mwosa

name	mwosa
description	Use when working in the mwosa repo or explaining mwosa CLI workflows, especially Datago ETF daily collection, canonical SQLite data, jq-based ETF screening, raw Datago JSON experiments, and saved screening strategies.

mwosa

First checks

• Work from the repository root.
• Before code edits, read RULE.md and check git status --short --branch.
• Prefer the installed mwosa CLI for user-facing commands. Use go run only when testing local source changes or running experiment-only tools that are not installed as CLI commands.
• When you need the complete installed command surface, read references/cli-command-help.md; it is generated from mwosa --help plus subcommand help.
• Keep stdout machine-readable for json, ndjson, csv, and jq pipelines. Put progress, diagnostics, and explanations on stderr or in chat.

ETF daily collection

Use canonical SQLite collection when the user asks to collect ETF data for analysis:

mwosa backfill daily \
  --provider datago \
  --security-type etf \
  --from YYYY-MM-DD \
  --to YYYY-MM-DD \
  --workers 4 \
  -o json

For one trading date, set --from and --to to the same date. If datago is already the configured/default provider, --provider datago can be omitted, but keep it in handoff commands when clarity matters.

Verify a known ETF after collection:

mwosa get daily 069500 \
  --security-type etf \
  --from YYYY-MM-DD \
  --to YYYY-MM-DD \
  -o json

If provider auth is suspect, inspect or validate the provider without printing secrets:

mwosa provider doctor datago -o json

Current jq screening surface

Check references/cli-command-help.md before recommending jq commands. The installed mwosa v0.1.0 exposes saved strategy commands, while newer source checkouts may also expose one-off mwosa screen etfs --jq/--jq-file.

Installed v0.1.0 strategy flow:

mwosa create strategy etf-weekly-leaders \
  --engine jq \
  --input etf_daily_metrics \
  --jq-file strategies/etf-weekly-leaders.jq \
  -o json

mwosa screen strategy etf-weekly-leaders \
  --alias YYYY-MM-DD-weekly-leaders \
  -o json

mwosa history screen -o table
mwosa inspect screen YYYY-MM-DD-weekly-leaders -o json

Do not promise runtime --argjson support for saved strategy execution unless the codebase has been checked again; the current CLI flags only expose --alias on screen strategy.

Dataset schema and keys

etf_daily_metrics currently reads canonical ETF daily bars from SQLite. Despite the name, treat it as daily-bar records unless the codebase has added derived metrics.

Canonical strategy input rows use these JSON keys:

Key	Meaning	Type
provider	provider id, usually datago	string
provider_group	provider group, usually securitiesProductPrice	string
operation	source operation, e.g. getETFPriceInfo	string
market	market id, usually krx	string
security_type	etf, etn, or elw	string
trading_date	trading date, YYYY-MM-DD	string
symbol	KRX short code such as 069500	string
isin	ISIN code	string
name	item name	string
currency	currency, usually KRW	string
opening_price	open price	numeric string
highest_price	high price	numeric string
lowest_price	low price	numeric string
closing_price	close price	numeric string
price_change_from_previous_close	absolute change from previous close	numeric string
price_change_rate_from_previous_close	percent change from previous close	numeric string
traded_volume	traded volume	numeric string
traded_amount	traded value/amount	numeric string
market_capitalization	market capitalization	numeric string
extensions	provider-specific extra fields	object of strings

Convert numeric strings inside jq with tonumber? // 0.

Common Datago ETF extras are stored under extensions after canonical storage:

Extension key	Meaning
nPptTotAmt	ETF net asset total amount
stLstgCnt	ETF listed share count
nav	ETF NAV
bssIdxIdxNm	underlying index name
bssIdxClpr	underlying index close

For raw Datago JSON files, use provider-native keys instead of canonical keys:

Raw key	Canonical key
basDt	trading_date
srtnCd	symbol
isinCd	isin
itmsNm	name
mkp	opening_price
hipr	highest_price
lopr	lowest_price
clpr	closing_price
vs	price_change_from_previous_close
fltRt	price_change_rate_from_previous_close
trqu	traded_volume
trPrc	traded_amount
mrktTotAmt	market_capitalization

When writing jq for canonical SQLite screens, use canonical snake_case keys. Use raw camel-case Datago keys only inside testing/experiments/datago_daily_json_collector scripts or raw snapshot files.

Practical jq examples

Latest-date liquidity leaders query:

def n: tonumber? // 0;
group_by(.symbol)
| map(max_by(.trading_date))
| sort_by((.traded_amount | n))
| reverse
| .[:50]

Save and run a query file on installed v0.1.0:

mwosa create strategy latest-liquidity-leaders \
  --engine jq \
  --input etf_daily_metrics \
  --jq-file strategies/latest-liquidity-leaders.jq \
  -o json

mwosa screen strategy latest-liquidity-leaders \
  --alias YYYY-MM-DD-liquidity-leaders \
  -o json

Period return leaders query from stored daily bars:

def n: tonumber? // 0;
map(select(.trading_date >= "YYYY-MM-DD" and .trading_date <= "YYYY-MM-DD"))
| group_by(.symbol)
| map(
    sort_by(.trading_date) as $rows
    | ($rows[0].closing_price | n) as $start
    | ($rows[-1].closing_price | n) as $end
    | select(length >= 2 and $start > 0 and $end > 0)
    | {
        symbol: $rows[-1].symbol,
        name: $rows[-1].name,
        from: $rows[0].trading_date,
        to: $rows[-1].trading_date,
        return_pct: (($end / $start - 1) * 100),
        traded_amount: ($rows[-1].traded_amount | n)
      }
  )
| sort_by(.return_pct)
| reverse
| .[:50]

For CSV output from one-off screens, prefer -o csv only when rows are flat. If payloads are nested, use -o json | jq -r ... to choose exact columns.

Raw Datago JSON experiment scripts

Use the raw collector only when the user specifically wants date-partitioned source JSON or the existing experiment screeners. It writes one JSON file per date under tmp/testing/datago-daily-json-collector/raw.

DATAGO_SERVICE_KEY="$DATAGO_SERVICE_KEY" \
go run ./testing/experiments/datago_daily_json_collector \
  --products etf \
  --start-date YYYY-MM-DD \
  --end-date YYYY-MM-DD \
  --workers 2 \
  --overwrite=true

Run the momentum screener:

testing/experiments/datago_daily_json_collector/scripts/next_day_momentum/screen_next_day_momentum.sh \
  --raw-dir tmp/testing/datago-daily-json-collector/raw \
  --preset balanced \
  --format csv

Run the low-volatility uptrend screener:

testing/experiments/datago_daily_json_collector/scripts/low_vol_uptrend/screen_low_vol_uptrend.sh \
  --raw-dir tmp/testing/datago-daily-json-collector/raw \
  --format csv

Always pass --raw-dir tmp/testing/datago-daily-json-collector/raw explicitly for these scripts. When results look stale or empty, inspect the target date file and rerun with overwrite; old empty date files have caused misleading screens before.

How to answer users

• If the user asks for a command, give the installed mwosa command first.
• If the user asks how jq screening works, explain the two layers: canonical SQLite strategy screens through installed create/screen strategy, raw JSON experiments through testing/experiments/.../scripts.
• Keep Korean explanations concise and command-first.
• When producing candidate files for review, create both machine-friendly JSON and small human-openable CSV when the output could be large.