Backlog item: research/projects/thesis-ideas.md:9 — "Tungsten Supply Chain" (Priority 1 BLIND TEST #1) Source claim: @kawzinvests, 2026-04-11 capture — tungsten $320 → $3,000/mtu in 18 months (837%); China exports zero, now net importer; "China spent 40 years deliberately running western tungsten mines into closure. Below-cost production. Sustained market flooding. Hemerdon. Hollister. Strawberry. One by one they went uneconomic." Verdict: Thesis confirmed; western pure-play surfaced via blind trace (ALMTF/ALMU); actionability is mixed — one clean toll-booth (KMT), one binary pure-play (ALMU, gated on user), one ticker that converged but now belongs to a different chain (CRS).

---

What we're asking

The backlog item posed four questions:

1. Who mines tungsten outside China?
2. What products need tungsten? (cutting tools, military, electronics?)
3. What companies are bottlenecked?
4. Are there US-listed pure-plays?

These are physics-first questions. The protocol called for a blind trace — reason from the chain, not from "what's already running in our monster scan." That trace ran on 2026-04-11 (3 agents, blind config) at research/projects/autonomous-research/traces/2026-04-11-tungsten-supply-chain.json. This investigation closes out the blind-test cycle: did the trace land on the right names, what's left ungated, and what's the trade today.

The chain (physics)

Tungsten flows through four distinct value-chain layers, each with different concentration and substitution risk:

The bottleneck logic: Layer 1 (mine output) was deliberately destroyed by China over 30+ years. Restart is slow (Hemerdon took ~6 years; Sangdong ~10 years). Layer 2 (APT/WC powder) is capital-intensive and largely private. Layer 3 (cutting tools) is the toll-booth — whoever owns the cemented-carbide tooling franchise extracts margin on every reshored / nearshored manufacturing operation that consumes a drill bit or cutting insert.

Substitution insight: WC has no scale substitute for hard-metal cutting (the geometry + hardness + cost combination doesn't replicate in CBN/PCD at industrial volume). That's the moat. Military KE penetrators have a depleted-uranium substitute — but politically constrained outside specific platforms (M829 tank rounds historically used DU; many programs are migrating back to WC for ITAR / non-DU reasons).

What the blind trace found

research/projects/autonomous-research/traces/2026-04-11-tungsten-supply-chain.json (3 agents, blind, config=claude):

What the trace missed (international): Sandvik (SAND.ST), Mitsubishi Materials (5711.T), AMG Critical Materials (AMG.AS). These are non-US-listed and the blind agent was likely biased toward US tickers. The miss is a methodology footnote, not a thesis failure — KMT is the cleanest US-listed Layer-3 toll-booth and the trace got that right.

What the trace missed (private): HC Starck, Global Tungsten & Powders (Plansee subsidiary), Buffalo Tungsten — Layer 2 is almost entirely private. No actionable public pure-play on the APT/WC-powder chemistry layer in the US-listed universe. This is a real structural gap, not a trace failure.

Data note: ALMTF showed "no data" in yfinance for the 2026-04-26 supply-chain-traces scan. The Almonty Nasdaq dual-listing (ALMU) resolves the data problem — ALMU is now in stonks/watchlists/supply-chain-traces.json and on stonks/watchlists/watching.json (per MEMORY note: "ALMU is now in watching.json"). Use ALMU as the trackable ticker; ALMTF is the OTC/TSX legacy listing.

State of each name (point-in-time receipts, not fresh tape)

These are the most recent observations on file. Per the backlog rule, fresh tape was not required for this verdict.

ALMU (Almonty Industries) — Layer 1 pure-play

• TASKS-RESEARCH.md:19 — "ALMU binary-outcome deep-dive" — deferred to user. Reinforced by @AlmaCap114204 + @Speculator_io. Open questions: TSEM partnership/contract details, IP/patent posture on the QD-laser-on-silicon pivot, revenue split today (tungsten mining vs. photonics), management track record. Tagged #binary because Sangdong restart + photonics pivot are both step-functions, not gradual ramps.
• Already in supply-chain-traces.json and watching.json watchlists.
• Earnings scheduled 2026-05-13 (per research/api-data/events/2026-05-13-almu-earnings.json — status "captured" but body indicates auto-scheduled, not yet populated with prints).
• No new action this investigation. ALMU stays on user's plate per the existing #defer-to-user tag.

KMT (Kennametal) — Layer 3 toll-booth

• 2026-04-26 supply-chain-traces scan: $39.20, RSI 57.3, +14.39% 3M, +36% vs SMA200. Status: 🟡 Healthy / 📈 Accumulate. Earlier 2026-04-13 scan: $39.57, RSI 60.9, +18.7% 3M, +40.8% vs SMA200 — golden cross intact.
• No deep dive on file. This is the cleanest follow-up: KMT is a public US-listed industrial with a dividend, structurally exposed to tungsten carbide pricing AND to reshoring/CHIPS/IRA capex, with no binary-outcome risk attached. Filed below as [ ] in TASKS-RESEARCH.md.

CRS (Carpenter Technology) — Layer 4-ish, reclassified

• 2026-05-11 deep-dive done: $429.41, +110% TTM, $21.3B market cap, op margin 19% → 23% in 4 quarters. Conviction medium. Thesis: nickel-superalloy supplier to IGT-blade casters (HWM / PCC / OEM-mediated), confirmed by ATI 10-K naming CRS as the legacy-nickel-superalloy competitor. Tungsten is a minor ingredient (some specialty steels, X-ray-target alloys) — not the demand driver. Already on ai-power watchlist + ai-power-bottleneck perspective.
• Reclassification action: the trace surfaced CRS for tungsten reasons but the IGT-blade thesis has supplanted that framing. Treat CRS as an ai-power-bottleneck ticker, not a tungsten ticker. CRS should remain in supply-chain-traces.json as a historical trace artifact (it was found by the agents and the call worked) but its tungsten association is incidental, not load-bearing.

Verdict + reasoning

Thesis (kawzinvests, 2026-04-11): China spent 40 years engineering western tungsten supply destruction → price 320 → 3,000/mtu in 18 months → western mine restart + downstream toll-booth thesis exists.

Verdict: CONFIRMED on the structural claim. MIXED on actionability.

Why "mixed actionability":

• ALMU is binary-outcome (Sangdong scale-up + photonics pivot are step-functions) and gated on user attention per TASKS-RESEARCH.md:19. Not autonomous-grade.
• KMT is the clean trade-shape: industrial, dividend, no binary catalyst, structurally levered to reshoring + tungsten pricing + WC moat. But the price has already moved — 2026-04-26 scan had it +36% vs SMA200, RSI 57. Entry zone in that scan was $36-38; April price was $39.20. So even the clean toll-booth has run.
• CRS is reclassified — don't count it as a tungsten trade.
• AMG.AS / SAND.ST / 5711.T are coverage gaps. Whether to add them depends on whether the user wants international tungsten exposure (most US scans don't pick them up).

The trace did its job. Three-agent blind convergence found the only US-listed pure-play (ALMU) and the only US-listed toll-booth (KMT), correctly identified the cutting-tool layer as the demand bottleneck, and surfaced a third name (CRS) that turned out to belong to a different chain — that's a feature, not a bug, because CRS is now a confirmed ai-power-bottleneck ticker that the system might not have found otherwise. This trace is a positive datapoint for the convergence-then-verify methodology.

The non-actionable claim: "Tungsten 320→3,000/mtu in 18 months" — verified at the captured-source level (@kawzinvests, screenshot in research/inputs/2026-04-11-follows-roundup.md). Source-of-source claim ("China spent 40 years...") is plausible — Hemerdon UK closed 2018, Hollister NV mothballed mid-2010s, Strawberry CA closed 2010s. Specific dates not independently verified in this investigation (no fresh web fetches per the no-broad-scan rule). If a fresh verification pass is needed, fetch USGS Tungsten Mineral Commodity Summaries 2024-2025 + LME/Argus tungsten APT pricing — that's a focused, narrow task, not a tape-scan.

Coverage-gap follow-ups (filed as [ ] in TASKS-RESEARCH.md)

• KMT deep-dive — cleanest follow-up. Layer 3 toll-booth, US-listed, no binary risk, structurally exposed to tungsten + reshoring. Run standard /deep-dive KMT. Specific questions: cemented-carbide pricing pass-through (WC powder cost → tool ASP), customer-concentration disclosure (industrial vs. mining vs. aerospace mix), capacity-expansion capex in the 2025-2026 reshoring window, dividend sustainability through a downcycle. ~30-45 min.
• Sandvik (SAND.ST) coverage decision — international Layer-3 peer of KMT. Larger and more diversified. Decide whether to add to a watchlist (supply-chain-traces or a new tungsten watchlist) or pass. ~10 min.
• AMG Critical Materials (AMG.AS) — Layer-2 closest-to-pure-play public name on the powder/chemistry side. Tungsten exposure is one of several critical materials. Decide watchlist coverage. ~15 min.
• Mitsubishi Materials (5711.T) — Japanese Layer-3 peer. Probably won't add (Japan listing, low US tracking), but worth noting as a coverage gap. ~5 min.

What didn't get done (transparent)

• No fresh price/RSI fetch — by the backlog rule. Prices cited are point-in-time from existing scan events.
• No web verification of the "Hemerdon / Hollister / Strawberry" closure timeline or USGS tungsten production share figures. The structural claim is plausible and consistent with industry knowledge but not independently re-verified for this investigation. Filed as [ ] if it ever becomes load-bearing.
• ALMU binary-outcome deep-dive remains deferred to user (TASKS-RESEARCH.md:19) — not reopened by this investigation.
• No KMT deep-dive run today — filed as follow-up below.

Receipts

• Source claim capture: research/inputs/2026-04-11-follows-roundup.md (kawzinvests, $320→$3,000/mtu screenshot).
• Blind trace artifact: research/projects/autonomous-research/traces/2026-04-11-tungsten-supply-chain.json (3 agents, blind config, ALMTF/KMT/CRS convergence).
• Trace scan results (point-in-time prices): research/api-data/events/2026-04-13-supply-chain-traces.json, research/api-data/events/2026-04-26-supply-chain-traces.json.
• CRS reclassification: research/deep-dives/2026-05-11-crs-deep-dive.md (CRS = nickel-superalloy / IGT-blade thesis, not tungsten).
• ALMU deferral: TASKS-RESEARCH.md:19 (#defer-to-user, #binary, #optical-supercycle).
• Watchlist state: stonks/watchlists/supply-chain-traces.json (ALMU, KMT, CRS already present); stonks/watchlists/watching.json (ALMU per MEMORY note).
• Backlog item closed: research/projects/thesis-ideas.md:9 (Priority 1 BLIND TEST #1 → marked resolved with link to this investigation).
• Follow-up filed: TASKS-RESEARCH.md — KMT deep-dive [ ] (+ optional SAND.ST / AMG.AS / 5711.T coverage decisions).

Run deno task research:feeds after this commits to project into ticker/perspective feeds.