ewin-reg's picture
Real Excel test 2: bug fixes in src, new example, text
bd3cf61 verified
Raw History Blame Contribute Delete
3.78 kB
"""Draft state: the model fills the fixed text fields, checked against the material items (spec 6.2 step 7)."""
from weekly_analyst.analysis.draft import DraftFailed, DraftFields, get_draft
from weekly_analyst.analysis.plan import _item_key as plan_item_key
from weekly_analyst.core.components import DraftC, EvidenceC, ItemsC, ValidationC
from weekly_analyst.core.context import RunContext
from weekly_analyst.core.entity import Entity
from weekly_analyst.core.system import Halt, System
from weekly_analyst.core.world import World
from weekly_analyst.data.metrics_calc import compute
from weekly_analyst.systems.common import frame, metric_map, prev_period, public_items, raw_items
class DraftSystem(System):
name = "draft"
requires = (ValidationC, ItemsC, EvidenceC)
produces = (DraftC,)
def run(self, world: World, entity: Entity, ctx: RunContext) -> None:
draft = request_draft(world, entity, ctx, [])
world.add(entity, DraftC(draft=draft.model_dump(mode="json")))
def request_draft(world: World, entity: Entity, ctx: RunContext, failures: list[str]) -> DraftFields:
"""Ask the model for the draft text. Also used by the check state's redraft rounds."""
try:
draft = get_draft(ctx.client, _evidence(world, entity, ctx, failures), ctx.settings)
except DraftFailed as err:
raise Halt(f"draft: {err}") from None
wanted = sorted(plan_item_key(item) for item in public_items(world, entity))
given = sorted(f.item_key for f in draft.findings)
if wanted != given:
raise Halt("draft: findings do not match the material items")
return draft
def _direction(df, ctx: RunContext, item) -> str:
"""up, down or unchanged for the item's metric against the prior week: words only, no numbers for the model."""
metric = metric_map(ctx)[item.metric]
segment = dict(item.segment)
now = compute(df, metric, segment, item.period, ctx.metrics.date_column)
before = compute(df, metric, segment, prev_period(item.period), ctx.metrics.date_column)
if now is None or before is None:
return "unknown"
return "up" if now > before else "down" if now < before else "unchanged"
def _largest_segments(result: dict) -> list[str]:
"""Names of the segments that moved most, from an attribution result. Labels only, never numbers."""
rows = result.get("segments") or []
ranked = sorted(rows, key=lambda r: abs(r.get("delta") or 0), reverse=True)
return [str(r["segment"]) for r in ranked[:2]]
def _evidence(world: World, entity: Entity, ctx: RunContext, failures: list[str]) -> dict:
"""What the model sees: words and labels only. Every number in the report comes from code, so none goes in."""
validation = world.get(entity, ValidationC)
df = frame(world, entity, ctx)
items = [
{
"item_key": plan_item_key(public),
"metric": public.metric,
"segment": dict(public.segment),
"direction": _direction(df, ctx, raw),
}
for public, raw in zip(public_items(world, entity), raw_items(world, entity))
]
investigation = [
{
"tool": r["tool"],
"item_key": r["item_key"],
"completed": "error" not in r,
"largest_segments": _largest_segments(r.get("result", {})) if r["tool"] == "attribute_change" else [],
}
for r in world.get(entity, EvidenceC).results
]
return {
"cold_start": validation.cold_start,
"items": items,
"validation": [{"kind": f["kind"], "metric": f.get("metric")} for f in validation.findings],
"investigation": investigation,
"check_failures": failures,
}