Verdict: Keep current scanner routing as-is; formalize DISCO + Samsung as research_only in _securities.json; do not add 6146.T/005930.KS to a tracked watchlist. BESI and Advantest already route correctly through OTC proxies (BESIY, ATEYY) and need no further work. SK hynix is routed through the German GDR (HY9H.F) as a tracked foreign_primary and stays as-is. The policy is: ADR/OTC proxy where one exists with adequate float; research_only security-master row for the JP/KR primary when scanner-fetchable; only add a tracked foreign primary when an ADR is genuinely unavailable AND the name is load-bearing for an active perspective.

What we're asking

The 2026-05-25 HBM verdict (research/investigations/2026-05-25-hbm-memory-supercycle-blind-test-verdict.md) confirmed five equipment names at the right physics layers and surfaced four foreign-primary names sitting around the edges of our coverage:

• DISCO Corp (6146.T, JP) — Layer-1 wafer thinning leader; the only major HBM-layer name not currently in any watchlist. No US ADR.
• Samsung Electronics (005930.KS, KR) — One of the three HBM memory makers; demand-side. No 005930 ADR; the closest US line is SSNLF (heavily restricted on most US brokers).
• BE Semiconductor Industries (BESI.AS / BESIY) — Already routed through BESIY (OTC proxy) and tracked in supply-chain-traces; deep-dive done.
• Advantest (6857.T / ATEYY) — Already routed through ATEYY (OTC proxy) and tracked in supply-chain-traces; deep-dive filed but not yet written.

Cross-checking SK hynix completes the set: it's routed through HY9H.F (Frankfurt GDR) as a foreign_primary in _securities.json and lives in memory.json.

The TASKS-RESEARCH.md:41 entry asks for a policy decision before we add more foreign primaries piecemeal:

(a) keep 6146.T DISCO and 005930.KS Samsung as research_only in stonks/watchlists/_securities.json; (b) add 6146.T to supply-chain-traces.json only if JP primary listings become an authorized scanner pattern; (c) during ATEYY deep-dive, compare ATEYY ADR liquidity vs 6857.T primary; (d) update any report prose to show scanner symbol / primary listing / venue / currency on first mention.

The question is whether to (i) preserve the OTC-proxy-first routing that's been the de-facto pattern (BESIY/ATEYY), (ii) start tracking JP/KR primaries directly, or (iii) some hybrid. The cost of getting this wrong is research debt: scanner output drifting away from how research notes cite names, OR scanner output filling with names US users can't actually trade on Schwab/Fidelity.

What we found

Current security-master state (stonks/watchlists/_securities.json)

The good news is the file is already in the shape this task asks for. DISCO and Samsung are already tagged research_only; BESI/Advantest/Kulicke are already routed correctly. The decision the task points at is therefore mostly a policy ratification — write down the rule that drove these settings so the next foreign-primary case (whether HBM-related or not) gets handled the same way.

Cross-check against the watchlists

• memory.json: MU, STX, WDC, DRAM, HY9H.F. SK hynix flows in via the German GDR. Samsung is not in the watchlist — it would be redundant with the SK hynix/MU pair on the demand-side memory-supercycle thesis.
• supply-chain-traces.json: includes BESIY, ATEYY, KLIC, AMKR, ASX (plus tungsten + other-trace names). No 6146.T. The watchlist is uniformly ADR/US-listed; adding a JP primary would be the first deviation from that pattern.
• _aliases.json: lightweight capture dedupe only — not the right place for any of this. The _securities.json registry is the canonical answer.

Liquidity / accessibility — the practical case for ADR-first

The "OTC proxy first" pattern persists because it solves three problems at once:

1. Most US brokers (Schwab, Fidelity, IBKR, paper accounts) can route an OTC ADR like BESIY/ATEYY without a foreign-account upgrade. JP/KR primary tickets require either a global account or a specialized broker.
2. yfinance OHLC is generally reliable on US-listed OTC ADRs; JP/KR primaries can break on weekends, holidays, or split-adjustment events without warning. Verified working today (6146.T, 005930.KS, 6857.T all fetched 30d cleanly) but that's not a stability guarantee.
3. Prose citation lines up: writing "BESIY (OTC proxy for AMS:BESI)" once at first mention is cheaper than maintaining a parallel 6146.T/6857.T shadow column in every report.

The case against ADR-first surfaces in two specific cases:

• No ADR exists with adequate float. DISCO has no US ADR. Its closest US line (DSCSY) has historically been spotty; not in our security master at all. So DISCO is a genuine "foreign primary or nothing" name.
• ADR is so illiquid it breaks scanner math. ATEYY at ~$169 with ATR/spreads on US OTC is fine; some Japanese small-caps trade single-digit ADR volumes that make daily summaries noisy.

The SK hynix precedent

SK hynix is the existing template for what to do when an ADR is unavailable AND the name is load-bearing. The German Frankfurt GDR (HY9H.F) was chosen as the scanner symbol because: (a) no liquid US ADR; (b) Frankfurt GDR has more daily volume and cleaner OHLC than Pink Sheet alternatives; (c) tagged as foreign_primary (not otc_proxy, not primary_listing) which signals "this is the best available US-accessible line for a foreign primary." This is the pattern to copy if a future name genuinely needs the same treatment.

The DISCO question specifically

DISCO is the only HBM-layer name currently sitting as research_only with active research need. Three options:

Option A is the right answer until JP-listing coverage becomes an authorized scanner pattern — which is the same thing the security-master note already says. The right gate to add Japanese primaries to tracked watchlists is explicit user authorization for the listing class, not name-by-name discretion. Today, DISCO stays as a research-only reference in the HBM verdict body, identical to how 5711.T (Mitsubishi Materials) is handled for the tungsten thesis (see 2026-05-25-tungsten-supply-chain-blind-test-verdict.md for the parallel case).

The Samsung question

Samsung is even more clearly research-only correct. The memory-supercycle thesis is already captured on the cohort via MU + HY9H.F + STX + WDC + DRAM; Samsung adds no incremental coverage and trading it via US brokers is heavily restricted. Reference in the memory-supercycle perspective body as "Samsung (005930.KS, research-only — tracked indirectly via SK hynix as cohort peer)" and move on.

The Advantest-deep-dive cross-check

The deep-dive can include a one-paragraph ADR-vs-primary comparison: ATEYY OTC volume + spread, 6857.T primary volume, currency-translation drift between the two on a 30D window. If the spread is wide enough that ATEYY-printed prices are misleading the cohort math, that's a reason to revisit. If not (the typical case), the dual-listing footnote stays informational. That's a normal step in any deep-dive on an OTC-proxy name — it doesn't require a security-master change.

Verdict and reasoning

Decision: keep current routing as-is and ratify the policy.

Policy in one line. For a foreign issuer covered in our research:

OTC/ADR proxy if one exists with adequate float and reasonable spread; foreign_primary (e.g. HY9H.F-style GDR) if no US line exists and the name is load-bearing; research_only if no US-accessible line exists and the name is informational/peer-context only. Do not add JP/KR/non-EU primaries to tracked watchlists name-by-name; wait for an authorized listing-class decision that takes multiple names at once.

Prose convention (to update the perspectives editorial playbook on the next refresh): on first mention of a foreign-issuer family, write scanner symbol (primary listing, venue, currency, role) — e.g. "BESIY (BESI.AS, AMS, EUR, OTC proxy)" or "DISCO (6146.T, JPX, JPY, research-only)." This is the convention already in use in _securities.json notes; the action is to mirror it in editorial prose for clarity, not to invent it.

Coverage gaps / follow-ups

• No new watchlist adds from this investigation. The "ratify the policy" outcome is the deliverable.
• Prose-convention codification in the editorial playbook — file as a single line in .brief/perspectives-editorial-playbook.md next time it's edited. Not its own task; bundle with the next playbook touch.
• ATEYY deep-dive should include an ADR-vs-primary check — already in scope for the ATEYY deep-dive task per the source TASKS-RESEARCH.md:37 (question (e)). No new task.
• If JP-listing coverage gets authorized as a scanner pattern in a future session, revisit DISCO (6146.T) + 6857.T (in addition to ATEYY) + 5711.T (tungsten cross-thesis) together. That's the batch-trigger condition; do not relitigate name-by-name.

What didn't get done (transparency)

• No fresh OHLC fetch on 6146.T / 005930.KS / 6857.T spread-vs-ADR check. This investigation is a policy ratification of the already-correct _securities.json rows, not a freshness exercise. The ATEYY-vs-6857.T liquidity comparison is deferred to the deep-dive.
• No update to .brief/perspectives-editorial-playbook.md. Prose convention is settled here; codification will happen on next editorial-playbook touch.
• No expansion of _securities.json rows — the relevant rows already exist with the right role tags as of the 2026-05-25 update.

Receipts

Source backlog item:

• TASKS-RESEARCH.md:41 — "HBM foreign-primary / ADR coverage backfill decision — DISCO, Samsung, BESI, Advantest." Closed by this investigation.

Cited and consolidated:

• research/investigations/2026-05-25-hbm-memory-supercycle-blind-test-verdict.md — HBM cohort coverage state map; surfaced the foreign-primary question.
• stonks/watchlists/_securities.json — security-family registry, already in the shape this verdict ratifies.
• stonks/watchlists/memory.json — memory cohort, already includes HY9H.F via the foreign_primary pattern.
• stonks/watchlists/supply-chain-traces.json — equipment cohort, includes BESIY/ATEYY/KLIC via OTC-proxy / direct patterns.

Filed follow-ups (in TASKS-RESEARCH.md):

• None new. ATEYY deep-dive already filed and will pick up the ADR-vs-primary check question. Editorial-playbook prose convention is a one-line addition on next touch, not its own task.

Commands run:

deno task beta:emit investigation \
  --topic "HBM foreign-primary / ADR coverage backfill decision — DISCO, Samsung, BESI, Advantest" \
  --slug "hbm-adr-foreign-primary-backfill-decision" \
  --tickers "BESIY,ATEYY,KLIC,MU,HY9H.F" \
  --perspectives "memory-supercycle" \
  --watchlists "memory,supply-chain-traces" \
  --cites "2026-05-25-hbm-memory-supercycle-blind-test-verdict"