HKUDS/Vibe-Trading · error · TypeError
'last_run_at' must be an integer (epoch ms) or null
Error message
'last_run_at' must be an integer (epoch ms) or null
What it means
Raised by Job from_dict (agent/src/scheduled_research/models.py:444) when the optional 'last_run_at' field is present but is neither an int (epoch ms) nor null. Unlike the required timestamps, absence is fine (None); only a present wrong-typed value raises.
Source
Thrown at agent/src/scheduled_research/models.py:444
The reconstructed ``ScheduledResearchJob``.
Raises:
KeyError: If a required field is missing.
TypeError: If a field has the wrong type.
ValueError: If ``status`` is not a recognized ``JobStatus`` value.
"""
job_id = data["id"]
prompt = data["prompt"]
schedule = data["schedule"]
if not isinstance(job_id, str) or not isinstance(prompt, str) or not isinstance(schedule, str):
raise TypeError("'id', 'prompt', and 'schedule' must be strings")
next_run_at = data["next_run_at"]
created_at = data["created_at"]
if not isinstance(next_run_at, int) or not isinstance(created_at, int):
raise TypeError("'next_run_at' and 'created_at' must be integers (epoch ms)")
last_run_at = data.get("last_run_at")
if last_run_at is not None and not isinstance(last_run_at, int):
raise TypeError("'last_run_at' must be an integer (epoch ms) or null")
consecutive_failures = data.get("consecutive_failures", 0)
if (
isinstance(consecutive_failures, bool)
or not isinstance(consecutive_failures, int)
or consecutive_failures < 0
):
raise TypeError("'consecutive_failures' must be a non-negative integer")
last_error = data.get("last_error")
failure_kind = data.get("failure_kind")
if last_error is not None and not isinstance(last_error, str):
raise TypeError("'last_error' must be a string or null")
if failure_kind is not None and failure_kind not in {"dispatch", "schedule"}:
raise ValueError("'failure_kind' must be 'dispatch', 'schedule', or null")
# Never raises: the store quarantines the whole file when a single
# record fails to load, so an unusable timezone value degrades that
# one job to UTC — the semantics it had before the field existed —
# instead of taking every other job down with it. Absent, blank, and
# non-string values all normalize to None.View on GitHub (pinned to 80ffdda44c)
Solutions
- Use the same int-epoch-ms helper for all timestamp fields including last_run_at
- Convert on write: int(dt.timestamp() * 1000)
- Set it to null rather than 0 or "" when unknown
Example fix
// before job["last_run_at"] = datetime.now(timezone.utc).isoformat() // after job["last_run_at"] = int(datetime.now(timezone.utc).timestamp() * 1000)
Defensive patterns
Strategy: validation
Validate before calling
def last_run_ok(d):
v = d.get("last_run_at")
return v is None or (isinstance(v, int) and not isinstance(v, bool)) Type guard
def safe_last_run(v):
if v is None or isinstance(v, bool) or not isinstance(v, int):
return None
return v Try / catch
try:
job = Job.from_dict(record)
except TypeError as exc:
if "last_run_at" in str(exc):
record = dict(record); record["last_run_at"] = None
job = Job.from_dict(record)
else:
raise Prevention
- Null out unknown timestamps instead of writing 0 or ''
- Reuse the same timestamp helper for optional fields
- Test deserialization after the first completed run
When it happens
Trigger: {"last_run_at": "1767225600000"} or a float/ISO string in the last-run slot after a job has run at least once; absent key or explicit null is accepted.
Common situations: Same timestamp format drift as other fields but only appearing after the first run, so it surfaces late; retry writers formatting timestamps differently than the creation path.
Related errors
- 'delivery.updated_at' must be an integer (epoch ms) or null
- 'next_run_at' and 'created_at' must be integers (epoch ms)
- 'end_at' must be an integer (epoch ms) or null
- 'delivery' must be an object or null
- unknown delivery status {raw_status!r}
AI-assisted analysis of HKUDS/Vibe-Trading@80ffdda44c (2026-08-28).
Data as JSON: /api/errors/e14481cc7f65a367.
Report an issue: GitHub.