# tracking_system.md — Speculation-Tracking System v1.0

A repeatable system for tracking thematic speculation across seven categories, with defined fields,
scan cadences, a markdown register and a JSON schema for automation.

**Last reviewed:** 1 August 2026

> **Market research and educational analysis, not personalised financial advice.** This system tracks
> what investors, analysts and fund managers are watching. It does not tell anyone what to buy or
> sell. Scores are this system's own assessments of *narrative and research interest*, not forecasts
> of return.

---

## 1. Categories

Seven categories. Each entry gets exactly one primary category — if an entry genuinely spans two, it
is split into separate entries rather than tagged twice, because a blurred category is a blurred
thesis.

| # | Category | Scope | Primary file |
|---|---|---|---|
| 1 | **AI Infrastructure** | Semiconductors, semi-cap equipment, memory, networking, data-centre hardware, power and thermal | `ai_infrastructure.md` |
| 2 | **AI Software** | Model providers, data layer, applied AI, agentic workflow platforms, observability | `ai_software.md` |
| 3 | **Frontier Tech** | Cross-cutting deep tech that doesn't sit in a narrower category | `frontier_tech.md` |
| 4 | **EV / Autonomy** | Electric vehicles, robotaxis, autonomy stacks, sensing | `frontier_tech.md` |
| 5 | **Energy Storage** | Batteries, solid-state, grid-scale storage, generation incl. uranium and SMR | `frontier_tech.md` |
| 6 | **Biotech** | AI drug discovery, platform biotech, autonomous labs | `frontier_tech.md` |
| 7 | **Moonshots** | Humanoid robotics, quantum, space, anything without a dated commercial milestone | `frontier_tech.md` |

**Category discipline note.** Miscategorisation is the most common failure in thematic tracking.
DroneShield is counter-drone defence, not AI infrastructure, despite appearing in "ASX AI" lists.
Uranium developers are Energy Storage, not AI Infrastructure, even though the demand narrative comes
from data centres. Put the entry where the *business* sits, and record the crossover in the narrative
field.

---

## 2. Tracking fields

Seven fields per company or theme. Scored fields use 1–5; three are descriptive.

### 2.1 Narrative strength (1–5)

How coherent, externally driven and multi-sourced the story is.

| Score | Meaning |
|---|---|
| 5 | Externally driven by policy, physics or verified demand; visible across many unrelated sources |
| 4 | Strong and widely held, with real evidence underneath |
| 3 | Plausible, but partly company-promoted or thinly sourced |
| 2 | Mostly manufactured by the company or its promoters |
| 1 | No coherent narrative — a ticker in search of a story |

### 2.2 Analyst sentiment (1–5)

What published, named analysts say. **Attribution required** — record who said it.

| Score | Meaning |
|---|---|
| 5 | Broad, strong published coverage with consensus support |
| 4 | Positive coverage with some dispersion |
| 3 | Mixed or actively debated |
| 2 | Thin coverage, or coverage is largely sponsored |
| 1 | No independent coverage exists |

### 2.3 Hedge-fund interest (1–5)

Positioning evidence from 13F filings, prime-brokerage data and fund commentary.

| Score | Meaning |
|---|---|
| 5 | Record or near-record positioning; named repeatedly in fund commentary |
| 4 | Meaningful and rising institutional interest |
| 3 | Present but unremarkable |
| 2 | Minimal institutional participation |
| 1 | Effectively retail-only |

**Read this field in both directions.** A 5 means the informed money agrees *and* that the exit is
crowded. A 1 with a high narrative score is what an Emergent-stage opportunity looks like — see the
software-under-owned observation in `ai_software.md`.

### 2.4 Catalyst timeline

Descriptive, not scored. Record the **next dated event** and its date. If there is no date, write
**"Undated"** — never "soon," "H2," or "upcoming."

Grade the catalyst using the taxonomy in `spec.md` Part 2: scheduled binary, scheduled mechanical,
policy, corporate action, or emergent.

**Undated is a red flag, not a neutral value.** Positions held against undated catalysts are where
speculative capital bleeds out quietly.

### 2.5 Risk tier

Maps to the four tiers used across this site, assigned on **structure** — liquidity, disclosure,
revenue, dilution history — never on price performance.

- **Low (Tier 1)** — real revenue or assets, verifiable disclosure, adequate liquidity
- **Medium (Tier 2)** — listed and liquid, valued on a future story
- **High (Tier 3)** — thin liquidity, serial dilution, minimal independent coverage
- **Extreme (Tier 4)** — attention is the asset; assume zero

### 2.6 Volatility profile

Descriptive: Low / Medium / High / Very high. Records observed behaviour so that a 30% move can be
read as normal or abnormal for that name.

**Volatility raises the risk tier. It never raises the score.**

### 2.7 Long-term theme alignment

Whether the entry is genuinely exposed to a structural change or is a proxy that will decouple.

Record as a short phrase plus a strength: e.g. *"Very high — the binding constraint"* or
*"Medium — input exposure only."*

The test: **if the theme plays out exactly as described, does this specific entity capture value?**
Rare earths mattering says nothing about whether one explorer has an economic deposit. Humanoids
mattering says nothing about whether a listed supplier captures the margin.

---

## 3. Update workflow

Five cadences. Each has a defined trigger, scope and output.

### 3.1 Daily — news scan

*Runs with the existing 06:30 AWST speculation scan.*

- Scan announcements, filings and news across all seven categories
- Update the **catalyst timeline** field where a date is confirmed, moved or passed
- Flag any entry whose narrative was contradicted by a primary source
- Log new entrants; do not manufacture them — a quiet day is a valid result
- **Output:** updated register, appended `history.md` entry

### 3.2 Weekly — earnings and catalyst scan

*Mondays.*

- Roll the 90-day catalyst calendar forward
- Record outcomes of catalysts that resolved, and the price/narrative reaction
- Apply the **saturation test**: did good news produce a flat or negative close?
- Re-score **narrative strength** where evidence changed
- **Output:** catalyst calendar, resolved-catalyst log

### 3.3 Monthly — hedge-fund sentiment scan

*First business day.*

- Update **hedge-fund interest** from the latest 13F, prime-brokerage and fund commentary
- Note positioning extremes in both directions — record highs *and* multi-year lows
- Re-score **analyst sentiment** with attribution
- Re-tier any entry whose **structure** changed (dilution, liquidity, revenue arrival)
- **Output:** positioning summary, tier-migration log

> 13F filings are quarterly and lag ~45 days. Q2 2026 filings were due **14 August 2026**. Treat
> positioning as a picture of the past, not the present.

### 3.4 Quarterly — narrative re-evaluation

*After each reporting season.*

- Re-place every narrative on the five-stage lifecycle; record direction of travel
- Retire entries that reached Decay, whose catalyst passed, or whose score fell 20+ points
- Re-score **long-term theme alignment** against a full quarter of evidence
- Review the register's own hit rate: which *reasons* worked, not which tickers
- **Output:** narrative register update, retirement log, self-assessment

### 3.5 Annual — theme realignment

*January.*

- Ask whether the seven categories still describe the market. The AI Infrastructure/AI Software split
  is a 2025–26 distinction and will not stay useful forever
- Retire dead themes; promote sub-themes that outgrew their parent
- Re-examine the field definitions and the 1–5 anchors for drift
- Compare a full year of scores against what actually happened — the only real validation available
- **Output:** revised category set, versioned system document

### 3.6 Cadence summary

| Cadence | Trigger | Fields touched | Output |
|---|---|---|---|
| Daily | 06:30 AWST auto-run | Catalyst timeline; new entrants | Register + history |
| Weekly | Monday | Catalyst calendar; narrative strength | Calendar + resolution log |
| Monthly | 1st business day | HF interest; analyst sentiment; risk tier | Positioning + tier log |
| Quarterly | Post-earnings | Lifecycle stage; theme alignment; retirements | Register + self-assessment |
| Annual | January | Category set; field definitions | Versioned system doc |

---

## 4. Output format — markdown register

The canonical table. One row per entry.

| Company / theme | Category | Narrative | Analyst | HF | Catalyst timeline | Risk tier | Volatility | Theme alignment |
|---|---|---|---|---|---|---|---|---|
| Power & thermal (VRT, ETN, TT) | AI Infrastructure | 5 | 4 | 4 | Q3 earnings, Aug–Sep 2026 | Low–Medium | Medium | Very high — binding constraint |
| Semi cap equipment (LRCX, AMAT, ASML) | AI Infrastructure | 4 | 4 | 5 | Q3 earnings; capex guidance | Low | Medium–High | High — total capacity exposure |
| Memory (MU) | AI Infrastructure | 4 | 4 | 5 | Quarterly; HBM pricing | Medium | High | High but cyclical |
| Grid-connected DC operators (NXT, MAQ) | AI Infrastructure | 4 | 4 | 2 | Contracted-utilisation updates | Medium | Medium | High — energisable sites |
| AI-native cloud (CRWV) | AI Infrastructure | 4 | 3 | 3 | Contract + debt announcements | Medium–High | Very high | Medium — concentration risk |
| Data layer (SNOW, MDB) | AI Software | 4 | 4 | 2 | Quarterly consumption metrics | Low–Medium | High | Very high — durable toll booth |
| Applied AI (PLTR) | AI Software | 5 | 4 | 3 | Q2 earnings; contract news | Medium | Very high | High — multiple is the risk |
| Agentic workflow (CRM, NOW, HUBS) | AI Software | 4 | 3 | 2 | Results — separate AI line item? | Low–Medium | Medium | High — pricing-shift test |
| Big-tech platforms (MSFT, GOOGL, AMZN) | AI Software | 4 | 5 | 4 | Quarterly capex + AI commentary | Low | Medium | Medium — diluted by size |
| "Software is under-owned" thesis | AI Software | 3 | 2 | 1 | Q3 13F filings, mid-Nov 2026 | Medium | High | Very high if it re-rates |
| Robotaxi commercialisation (UBER/LCID/Nuro) | EV / Autonomy | 5 | 4 | 3 | **Dated** — SF rollout later 2026 | Medium–High | High | Very high |
| Solid-state batteries (SLDP, QS) | Energy Storage | 4 | 2 | 2 | OEM qualification; EVE Dec 2026 | High | Very high | High — niche end market first |
| Uranium developers (BOE, PDN, DYL, BMN) | Energy Storage | 4 | 3 | 2 | Contracting; Citi US$100–125/lb | Medium | High | Very high — AI crossover |
| Grid-scale storage & generation (CEG, BE) | Energy Storage | 4 | 4 | 3 | PPA announcements | Low–Medium | Medium | Very high — dual narrative |
| Battery feedstock (ELV, LTR, PLS) | Energy Storage | 3 | 3 | 2 | Spodumene pricing; quarterlies | Medium | High | Medium — input exposure |
| AI drug discovery (RXRX + Vertex) | Biotech | 4 | 3 | 2 | Trial readouts — dated, binary | High | Very high | Very high — countable metric |
| Humanoid robotics exposure | Moonshots | 5 | 3 | 2 | Undated — platform milestones | High | Very high | High narrative, thin access |
| SMR / advanced nuclear | Moonshots | 3 | 2 | 2 | Undated — regulatory | High | Very high | High, 2030s revenue |
| Quantum pure-plays (IONQ, RGTI, QBTS) | Moonshots | 3 | 2 | 2 | Q2 earnings, 5–6 Aug 2026 | Medium | Very high | Medium — saturated |

---

## 4b. Output format — sentiment tables

Each agent produces three sentiment tables, in both Markdown and HTML. Markdown versions live in the
agent's own `.md` file; a combined clean-HTML version of all nine is in `tables.html`.

| Table | Columns | What it reports |
|---|---|---|
| **Speculative Picks** | Company, Sector, Why Investors Talk About It, Common Narrative, Risk Level | Retail and forum-level chatter — "investors often say…" |
| **Long-Hold Picks** | Company, Sector, Analyst Commentary, Strengths, Risk Level | Published, named analyst opinion — "analysts commonly highlight…" |
| **Solid Picks** | Company, Sector, Institutional Behaviour, Why Funds Accumulate, Risk Level | 13F and fund positioning — "hedge funds often accumulate…" |

**Rules**

- Every narrative or commentary cell is a **reported view with attribution**, phrased as what a group
  says. A bare assertion in one of these cells is a defect.
- **Risk Level** maps to the four-tier structural scale in §2.5 — assigned on balance sheet,
  liquidity and disclosure, never on price performance.
- "Speculative", "Long-hold" and "Solid" describe **how each group of market participants is
  behaving**, not a suggested holding period.
- HTML output must be semantic only: `<table>`, `<thead>`, `<tbody>`, `<tr>`, `<th>`, `<td>`. No
  inline styles, no scripts, no external dependencies. It inherits whatever styling the host page has.
- Read the three tables *against* each other. When Solid Picks fills with incumbents while pure-plays
  sit in Speculative, that gap is itself the finding — it is what the ~6% hedge-fund software weight
  looks like rendered as a table.

---

## 5. Output format — JSON schema

`schema.json` in this folder holds the machine-readable version. Structure:

```json
{
  "system": "speculation-tracking-system",
  "version": "1.0",
  "generated": "2026-08-01",
  "disclaimer": "Market research and educational analysis, not personalised financial advice.",
  "categories": [
    "AI Infrastructure", "AI Software", "Frontier Tech",
    "EV / Autonomy", "Energy Storage", "Biotech", "Moonshots"
  ],
  "entries": [
    {
      "id": "ai-infra-power-thermal",
      "name": "Power & thermal suppliers",
      "tickers": [
        { "exchange": "NYSE", "code": "VRT", "filings_url": "https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&ticker=VRT" }
      ],
      "category": "AI Infrastructure",
      "narrative_strength": 5,
      "analyst_sentiment": 4,
      "hedge_fund_interest": 4,
      "catalyst": {
        "next_event": "Q3 2026 earnings",
        "date": "2026-08-15",
        "dated": true,
        "type": "scheduled_binary"
      },
      "risk_tier": "low-medium",
      "volatility_profile": "medium",
      "theme_alignment": {
        "strength": "very high",
        "note": "Power and cooling is the binding constraint of the AI buildout"
      },
      "narratives": [
        {
          "claim": "The bottleneck has moved from chips to electrons",
          "attribution": "Morgan Stanley; BlackRock",
          "source_url": "https://www.morganstanley.com/insights/articles/powering-ai-energy-market-outlook-2026",
          "source_tier": "B"
        }
      ],
      "first_logged": "2026-08-01",
      "last_reviewed": "2026-08-01",
      "status": "active"
    }
  ]
}
```

### Field rules for automation

- `narrative_strength`, `analyst_sentiment`, `hedge_fund_interest` — integers 1–5
- `catalyst.dated` — boolean. If `false`, `catalyst.date` must be `null` and `next_event` must
  describe the event honestly rather than implying a timeframe
- `catalyst.type` — one of `scheduled_binary`, `scheduled_mechanical`, `policy`,
  `corporate_action`, `emergent`
- `risk_tier` — one of `low`, `low-medium`, `medium`, `medium-high`, `high`, `extreme`
- `volatility_profile` — one of `low`, `medium`, `high`, `very-high`
- `source_tier` — `A` primary, `B` reputable secondary, `C` sentiment only, `D` excluded.
  **Every narrative claim requires an `attribution` string.** An entry with `source_tier: "C"`
  supporting a factual claim is invalid and must fail validation
- `status` — `active`, `retired`, or `watch`
- `tickers[].filings_url` must point at an exchange announcements page or a regulator's filing
  system. **Broker, trading-platform and purchase URLs are invalid** and must fail validation

---

## 6. Known limitations

1. **No live market data.** No prices, market caps or volumes. Everything is narrative, catalyst and
   sentiment.
2. **Scores are judgements, not measurements.** They have never been backtested against forward
   returns. Their real use is comparative and longitudinal.
3. **13F data lags ~45 days** and shows quarter-end positions. It is a photograph of the past.
4. **Analyst sentiment is a summary of published opinion**, which carries its own conflicts —
   sponsored coverage, banking relationships, and a structural bias toward positive ratings.
5. **Coverage bias.** The system sees what is written about. Genuinely latent themes — the most
   valuable stage — are the hardest for it to catch.
6. **Category boundaries will age.** The AI Infrastructure/AI Software split is a 2025–26 distinction.
   The annual review exists to catch that.
7. **Not personalised.** The system knows nothing about anyone's finances, timeframe, tax position or
   risk capacity.

---

## Related files

- `ai_infrastructure.md` — AI Infrastructure Agent output
- `ai_software.md` — AI Software Agent output
- `frontier_tech.md` — Frontier Tech Agent output
- `schema.json` — machine-readable register
- `plan.md`, `spec.md`, `skills.md`, `history.md` — the underlying speculation-research system
- `index.html` — the live page

---

*Market research and educational analysis, not personalised financial advice.*