Bandar check

Broker flow is the one thing this server has that no general market API does. It is also the easiest thing to over-read, so the order below is not optional: establish that the tape is meaningful before saying anything about who is on it.

The sequence

1. market_session — is the market open, pre-open, or closed? Today's flow is partial until it closes. Say so rather than quietly presenting half a day as a day.
2. broker_summary symbol=… — the ledger. Every broker's buy and sell over the window.
   • period takes one of FIVE windows, and only these five: LATEST, YESTERDAY, LAST_7_DAYS, LAST_3_MONTHS, YEAR_TO_DATE. The longer ten-value list belongs to broker_top and broker_distribution, which are different endpoints; sending one of those here is rejected, and LAST_1_DAY / LAST_30_DAYS return a 400 from Stockbit itself. Or give from/to for an explicit range. Do not do both — the API ignores the dates and answers with the latest session, silently.
   • broker_activity has no window at all. Measured 2026-09-01: every value of period answers 400 there, so the tool refuses it rather than sending it. Its window is the server's, and it is reported back in from/to on the result — read those to date the rows.
   • transaction_type is NET or GROSS. NET is the question people mean. GROSS tells you who was busy; NET tells you who ended up holding.
   • market_board defaults to the regular board. NEGO and TUNAI are crossings and cash settlement — real, but not the same signal. If the user's question is about accumulation, say which board the number came from.
   • investor_type splits Asing (foreign), Lokal (domestic) and Pemerintah (government).
3. broker_distribution symbol=… — who bought from whom. This is the part that turns a list of names into a story: it pairs the accumulating brokers with the ones supplying them.
4. bandar_detector symbol=… for a ranked read, or broker_activity broker_code=… to ask the opposite question — what else has this broker been doing.
   • In the ranked read, topDistributors is largest seller FIRST, and sell figures are NEGATIVE because that is how the wire sends them. Do not negate them again.

workflow_run name=bandar_watch runs steps 2 and 3 in one call. It does NOT call market_session or bandar_detector, so if you use it, check the session yourself before presenting today's flow as a day's flow.

Reading it honestly

• A broker is not an investor. Brokers execute for clients. A house code accumulating means its clients, in aggregate, accumulated. Institutional-looking flow through a retail-heavy broker is usually many small orders, not one big one.
• A floor-locked stock carries no flow signal. If the price sat at ARB (auto-reject bottom) all session, the only trades that printed were the ones that could. Check price_bands before reading intent into the tape. The same goes for ARA on the way up.
• Foreign flow is not smart money. It is one investor class with its own mandate and its own redemption calendar.
• One day is noise. Persistence over a week or a month is the claim worth making. If the user asked about today, give today, and then say what the longer window shows. Institutions build over weeks; a single session of net buying sits inside ordinary two-way flow.
• Low concentration is not evidence of absence. The same beneficial owner can spread orders across several brokers precisely so this reading shows nothing. "No one broker dominates" and "nobody is accumulating" are different statements, and only the first is supported.
• Net flow is not a motive. Nothing here separates accumulation from a client rotating between accounts, a market maker hedging, an index fund tracking a reweight, or a block crossed by prior agreement. bandar_detector returns no verdict, no score and no direction, and neither should you.
• The board changes the numbers. The default REGULER excludes negotiated blocks; ALL can multiply the figures several times over. Say which board a number came from.

How to phrase it

Lead with the net position over the window, name the top accumulators and distributors with their lot counts, then say what would change your mind. Confidence is a description of the evidence ("five straight sessions of net buying by the same three brokers"), never a percentage. Do not predict a price. End anything that reads like a recommendation with a reminder that this is historical flow, not advice.