mirror of
https://github.com/hanzoai/insights.git
synced 2026-10-30 16:15:31 +00:00
zeekay
1b367e2b62
The tree was a debranded snapshot of upstream taken around 2026-02-20, and upstream has moved about five months since. This replaces it with upstream master, mechanically re-derived: bin/debrand does the rename, and the pipeline and squashed migrations are restored from the shipped tree. Not a merge. The previous import replaced the tree wholesale in a single commit, so there is no shared content history to merge against. Reproducing the rename on the new tree is what actually converges. What survives: .hanzo/workflows, the Dockerfiles and bin/docker* entrypoints, compose files, LLM.md, and insights/migrations. The squashed initial migration is load-bearing — the live database records it by name, so upstream's ~918 individual migrations cannot come back. Nine products are dropped as unreferenced. Twenty-one more that we do not run are kept anyway because the core app still imports them; unpicking those is a refactor and does not belong in a rebase. This base does not build yet, and the reasons are not in the rename: upstream now wants Python 3.13.13 against our 3.12.12 and Django 5.2 against our 4.2, and the dependency manifests still name upstream packages that we republish under our own names. Those are decisions, not renames.
91 lines
3.7 KiB
Python
91 lines
3.7 KiB
Python
import datetime as dt
|
|
from dataclasses import dataclass, replace
|
|
|
|
from products.pulse.backend.models import BriefConfig
|
|
|
|
# System-level constants — not per-config tunables. These bound the Temporal workflow and its
|
|
# activities, the LLM call, and the stale-brief reaper. They live here (not scattered across
|
|
# modules) so the workflow, activities, api, and reaper all read one source of truth.
|
|
|
|
# Caps total wall-clock across Temporal retries/re-executions. Worst-case activity budget in
|
|
# temporal/workflow.py is ~18min (gather 2 attempts x 5min + synthesize 5min + mark-failed
|
|
# 3 x 1min); 20 keeps the in-workflow failure path authoritative.
|
|
WORKFLOW_EXECUTION_TIMEOUT = dt.timedelta(minutes=20)
|
|
|
|
GATHER_ACTIVITY_TIMEOUT = dt.timedelta(minutes=5)
|
|
GATHER_MAX_ATTEMPTS = 2
|
|
SYNTHESIZE_ACTIVITY_TIMEOUT = dt.timedelta(minutes=5)
|
|
# A failed synthesis is not retried: retrying double-spends LLM calls.
|
|
SYNTHESIZE_MAX_ATTEMPTS = 1
|
|
MARK_FAILED_ACTIVITY_TIMEOUT = dt.timedelta(minutes=1)
|
|
MARK_FAILED_MAX_ATTEMPTS = 3
|
|
|
|
# 5-min headroom above the 20-min WORKFLOW_EXECUTION_TIMEOUT so a brief still legitimately running
|
|
# is never reaped early, while a truly-stranded row is reconciled promptly.
|
|
STALE_AFTER = dt.timedelta(minutes=25)
|
|
# The reaper sweeps at most this many stranded rows per run; a later run drains the rest.
|
|
REAP_BATCH_CAP = 500
|
|
|
|
SYNTHESIS_MODEL = "gpt-4.1"
|
|
LLM_TIMEOUT_SECONDS = 120
|
|
# Worst case 2 attempts x 120s stays under the synthesize activity timeout.
|
|
LLM_MAX_RETRIES = 1
|
|
|
|
# Cap on items passed into synthesis — keeps the activity payload well under Temporal's ~2 MiB cap.
|
|
MAX_ITEMS = 50
|
|
|
|
# Per-source gather caps — safety bounds on query/payload size.
|
|
MAX_ITEMS_PER_DETECTOR = 10
|
|
# Mirrors insights.caching.insight_cache.MAX_ATTEMPTS (not imported — that module pulls the celery
|
|
# task graph onto the pulse import path). The point at which the cache updater has given up.
|
|
STUCK_REFRESH_ATTEMPTS = 3
|
|
|
|
|
|
def _clamp(value: float, lo: float, hi: float) -> float:
|
|
return max(lo, min(hi, value))
|
|
|
|
|
|
# (key, min, max, cast) — the allowed override knobs and their safe ranges. Kept in sync with
|
|
# BriefSettings fields and the API's BriefSettingsSerializer.
|
|
_RANGES: list[tuple[str, float, float, type]] = [
|
|
("min_abs_change_pct", 1.0, 1000.0, float),
|
|
("min_baseline_value", 0.0, 1_000_000.0, float),
|
|
("max_anchor_insights", 1, 100, int),
|
|
("fallback_dashboard_count", 1, 20, int),
|
|
("confidence_threshold", 0.0, 1.0, float),
|
|
("max_opportunities", 1, 20, int),
|
|
("max_annotations", 1, 100, int),
|
|
]
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class BriefSettings:
|
|
"""Per-config tunables for gathering and synthesis. Defaults in DEFAULT_BRIEF_SETTINGS;
|
|
a BriefConfig can override any subset via its `settings` JSON, clamped to safe ranges."""
|
|
|
|
min_abs_change_pct: float = 20.0
|
|
min_baseline_value: float = 10.0
|
|
max_anchor_insights: int = 10
|
|
fallback_dashboard_count: int = 3
|
|
confidence_threshold: float = 0.6
|
|
max_opportunities: int = 3
|
|
max_annotations: int = 20
|
|
|
|
@classmethod
|
|
def from_config(cls, config: BriefConfig | None) -> "BriefSettings":
|
|
"""Merge a config's `settings` overrides over the defaults, clamping each to its range.
|
|
Unknown keys and out-of-range values are ignored/clamped, never raised — gathering must
|
|
not fail on a bad stored setting."""
|
|
settings = (config.settings if config else None) or {}
|
|
overrides: dict[str, float | int] = {}
|
|
for key, lo, hi, cast in _RANGES:
|
|
if key in settings:
|
|
try:
|
|
overrides[key] = _clamp(cast(settings[key]), lo, hi)
|
|
except (TypeError, ValueError):
|
|
continue
|
|
return replace(DEFAULT_BRIEF_SETTINGS, **overrides) # type: ignore[arg-type]
|
|
|
|
|
|
DEFAULT_BRIEF_SETTINGS = BriefSettings()
|