runloopable / platform /modules /amazon_sku_metrics.py
fsanyoto's picture
Deploy AIOS web (React glide grid + FastAPI slice)
8133df9 verified
Raw History Blame Contribute Delete
37.8 kB
"""Per Amazon seller SKU ads, stock, returns and listing metrics, four health labels, a score and a priority.
Every ratio is a fraction (0.15 means 15%), the scale `amazon_profit.compute_row` stores `margin` in.
`rollup` builds the ASIN and Parent ASIN grains from SKU rows. It sums first, then recomputes every
ratio from the sums. Ad spend and its trend come from the economics ad charges when economics carries
them, else from the Ads tables. `validate_metrics` ties ad spend, stock and returns to independent
mirror totals (rule 8). No source read carries a LIMIT (rule 1).
"""
from __future__ import annotations
import datetime as dt
import json
import math
# Data-layer readers of the Amazon mirror, shared on purpose so the stock and listing SQL is not forked.
from modules import ranking_diagnostic, restock_planner
#: Cut-offs and weights (owner ruling). Ratios are fractions: 0.005 is 0.5%.
BENCHMARKS = {
"ctr_strong": 0.005, # CTR at least 0.5% is Strong
"ctr_weak": 0.003, # CTR under 0.3% is Weak
"acos_strong": 0.25, # ACOS at most 25% is Strong
"acos_weak": 0.40, # ACOS over 40% is Weak
"tacos_strong": 0.10, # TACOS at most 10% is Strong
"tacos_weak": 0.20, # TACOS over 20% is Weak
"cvr_strong": 0.10, # CVR at least 10% is Strong
"cvr_weak": 0.05, # CVR under 5% is Weak
"margin_strong": 0.15, # margin at least 15% is Strong
"margin_weak": 0.05, # margin under 5% is Weak
"cover_strong_min": 30, # days of cover 30 to 90 is Strong
"cover_strong_max": 90,
"cover_weak_under": 14, # days of cover under 14 is Weak
"cover_weak_over": 180, # days of cover over 180 is Weak
"return_rate_strong": 0.03, # return rate at most 3% is Strong
"return_rate_weak": 0.08, # return rate over 8% is Weak
"listing_cvr_strong": 0.10, # unit session percentage at least 10%: mirrors the CVR ruling
"listing_cvr_weak": 0.05, # unit session percentage under 5%: mirrors the CVR ruling
"points_strong": 25, # each of the four dimensions is worth 25 points
"points_fair": 12.5, # Fair scores half points
"points_weak": 0,
"points_no_data": 12.5, # No data scores neutral, half points, never zero
"priority_high_under": 40, # score under 40 is High priority
"priority_low_over": 70, # score over 70 is Low priority; 40 to 70 inclusive is Medium
}
GRAINS = ("sku", "asin", "parent_asin")
#: The words a person reads and picks for each grain key. Stored rows may hold either form.
GRAIN_LABELS = {"sku": "SKU", "asin": "ASIN", "parent_asin": "Parent ASIN"}
_GRAIN_BY_WORD = {**{key: key for key in GRAINS},
**{label.lower(): key for key, label in GRAIN_LABELS.items()}}
def grain_key(value):
"""The grain key for a key or its label, case-insensitive; "" for anything else."""
word = " ".join(str(value or "").split()).lower()
return _GRAIN_BY_WORD.get(word, "")
HEALTH_OPTIONS = ("Strong", "Fair", "Weak", "No data")
PRIORITY_OPTIONS = ("High", "Medium", "Low")
NO_ADS_NOTE = "Connect Amazon Ads on the Connectors page"
NO_STOCK_NOTE = "No inventory planning data yet"
NO_RETURNS_NOTE = "No returns data yet"
#: Reason texts for a blank figure. The Grid cannot say why a cell is blank, so the reason is data.
NOT_ADVERTISED_NOTE = "Not in the Ads reports for the period"
NO_AD_SPEND_NOTE = "No ad spend"
NO_NET_SALES_NOTE = "No net sales in the period"
NOT_PLANNED_NOTE = "Not in the FBA planning report"
NO_RECENT_SALES_NOTE = "No sales in the last 30 days"
NO_UNITS_NOTE = "No units sold in the period"
NO_TRAFFIC_NOTE = "Not in the traffic report"
NO_SESSIONS_NOTE = "No sessions in the traffic report"
NO_BUY_BOX_NOTE = "No Buy Box share in the traffic report"
SERIES_DAYS = 30
#: Ad spend per advertised product may trail the campaign report by a few cents per day.
ADS_TIE_TOLERANCE = 0.02
#: Economics ad spend ties to its own independent sum within one cent.
SPEND_TIE_TOLERANCE = 0.01
COUNT_TIE_TOLERANCE = 0.01
ECONOMICS = "amzn_economics"
PLANNING = "amzn_inventory_planning"
FBA_RETURNS = "amzn_fba_customer_returns"
MFN_RETURNS = "amzn_mfn_returns"
AD_TYPES = ("sp", "sb", "sd")
ADS_PRODUCTS = tuple(f"amzn_ads_{kind}_advertised_products" for kind in AD_TYPES)
ADS_CAMPAIGNS = tuple(f"amzn_ads_{kind}_campaigns" for kind in AD_TYPES)
_HEALTH = {"options": list(HEALTH_OPTIONS), "format": {"display": "dot"}}
#: (key, label, type, description, extra) in display order. H1b wraps each with `amazon_profit._field`.
METRIC_COLUMNS = (
("parent_asin", "Parent ASIN", "text", "Variation family. An ASIN with no parent is its own.", {}),
("ad_impressions", "Ad impressions", "int", "Sponsored ad impressions in the period.", {}),
("ad_clicks", "Ad clicks", "int", "Sponsored ad clicks in the period.", {}),
("ad_spend", "Ad spend", "currency", "Sponsored ad cost in the period.", {}),
("ad_sales", "Ad sales", "currency", "Sales Amazon credits to the ads.", {}),
("ad_orders", "Ad orders", "int", "Orders Amazon credits to the ads.", {}),
("ctr", "CTR", "pct", "Ad clicks over ad impressions.", {}),
("cvr", "CVR", "pct", "Ad orders over ad clicks.", {}),
("cpc", "CPC", "currency", "Ad spend over ad clicks.", {}),
("acos", "ACOS", "pct", "Ad spend over ad sales.", {}),
("tacos", "TACOS", "pct", "Sponsored ad charges over net sales.", {}),
("tacos_note", "TACOS note", "text", "Why TACOS is blank.", {}),
("roas", "ROAS", "int", "Ad sales over ad spend.", {"format": {"decimals": 2}}),
("ads_note", "Ads note", "text", "Why ads figures are blank.", {}),
("fba_on_hand", "FBA on hand", "int", "Units available at Amazon in the newest snapshot.", {}),
("inbound_units", "Inbound units", "int", "Units on the way to Amazon.", {}),
("days_of_cover", "Days of cover", "int", "Days the FBA stock lasts at the 30 day sales pace.", {}),
("stock_note", "Stock note", "text", "Why stock figures are blank.", {}),
("return_units", "Returned units", "int", "Customer returns in the period.", {}),
("return_rate", "Return rate", "pct", "Returned units over units sold.", {}),
("returns_note", "Returns note", "text", "Why return figures are blank.", {}),
("sessions", "Sessions", "int",
"Product page sessions of the ASIN from the newest traffic report. Not aligned to the period.", {}),
("unit_session_pct", "Unit session percentage", "pct", "Units sold over sessions.", {}),
("buy_box_pct", "Buy Box", "pct", "Share of sessions that showed your offer.", {}),
("listing_note", "Listing note", "text", "Why traffic figures are blank.", {}),
("stock_health", "Stock health", "select", "Days of cover against the benchmark.", _HEALTH),
("profit_health", "Profit health", "select", "Margin against the benchmark.", _HEALTH),
("ads_health", "Ads health", "select", "CTR, ACOS, TACOS and CVR against the benchmarks.", _HEALTH),
("listing_health", "Listing health", "select", "Unit session percentage and return rate.", _HEALTH),
("score", "Score", "int", "Health score from 0 to 100.", {"format": {"display": "bar"}}),
("priority", "Priority", "select", "High needs attention first.",
{"options": list(PRIORITY_OPTIONS), "format": {"display": "dot"}}),
("units_30d", "Units trend", "chart", "Units sold per day, last 30 days.", {}),
("net_sales_30d", "Net sales trend", "chart", "Net sales per day, last 30 days.", {}),
("ad_spend_30d", "Ad spend trend", "chart", "Ad spend per day, last 30 days.", {}),
)
ROLLUP_COLUMNS = (
("sku_count", "SKUs", "int", "Seller SKUs in this row.", {}),
("skus", "SKU list", "text", "Seller SKUs in this row, comma separated.", {}),
)
ADDITIVE_KEYS = ("units_sold", "net_sales", "fees", "sponsored_ads", "net_proceeds", "cogs_total",
"net_profit", "ad_impressions", "ad_clicks", "ad_spend", "ad_sales", "ad_orders",
"fba_on_hand", "inbound_units", "return_units", "velocity_30d")
SERIES_KEYS = ("units_30d", "net_sales_30d", "ad_spend_30d")
_INT_KEYS = frozenset(("units_sold", "ad_impressions", "ad_clicks", "ad_orders", "fba_on_hand",
"inbound_units", "return_units"))
#: A sum over a partly unlinked group is not a profit: one costless active child blanks it.
_STRICT_KEYS = frozenset(("cogs_total", "net_profit"))
_SCORED = ("tacos", "return_rate", "stock_health", "profit_health", "ads_health", "listing_health",
"score", "priority")
#: Each reason column and the figures it explains. A blank figure needs its note set.
ADS_FIGURES = ("ad_impressions", "ad_clicks", "ad_spend", "ad_sales", "ad_orders", "ctr", "cvr", "cpc", "acos",
"roas")
NOTE_GROUPS = {
"ads_note": ADS_FIGURES,
"tacos_note": ("tacos",),
"stock_note": ("fba_on_hand", "inbound_units", "days_of_cover"),
"returns_note": ("return_units", "return_rate"),
"listing_note": ("sessions", "unit_session_pct", "buy_box_pct"),
}
NOTE_KEYS = tuple(NOTE_GROUPS)
#: What the last `rollup` call grouped and skipped. Updated in place.
last_report: dict = {}
# ---------------------------------------------------------------- pure arithmetic
def _f(value):
try:
number = float(value)
except (TypeError, ValueError):
return None
return number if math.isfinite(number) else None
def _ratio(num, den):
"""num / den, or None when either is missing or the denominator is not positive."""
num, den = _f(num), _f(den)
if num is None or den is None or den <= 0:
return None
return num / den
def _half_up(value):
return int(math.floor(value + 0.5))
def _text(value):
return str(value).strip() if value is not None and str(value).strip() else None
def band(metric, value):
"""Strong, Fair or Weak for one ratio against BENCHMARKS; None when the value is None."""
value = _f(value)
if value is None:
return None
if metric == "days_of_cover":
return cover_band(value)
name = "listing_cvr" if metric == "unit_session_pct" else metric
strong, weak = BENCHMARKS[f"{name}_strong"], BENCHMARKS[f"{name}_weak"]
value = round(value, 9)
if name in ("acos", "tacos", "return_rate"):
return "Strong" if value <= strong else "Weak" if value > weak else "Fair"
return "Strong" if value >= strong else "Weak" if value < weak else "Fair"
def cover_band(days):
"""Strong for 30 to 90 days, Weak under 14 or over 180, else Fair; None when unknown."""
days = _f(days)
if days is None:
return None
if BENCHMARKS["cover_strong_min"] <= days <= BENCHMARKS["cover_strong_max"]:
return "Strong"
if days < BENCHMARKS["cover_weak_under"] or days > BENCHMARKS["cover_weak_over"]:
return "Weak"
return "Fair"
def combine(labels):
"""Weak if any is Weak, Strong if all available are Strong, else Fair; No data when none is available."""
found = [label for label in labels if label]
if not found:
return "No data"
if "Weak" in found:
return "Weak"
return "Strong" if all(label == "Strong" for label in found) else "Fair"
def points(label):
return BENCHMARKS[{"Strong": "points_strong", "Fair": "points_fair", "Weak": "points_weak"}
.get(label, "points_no_data")]
def priority_for(score):
if score < BENCHMARKS["priority_high_under"]:
return "High"
return "Low" if score > BENCHMARKS["priority_low_over"] else "Medium"
def score_row(row: dict) -> dict:
"""Health labels, score and priority for one row that already carries the economics and metric keys."""
net_sales = _f(row.get("net_sales"))
spend = _f(row.get("ad_spend"))
if spend is None:
spend = _f(row.get("sponsored_ads"))
# R5: a SKU with no ad spend keeps TACOS blank.
tacos = _ratio(spend, net_sales) if spend and spend > 0 else None
return_rate = _ratio(row.get("return_units"), row.get("units_sold"))
stock = combine([cover_band(row.get("days_of_cover"))])
profit = combine([band("margin", row.get("margin"))]) if net_sales else "No data"
ads = combine([band("ctr", row.get("ctr")), band("acos", row.get("acos")),
band("tacos", tacos), band("cvr", row.get("cvr"))])
listing = combine([band("unit_session_pct", row.get("unit_session_pct")),
band("return_rate", return_rate)])
score = _half_up(sum(points(label) for label in (stock, profit, ads, listing)))
return {"tacos": tacos, "return_rate": return_rate, "stock_health": stock, "profit_health": profit,
"ads_health": ads, "listing_health": listing, "score": score, "priority": priority_for(score)}
def _ads_reason(row):
if row.get("ads_note") == NO_ADS_NOTE:
return NO_ADS_NOTE
if row.get("ad_impressions") is None:
return NOT_ADVERTISED_NOTE
names = [word for keys, word in ((("ctr",), "ad impressions"), (("cvr", "cpc"), "ad clicks"),
(("acos",), "ad sales"), (("roas", "ad_spend"), "ad spend"))
if any(row.get(k) is None for k in keys)]
return f"No {' or '.join(names)} in the period" if names else NOT_ADVERTISED_NOTE
def _tacos_reason(row):
spend = _f(row.get("ad_spend"))
if spend is None:
spend = _f(row.get("sponsored_ads"))
return NO_AD_SPEND_NOTE if not spend or spend <= 0 else NO_NET_SALES_NOTE
def _stock_reason(row):
if row.get("fba_on_hand") is None or row.get("inbound_units") is None:
return NOT_PLANNED_NOTE
return NO_RECENT_SALES_NOTE
def _returns_reason(row):
return NO_RETURNS_NOTE if row.get("return_units") is None else NO_UNITS_NOTE
def _listing_reason(row):
sessions = _f(row.get("sessions"))
if sessions is None:
return NO_TRAFFIC_NOTE
if row.get("unit_session_pct") is None:
return NO_SESSIONS_NOTE if sessions <= 0 else NO_UNITS_NOTE
return NO_BUY_BOX_NOTE
_REASONS = {"ads_note": _ads_reason, "tacos_note": _tacos_reason, "stock_note": _stock_reason,
"returns_note": _returns_reason, "listing_note": _listing_reason}
def blank_notes(row: dict) -> dict:
"""Each note in NOTE_GROUPS: "" when its figures are all set, else why one is blank (rule 1).
`row` carries raw figures after `score_row`. An `ads_note` of NO_ADS_NOTE means no Ads table is mirrored."""
return {note: _REASONS[note](row) if any(row.get(k) is None for k in figures) else ""
for note, figures in NOTE_GROUPS.items()}
# ---------------------------------------------------------------- mirror reads
def _columns(con, table):
return {r[0] for r in con.execute(
"SELECT column_name FROM information_schema.columns WHERE table_name = ?", [table]).fetchall()}
def _q(name):
return '"' + str(name).replace('"', '""') + '"'
def _day(column):
"""A DATE from an ISO text or timestamp prefix, or from MM/DD/YYYY."""
text = f"CAST({_q(column)} AS VARCHAR)"
return (f"COALESCE(TRY_CAST(substr({text}, 1, 10) AS DATE), "
f"CAST(try_strptime({text}, '%m/%d/%Y') AS DATE))")
def _pick(cols, *choices):
return next((c for c in choices if c in cols), None)
def _num_sql(cols, name):
return f"TRY_CAST({_q(name)} AS DOUBLE)" if name in cols else "NULL"
def _sku_sql(column):
return f"TRIM(CAST({_q(column)} AS VARCHAR))"
def _all_skus(con):
if "msku" not in _columns(con, ECONOMICS):
return []
return [r[0] for r in con.execute(
f"SELECT DISTINCT {_sku_sql('msku')} FROM {ECONOMICS} WHERE NULLIF(TRIM(CAST(msku AS VARCHAR)), '') "
"IS NOT NULL ORDER BY 1").fetchall()]
def _economics(con, start, end, skus):
"""{sku: {asin, units}}: newest child ASIN and net units sold in [start, end]."""
cols = _columns(con, ECONOMICS)
if not {"msku", "start_date"} <= cols:
return {}
day = _day("start_date")
asin = (f"arg_max(NULLIF(TRIM(CAST(child_asin AS VARCHAR)), ''), {day})"
if "child_asin" in cols else "NULL")
units = (f"SUM(TRY_CAST(sales_net_units_sold AS DOUBLE)) FILTER (WHERE {day} BETWEEN "
"CAST(? AS DATE) AND CAST(? AS DATE))" if "sales_net_units_sold" in cols else "NULL")
args = [start, end] if "sales_net_units_sold" in cols else []
rows = con.execute(f"SELECT {_sku_sql('msku')}, {asin}, {units} FROM {ECONOMICS} "
f"WHERE {_sku_sql('msku')} IN (SELECT unnest(?)) GROUP BY 1",
[*args, skus]).fetchall()
return {r[0]: {"asin": r[1], "units": r[2]} for r in rows}
def _econ_ads_columns(con):
"""The economics ad charge totals, `_charge_` detail skipped; empty when economics has none."""
from modules import amazon_profit # amazon_profit imports this module at load, so import late
cols = _columns(con, ECONOMICS)
if not {"msku", "start_date"} <= cols:
return []
return amazon_profit.ads_columns(cols)
def _econ_ads_sql(columns):
return " + ".join(f"COALESCE(TRY_CAST({_q(c)} AS DOUBLE), 0)" for c in columns)
def _econ_spend(con, start, end, skus):
"""{sku: economics ad charges} in [start, end]; None when economics carries no ad amounts."""
columns = _econ_ads_columns(con)
if not columns:
return None
rows = con.execute(f"SELECT {_sku_sql('msku')}, SUM({_econ_ads_sql(columns)}) FROM {ECONOMICS} "
f"WHERE {_day('start_date')} BETWEEN CAST(? AS DATE) AND CAST(? AS DATE) "
f"AND {_sku_sql('msku')} IN (SELECT unnest(?)) GROUP BY 1", [start, end, skus]).fetchall()
return {r[0]: r[1] for r in rows}
def _ads_parts(con):
"""SELECT sku, day, impressions, clicks, cost, sales, orders over every present ads product table."""
parts = []
for table in ADS_PRODUCTS:
cols = _columns(con, table)
if not {"advertised_sku", "date"} <= cols:
continue
metrics = ", ".join(f"{_num_sql(cols, c)} AS {c}"
for c in ("impressions", "clicks", "cost", "sales", "orders"))
parts.append(f"SELECT {_sku_sql('advertised_sku')} AS sku, {_day('date')} AS day, {metrics} "
f"FROM {table}")
return " UNION ALL ".join(parts) or None
def _ads(con, start, end, skus):
"""{sku: summed ad figures} in [start, end]; None when no ads table is mirrored."""
union = _ads_parts(con)
if union is None:
return None
rows = con.execute(
f"SELECT sku, SUM(impressions), SUM(clicks), SUM(cost), SUM(sales), SUM(orders), COUNT(*) "
f"FROM ({union}) a WHERE day BETWEEN CAST(? AS DATE) AND CAST(? AS DATE) "
"AND sku IN (SELECT unnest(?)) GROUP BY 1", [start, end, skus]).fetchall()
return {r[0]: {"impressions": r[1], "clicks": r[2], "cost": r[3], "sales": r[4], "orders": r[5]}
for r in rows if r[6]}
def _returns_parts(con):
"""SELECT sku, day, units over the FBA and MFN customer returns tables that are present."""
parts = []
for table, skus, days, qty in ((FBA_RETURNS, ("sku",), ("return_date",), ("quantity",)),
(MFN_RETURNS, ("merchant_sku", "sku"),
("return_request_date", "return_date"), ("return_quantity", "quantity"))):
cols = _columns(con, table)
sku, day, units = _pick(cols, *skus), _pick(cols, *days), _pick(cols, *qty)
if sku is None or day is None:
continue
# A row without a quantity is one returned unit.
amount = f"COALESCE(TRY_CAST({_q(units)} AS DOUBLE), 1)" if units else "1"
parts.append(f"SELECT {_sku_sql(sku)} AS sku, {_day(day)} AS day, {amount} AS units FROM {table}")
return " UNION ALL ".join(parts) or None
def _returns(con, start, end, skus):
"""{sku: returned units} in [start, end]; None when no returns table is mirrored."""
union = _returns_parts(con)
if union is None:
return None
rows = con.execute(f"SELECT sku, SUM(units) FROM ({union}) r WHERE day BETWEEN CAST(? AS DATE) "
"AND CAST(? AS DATE) AND sku IN (SELECT unnest(?)) GROUP BY 1",
[start, end, skus]).fetchall()
return {r[0]: r[1] for r in rows}
def _daily(con, first, last, skus):
"""{sku: {day: [units, net_sales, ad_spend]}} for the series window."""
out = {}
def add(rows, index):
for sku, day, value in rows:
if day is not None and value is not None:
out.setdefault(sku, {}).setdefault(day.isoformat(), [0.0, 0.0, 0.0])[index] += value
cols = _columns(con, ECONOMICS)
spend = _econ_ads_columns(con)
if {"msku", "start_date"} <= cols:
sums = [(index, f"TRY_CAST({column} AS DOUBLE)") for index, column in
((0, "sales_net_units_sold"), (1, "sales_net_product_sales_amount")) if column in cols]
if spend:
sums.append((2, _econ_ads_sql(spend)))
for index, value in sums:
add(con.execute(
f"SELECT {_sku_sql('msku')}, {_day('start_date')} AS d, SUM({value}) "
f"FROM {ECONOMICS} WHERE {_day('start_date')} BETWEEN CAST(? AS DATE) AND CAST(? AS DATE) "
f"AND {_sku_sql('msku')} IN (SELECT unnest(?)) GROUP BY 1, 2",
[first, last, skus]).fetchall(), index)
union = None if spend else _ads_parts(con)
if union is not None:
add(con.execute(f"SELECT sku, day, SUM(cost) FROM ({union}) a WHERE day BETWEEN CAST(? AS DATE) "
"AND CAST(? AS DATE) AND sku IN (SELECT unnest(?)) GROUP BY 1, 2",
[first, last, skus]).fetchall(), 2)
return out
def _lineage_read(con, catalog, column, keys, projection):
"""Read only requested lineage keys in bounded SQL pages."""
keys = sorted({str(k).strip() for k in keys if k})
if catalog is None:
return
for start in range(0, len(keys), 300):
part = keys[start:start + 300]
source = "c" if column == "sku" else "p"
sql = catalog + f"SELECT {projection} FROM {source} WHERE {column} IN (" + ",".join("?" for _ in part) + ")"
offset = 0
while True:
rows = con.execute(sql + f" ORDER BY 1 LIMIT 500 OFFSET {offset}", part).fetchall()
yield from rows
if len(rows) < 500:
break
offset += len(rows)
def _parent_map(con, asins):
"""Use the project popup's pricing lineage for the requested ASINs."""
from core import amazon_projects as AP
return dict(_lineage_read(con, AP._catalog(con), "child", asins, "child, parent"))
def parent_children(con, parent):
"""Read the same pricing family used by persisted competitor cells."""
from core import amazon_projects as AP
return [child for child, _ in _lineage_read(con, AP._catalog(con), "parent", [parent], "child, parent")
if child and child != parent]
def lineage_rows(con, tables, *, children=False):
"""Project stored row identities onto the popup's current listings and pricing lineage."""
from core import amazon_projects as AP
catalog = AP._catalog(con)
out = {g: {rid: dict(row) for rid, row in rows.items()} for g, rows in tables.items()}
listed = {sku: (asin, parent) for sku, asin, parent in _lineage_read(
con, catalog, "sku", [r.get("sku") for r in out.get("sku", {}).values()], "sku, asin, parent")}
for row in out.get("sku", {}).values():
row["asin"], row["parent_asin"] = listed.get(str(row.get("sku") or "").strip(), (None, None))
asins = [str(r.get("asin") or "").strip().upper() for r in out.get("asin", {}).values()]
parents = dict(_lineage_read(con, catalog, "child", asins, "child, parent"))
for row in out.get("asin", {}).values():
row["asin"] = str(row.get("asin") or "").strip().upper() or None
row["parent_asin"] = parents.get(row["asin"])
if children:
wanted = [str(r.get("parent_asin") or "").strip().upper()
for r in out.get("parent_asin", {}).values()]
families = {parent: [] for parent in wanted}
for child, parent in _lineage_read(con, catalog, "parent", wanted, "child, parent"):
if child and child != parent:
families[parent].append(child)
# Rival popups use the parent itself when pricing names no children.
out["asin"] = {rid: dict(row, parent_asin=None) for rid, row in out.get("asin", {}).items()}
for parent, kids in families.items():
for child in kids or [parent]:
out["asin"][("lineage", parent, child)] = {"asin": child, "parent_asin": parent}
return out
def _out(value):
value = round(float(value), 2)
return int(value) if value == int(value) else value
def series_cell(values):
"""The stored chart value: a JSON list of numbers, oldest first."""
return json.dumps([_out(v) for v in values])
def _series(value):
if isinstance(value, str):
try:
value = json.loads(value)
except ValueError:
return None
if not isinstance(value, (list, tuple)):
return None
return [_f(v) or 0.0 for v in value]
def _int(value):
value = _f(value)
return None if value is None else _half_up(value)
def sku_metrics(con, *, start, end, skus, today) -> dict[str, dict]:
"""Per SKU metric cells for the period [start, end] (ISO dates), plus internal keys.
`skus` None reads every SKU in economics. Series cover the 30 days before `today`.
"""
wanted = [s for s in dict.fromkeys(_text(s) for s in (_all_skus(con) if skus is None else skus)) if s]
if not wanted:
return {}
if isinstance(today, str):
today = dt.date.fromisoformat(today[:10])
days = [(today - dt.timedelta(days=SERIES_DAYS - i)).isoformat() for i in range(SERIES_DAYS)]
econ = _economics(con, start, end, wanted)
planning = restock_planner._planning(con, wanted) or {}
pricing = ranking_diagnostic._pricing(con) or {}
parents = _parent_map(con, [r.get("asin") for r in econ.values()]
+ [r.get("asin") for r in planning.values()])
ads = _ads(con, start, end, wanted)
econ_spend = _econ_spend(con, start, end, wanted)
returns = _returns(con, start, end, wanted)
daily = _daily(con, days[0], days[-1], wanted)
out = {}
for sku in wanted:
e, p = econ.get(sku, {}), planning.get(sku)
asin = _text(e.get("asin")) or _text((p or {}).get("asin"))
row = {"asin": asin, "parent_asin": (parents.get(asin) or asin) if asin else None}
a = (ads or {}).get(sku)
impressions, clicks = (_f(a["impressions"]), _f(a["clicks"])) if a else (None, None)
spend, sales, orders = (_f(a["cost"]), _f(a["sales"]), _f(a["orders"])) if a else (None, None, None)
if econ_spend is not None:
spend = _f(econ_spend.get(sku)) or 0.0
row.update(ad_impressions=_int(impressions), ad_clicks=_int(clicks), ad_spend=spend, ad_sales=sales,
ad_orders=_int(orders), ctr=_ratio(clicks, impressions), cvr=_ratio(orders, clicks),
cpc=_ratio(spend, clicks), acos=_ratio(spend, sales), roas=_ratio(sales, spend),
ads_note="" if a else NO_ADS_NOTE if ads is None else NOT_ADVERTISED_NOTE)
available = _f(p.get("available")) if p else None
velocity = _f(p.get("units")) if p else None
cover = _f(p.get("days")) if p else None
if cover is None and available is not None:
cover = _ratio(available * 30, velocity)
row["fnsku"] = _text(p.get("fnsku")) if p else None
row.update(fba_on_hand=_int(available), inbound_units=_int(p.get("inbound")) if p else None,
velocity_30d=velocity, days_of_cover=_int(cover))
row["return_units"] = None if returns is None else _int(returns.get(sku) or 0)
traffic = pricing.get(asin) if asin else None
sessions = _int(traffic.get("sessions")) if traffic else None
box = _f(traffic.get("buyBox")) if traffic else None
row.update(sessions=sessions, buy_box_pct=None if box is None else box / 100.0,
unit_session_pct=_ratio(e.get("units"), sessions))
per_day = daily.get(sku, {})
for index, key in enumerate(SERIES_KEYS):
row[key] = series_cell(per_day.get(day, (0.0, 0.0, 0.0))[index] for day in days)
out[sku] = row
return out
# ---------------------------------------------------------------- grain rollups
def _sum(values):
found = [v for v in (_f(x) for x in values) if v is not None]
return sum(found) if found else None
def _active(row):
return bool(_f(row.get("net_sales")) or _f(row.get("units_sold")))
def rollup(rows: list[dict], *, key: str) -> list[dict]:
"""One row per distinct rows[i][key] (asin or parent_asin): sums, recomputed ratios, then score_row keys."""
if key not in ("asin", "parent_asin"):
raise ValueError("Roll up by asin or parent_asin.")
groups, skipped = {}, []
for row in rows:
value = _text(row.get(key))
if value is None:
skipped.append(row.get("sku"))
continue
groups.setdefault(value, []).append(row)
last_report.clear()
last_report.update(key=key, groups=len(groups), skipped=len(skipped),
skipped_skus=sorted(str(s) for s in skipped if s is not None))
out = []
for value in sorted(groups):
children = sorted(groups[value], key=lambda r: str(r.get("sku") or ""))
skus = [str(c.get("sku")) for c in children if c.get("sku") is not None]
row = {key: value, "title": next((c.get("title") for c in children if c.get("title")), None),
"sku_count": len(children), "skus": ", ".join(skus)}
row["parent_asin"] = value if key == "parent_asin" else next(
(_text(c.get("parent_asin")) for c in children if _text(c.get("parent_asin"))), value)
for k in ADDITIVE_KEYS:
if k in _STRICT_KEYS and any(c.get(k) is None and _active(c) for c in children):
row[k] = None
continue
total = _sum(c.get(k) for c in children)
row[k] = _half_up(total) if total is not None and k in _INT_KEYS else total
for k in SERIES_KEYS:
total = [0.0] * SERIES_DAYS
for c in children:
for i, v in enumerate((_series(c.get(k)) or [])[:SERIES_DAYS]):
total[i] += v
row[k] = series_cell(total)
row.update(ctr=_ratio(row["ad_clicks"], row["ad_impressions"]),
cvr=_ratio(row["ad_orders"], row["ad_clicks"]), cpc=_ratio(row["ad_spend"], row["ad_clicks"]),
acos=_ratio(row["ad_spend"], row["ad_sales"]), roas=_ratio(row["ad_sales"], row["ad_spend"]),
margin=_ratio(row["net_profit"], row["net_sales"]) if row["net_profit"] is not None else None,
ads_note=NO_ADS_NOTE if all(c.get("ads_note") == NO_ADS_NOTE for c in children) else "")
# Two seller SKUs on one FNSKU report the same inventory pool, so stock counts once per FNSKU.
pools = {}
for c in children:
pool = _text(c.get("fnsku")) or _text(c.get("sku"))
if pool and pool not in pools:
pools[pool] = (_f(c.get("fba_on_hand")), _f(c.get("inbound_units")))
on_hand = _sum(h for h, _i in pools.values())
inbound = _sum(i for _h, i in pools.values())
row["fba_on_hand"] = _int(on_hand) if on_hand is not None else None
row["inbound_units"] = _int(inbound) if inbound is not None else None
velocity = row["velocity_30d"]
cover = _ratio(row["fba_on_hand"], velocity / 30) if row["fba_on_hand"] is not None and velocity else None
row["days_of_cover"] = _int(cover)
# A SKU's sessions are its ASIN's, so they are summed once per distinct ASIN.
traffic = {}
for c in children:
asin = _text(c.get("asin"))
if asin and asin not in traffic:
traffic[asin] = (_f(c.get("sessions")), _f(c.get("buy_box_pct")))
sessions = _sum(s for s, _b in traffic.values())
boxes = [(b, s or 0.0) for s, b in traffic.values() if b is not None]
weight = sum(w for _b, w in boxes)
row["sessions"] = _int(sessions)
row["buy_box_pct"] = (sum(b * w for b, w in boxes) / weight if weight
else (sum(b for b, _w in boxes) / len(boxes) if boxes else None))
row["unit_session_pct"] = _ratio(row["units_sold"], row["sessions"])
if key == "asin":
row["asin"] = value
row.update(score_row(row))
row.update(blank_notes(row))
out.append(row)
return out
# ---------------------------------------------------------------- validate
def _check(name, ok, a, b, gap, note=""):
return {"check": name, "ok": bool(ok), "a": None if a is None else round(a, 2),
"b": None if b is None else round(b, 2), "gap": None if gap is None else round(gap, 4),
"note": note}
def validate_metrics(con, *, start, end, computed: dict[str, dict]) -> list[dict]:
"""Independent ties: [{check, ok, a, b, gap, note}] in amazon_profit.validate's shape."""
skus = [s for s in (_text(k) for k in computed) if s]
checks = []
spend = sum(_f(r.get("ad_spend")) or 0.0 for r in computed.values())
econ_columns = _econ_ads_columns(con)
if econ_columns:
# Summed one column at a time in Python, a different path from the per SKU row sum.
total = 0.0
for column in econ_columns:
total += float(con.execute(
f"SELECT COALESCE(SUM(TRY_CAST({_q(column)} AS DOUBLE)), 0) FROM {ECONOMICS} WHERE "
f"left(CAST(start_date AS VARCHAR), 10) BETWEEN ? AND ? AND "
f"TRIM(CAST(msku AS VARCHAR)) IN (SELECT unnest(?))", [start, end, skus or [""]]).fetchone()[0])
gap = abs(spend - total)
checks.append(_check("ads_spend_ties_economics", gap <= SPEND_TIE_TOLERANCE, spend, total, gap,
"Ad spend against the economics ad charges in the period, within one cent."))
else:
checks.append(_check("ads_spend_ties_economics", True, 0, 0, 0, "Economics carries no ad charges."))
campaigns = [t for t in ADS_CAMPAIGNS if {"date", "cost"} <= _columns(con, t)]
if econ_columns:
checks.append(_check("ads_spend_ties_campaigns", True, None, None, None,
"Ad spend comes from economics, so the campaign reports do not apply."))
elif not campaigns and _ads_parts(con) is None:
checks.append(_check("ads_spend_ties_campaigns", True, 0, 0, 0, NO_ADS_NOTE))
elif not campaigns:
checks.append(_check("ads_spend_ties_campaigns", False, spend, None, None,
"Campaign reports are missing, so ad spend cannot be tied."))
else:
union = " UNION ALL ".join(f"SELECT {_day('date')} AS day, TRY_CAST(cost AS DOUBLE) AS cost FROM {t}"
for t in campaigns)
total = float(con.execute(f"SELECT COALESCE(SUM(cost), 0) FROM ({union}) c WHERE day BETWEEN "
"CAST(? AS DATE) AND CAST(? AS DATE)", [start, end]).fetchone()[0])
gap = abs(spend - total) / abs(total) if total else abs(spend - total)
checks.append(_check("ads_spend_ties_campaigns", gap <= ADS_TIE_TOLERANCE, spend, total, gap,
"Advertised product spend against campaign spend, within 2%."))
on_hand = sum(_f(r.get("fba_on_hand")) or 0.0 for r in computed.values())
cols = _columns(con, PLANNING)
if not {"sku", "available"} <= cols:
checks.append(_check("stock_ties_planning", True, 0, 0, 0, NO_STOCK_NOTE))
else:
newest = ('CAST("_snapshot_date" AS VARCHAR) = (SELECT MAX(CAST("_snapshot_date" AS VARCHAR)) '
f"FROM {PLANNING})" if "_snapshot_date" in cols else "TRUE")
total = float(con.execute(f"SELECT COALESCE(SUM(TRY_CAST(available AS DOUBLE)), 0) FROM {PLANNING} "
f"WHERE {newest} AND {_sku_sql('sku')} IN (SELECT unnest(?))",
[skus or [""]]).fetchone()[0])
gap = abs(on_hand - total)
checks.append(_check("stock_ties_planning", gap <= COUNT_TIE_TOLERANCE, on_hand, total, gap,
"FBA on hand against the newest planning snapshot."))
returned = sum(_f(r.get("return_units")) or 0.0 for r in computed.values())
union = _returns_parts(con)
if union is None:
checks.append(_check("returns_tie", True, 0, 0, 0, NO_RETURNS_NOTE))
else:
total = float(con.execute(f"SELECT COALESCE(SUM(units), 0) FROM ({union}) r WHERE day BETWEEN "
"CAST(? AS DATE) AND CAST(? AS DATE) AND sku IN (SELECT unnest(?))",
[start, end, skus or [""]]).fetchone()[0])
gap = abs(returned - total)
checks.append(_check("returns_tie", gap <= COUNT_TIE_TOLERANCE, returned, total, gap,
"Returned units against the returns tables in the period."))
short = [sku for sku, r in computed.items() for k in SERIES_KEYS
if len(_series(r.get(k)) or []) != SERIES_DAYS]
checks.append(_check("trend_length", not short, len(short), 0, len(short),
f"Every trend has {SERIES_DAYS} days." if not short
else f"Short trends: {', '.join(sorted(set(map(str, short))))}."))
return checks