin"
},
{
"expires": null,
"id": "grant1",
"principal": "user1",
"resource": "*",
"role": "admin"
},
{
"expires": null,
... | 3 | 250,381,909,685,060 | test | security-holdout |
Evidence Triage Lab
Twenty entirely synthetic cases: fifteen train and five structural holdout task ids across finance, industrial diagnostics, science, office scheduling, and security auditing. Each row contains the canonical seed and all model-visible files, including instructions and corrections. No real user data, credentials, or answer labels are included.
The OpenEnv server provides read-only virtual files and a bounded exact rational calculator. The agent submits a report with three fields specified by the task. Reward is one-third per strictly typed correct field; a fully correct report earns one. Finish and action-budget exhaustion earn zero. Answers are recomputed deterministically from the visible records and revisions. This is an environment dataset, not evidence of training or private-task improvement.
Source and reproducible tests: https://github.com/ricable/openenv-evidence-triage-lab
Image: ghcr.io/ricable/openenv-evidence-triage-lab:v1 (linux/amd64). Arena SDK revision: 86a180ede21e044f7929b9a7783ad83aa67d83a3.
Canonical task ids are deterministic so multiple rollouts see comparable evidence; optional reset seeds generate additional independent cases. The holdout rows are excluded from the Arena training submission. Scientific differences are descriptive, and industrial/security outputs are advisory only.
- Downloads last month
- 83