Original samples smuggled in
A known raw-data key name (samples) anywhere in the capsule is rejected.
Rejected, exit code 1
- $.replay.samples: forbidden raw-measurement/data key
- replay.samples is not allowed by schema
1. The example capsule
Valid, exit code 0. Every value in it is synthetic. Note what is absent: no incident identifier, no device serial, no samples.
Show examples/example_capsule.json
{
"schema_version": "1.0",
"software": {
"name": "sample-monitor",
"version": "2.4.1"
},
"configuration": {
"version": "profile-r7"
},
"timestamp_policy": {
"policy": "relative_offset_seconds",
"value": 0
},
"device_channel_group": {
"device_category": "wearable",
"channel_group_category": "inertial"
},
"event_class": "threshold-transition",
"coarse_features": {
"duration_bucket": "1-2s",
"peak_level_bucket": "high",
"trend_bucket": "rising"
},
"replay": {
"generator": "synthetic-linear-drift-v1",
"seed": 417,
"recipe": {
"points": 32,
"baseline": 100.0,
"noise_stddev": 0.25,
"drift_start": 20,
"drift_per_step": 1.0
}
},
"behavior": {
"expected": "The alert remains clear while the synthetic baseline is stable.",
"observed": "The alert changes state as the synthetic drift increases."
},
"reproducibility_checklist": {
"software_version_recorded": true,
"configuration_version_recorded": true,
"synthetic_recipe_and_seed_recorded": true,
"expected_and_observed_behavior_written": true,
"independent_replay_completed": false
}
}2. Its replay, regenerated from the recipe
Generator synthetic-linear-drift-v1, seed 417, points=32, baseline=100.0, noise_stddev=0.25, drift_start=20, drift_per_step=1.0. The same recipe and
seed produce the same values on CPython 3.x. These numbers are invented by the generator; they are not a
recording of any device.
Show all 32 SYNTHETIC values
| Point | Value |
|---|---|
| 0 | 100.40791 |
| 1 | 99.8113 |
| 2 | 99.987953 |
| 3 | 99.929558 |
| 4 | 99.71914 |
| 5 | 100.238632 |
| 6 | 100.027112 |
| 7 | 100.008028 |
| 8 | 99.584359 |
| 9 | 100.100623 |
| 10 | 100.192604 |
| 11 | 100.103669 |
| 12 | 100.103232 |
| 13 | 100.003375 |
| 14 | 99.993674 |
| 15 | 100.224647 |
| 16 | 100.233417 |
| 17 | 99.842272 |
| 18 | 100.324066 |
| 19 | 100.054257 |
| 20 | 99.964421 |
| 21 | 100.884031 |
| 22 | 101.734373 |
| 23 | 102.979471 |
| 24 | 103.970244 |
| 25 | 104.861793 |
| 26 | 105.73547 |
| 27 | 107.150684 |
| 28 | 108.177859 |
| 29 | 108.903522 |
| 30 | 109.705877 |
| 31 | 111.369991 |
3. What the validator refuses
Each card below is the example capsule with one deliberate mistake. The messages are the validator's own output, captured when this page was built.
An extra identifier field
Unknown fields are rejected, so additions are visible and need a schema change.
Rejected, exit code 1
- $.incident_id is not allowed by schema
Numbers disguised as a category
Feature values that look like a list of numbers are rejected (a heuristic, not a guarantee).
Rejected, exit code 1
- coarse_features.peak_level_bucket looks like a list of numbers, not a categorical value
A timestamp more precise than its policy
A rounded-hour policy must carry a timestamp at the start of an hour.
Rejected, exit code 1
- rounded_hour_utc timestamps must be at the start of an hour
An unbounded replay recipe
Recipe values are bounded so every replay stays small and finite.
Rejected, exit code 1
- replay.recipe.points must be an integer from 1 through 10000
- replay.recipe.noise_stddev must not be negative
Limits
- It is a format guard, not a privacy scanner, anonymizer or de-identification tool. Passing validation does not make a capsule safe to publish; free text and categories can still identify people or places.
- All values on this page are SYNTHETIC. Use on real incident workflows is UNRUN.
- The checklist booleans are self-reported; the tool does not verify them.
- Replay determinism is stated for CPython 3.x only.