1a.   Scan PDFs (Explore agents for broad screening)
1b.   Extract insights (RLM for deep extraction - text, tables, AND charts)
1c.   Cross-doc synthesis (rlm-multi for insights across sources)
2.    Check KSVC holdings (preliminary - with known tickers)
3.    Write content (data backbone, Serenity-heavy)
4a.   AUDIT (verify draft claims against source PDFs with RLM)
4a.5. GEMINI CROSS-VALIDATION (web-verify FAIL/UNSOURCED inferences)
4b.   Final Holdings Verification (check ALL 7 models with discovered tickers)
4c.   Stylize (invoke kirk-mode skill for voice/character)
4d.   Humanize (remove AI patterns)
5.    Save draft for approval
6.    Chart decision & generation (after draft crystallizes thesis)
7.    PUBLISH to final folder (clean version for posting)

Never skip steps 4a-4d. Use 1a for multi-PDF screening, 1b for deep extraction, 1c for cross-doc synthesis, 4a for verification, 4a.5 for web cross-validation, 4b for final holdings check, 4c for character voice, 4d for AI pattern removal.

⚠️ CRITICAL: Step 1b extracts data. Step 1c synthesizes across docs. Step 4a VERIFIES the written content. Step 4a.5 CROSS-VALIDATES inferences.

1b: "What does each PDF say?" (per-doc extraction)
1c: "What patterns emerge across PDFs?" (cross-doc synthesis)
4a: "Does my draft accurately reflect the sources?" (source-locked verification)
4a.5: "Are the flagged inferences valid per public sources?" (web cross-validation)
4c: "Which Kirk mode fits this situation?" (character voice)

Subagent Permissions (CRITICAL)

Subagents CANNOT Read files outside the project directory. PDFs in /Users/Shared/ksvc/pdfs/ are blocked. The fix: symlink PDFs into the project directory before spawning subagents.

The main agent MUST create a symlink before Step 1a:

ln -sf "/Users/Shared/ksvc/pdfs/YYYYMMDD" ".claude/pdfs-scan"

Then subagents Read from .claude/pdfs-scan/filename.pdf — this works because the path resolves inside the project.

Access Method	`/Users/Shared/` path	Symlinked project path
Subagent Read tool (PDF)	❌ Auto-denied	✅ Works
Subagent Read tool (images)	❌ Auto-denied	✅ Works
Main agent Read tool	✅ User approves	✅ Works
Bash → RLM	✅ Any path	✅ Any path

Discovered 2026-02-07: Subagents fail with "Permission to use Read has been auto-denied (prompts unavailable)" on /Users/Shared/ paths. Symlink into project dir = full Read access. Tested: 19 PDFs, medium thoroughness, 125k tokens, zero errors.

Content Types & Voice Blends

Full guide: references/kirk-voice.md — Read this for templates and examples.

Kirk voice = Serenity's data + Citrini7's wit + Jukan's skepticism + Zephyr's energy.

Type	When	Blend	Key Element
Long Thread	Deep dive, multi-source	Serenity + Jukan	TLDR + skepticism
Quick Take	Single insight, one report	Citrini7 + Serenity	Punchy + one number
Breaking News	Just dropped	Zephyr + Jukan	Reaction word + number
Shitpost	Market absurdity	Citrini7 + Zephyr	Meme format
Personal Commentary	Opinion, question	Pure Jukan	First-person + uncertainty
Victory Lap	KSVC call worked	Pure Zephyr	Entry/Now + thesis

Quick Formulas

Long Thread: Hook → TLDR → Numbers → Skepticism → Position

Quick Take: Headline number → Context → "If you're looking now..."

Breaking News: "Huge." / "Well well well..." → Key number → Source

Victory Lap: "$TICKER up X% since KSVC added it" → Entry/Now → Thesis validated

Step 1a: Scan PDFs with Explore Agents

Use Explore agents for broad screening when you have many PDFs to review. This is faster than RLM for initial discovery.

Step 1a.0: Check Published Threads (MANDATORY - DO FIRST)

⚠️ Before scanning any PDFs, check what Kirk has already posted.

# List all published threads
ls /Users/Shared/ksvc/threads/

# Read recent thread.md files to understand what topics are covered

For each published thread, note:

Topic (what was the thesis?)
Source PDFs used (check _metadata.md)
Date (how recent?)

Then when selecting a topic after scanning, REJECT any topic that:

Uses the same primary source PDF as a published thread
Covers the same thesis/angle (even if from different sources)
Would read as a repeat to Kirk's followers

Acceptable overlap:

A follow-up/update to a previous thread with NEW data (e.g., earnings confirm the thesis)
A different angle on the same sector (e.g., posted about ABF shortage, now posting about specific company earnings)
Explicitly framed as "update: here's what changed since my last post on X"

Why this exists (Case Study — ABF Substrate, 2026-02-07):

Kirk published a 10-tweet thread on Feb 5 covering Goldman's ABF shortage report (10%→21%→42%, Kinsus/NYPCB/Unimicron). On Feb 7, the pipeline picked the same Goldman report and produced a 3-tweet quick take with the same numbers, same companies, same angle. We didn't check published threads first, so we wasted a pipeline run on duplicate content when 10 other fresh topic angles were available.

When to Use

Screening 10+ PDFs to find relevant ones
Finding cross-document connections
Building a thesis from multiple sources
Don't know which PDFs matter yet

How to Scan

1. Check published threads (Step 1a.0 above)

2. List recent PDF folders and count PDFs
   ls /Users/Shared/ksvc/pdfs/ | tail -5
   ls /Users/Shared/ksvc/pdfs/YYYYMMDD/ | wc -l

3. Symlink PDFs into project directory (REQUIRED for subagent access)
   ln -sf "/Users/Shared/ksvc/pdfs/YYYYMMDD" ".claude/pdfs-scan"

4. Split PDFs into groups and spawn parallel Explore agents
   TARGET: ~5 PDFs per agent. Spawn ALL agents in a single message.
   - Each agent gets a specific list of filenames to scan
   - All agents run simultaneously → total time = slowest agent
   - Haiku is cheap — more agents = faster with no meaningful cost increase

Agent Sizing

PDFs	Agents	PDFs/Agent	Expected Time
≤5	1	all	~25s
6-10	2	~5 each	~25s
11-15	3	~5 each	~25s
16-20	4	~5 each	~25s
21-30	5-6	~5 each	~30s

Why ~5 PDFs per agent? Sweet spot for speed. Each PDF takes ~4-8s to Read + summarize. 5 PDFs ≈ 25s per agent. Adding more PDFs per agent saves nothing (same total tokens) but makes wall-clock time worse.

Cost: Haiku is cheap. 4 agents × 5 PDFs × ~4k tokens = ~80k input tokens total — same as 1 agent doing all 20. Parallelism is free.

Cross-doc synthesis trade-off: Each agent only sees its batch, so cross-batch themes are the main agent's job. This is fine — the main agent merges all results anyway.

Example: Spawn Explore Agents

Step 1: Main agent creates symlink and lists PDFs:

ln -sf "/Users/Shared/ksvc/pdfs/20260205" ".claude/pdfs-scan"
/bin/ls ".claude/pdfs-scan/"

Step 2: Split filenames into groups and spawn agents in parallel (single message, multiple Task calls):

# Agent 1 — first batch
Task(subagent_type="Explore", prompt="""
**THOROUGHNESS: medium**

Scan these specific PDFs for content angles:
- file1.pdf
- file2.pdf
- file3.pdf
- file4.pdf
- file5.pdf
- file6.pdf
- file7.pdf

For each PDF, Read enough pages to understand the full thesis (use judgment — some need 1-2 pages, others 1-5):

Read(file_path="/Users/dydo/Documents/agent/ksvc-intern/.claude/pdfs-scan/FILENAME.pdf", pages="1-5")

For each PDF extract:
- Company/sector, ticker, rating, price target
- Key thesis and supporting numbers
- Supply chain connections
- Potential content angles

After scanning your batch, provide:
1. Per-PDF summary (2-3 sentences each)
2. Cross-document themes within your batch
3. Which PDFs are most relevant for deep extraction
""")

# Agent 2 — second batch (SPAWN IN SAME MESSAGE as Agent 1)
Task(subagent_type="Explore", prompt="""
... same prompt with file8.pdf through file14.pdf ...
""")

# Agent 3 — third batch (SPAWN IN SAME MESSAGE)
Task(subagent_type="Explore", prompt="""
... same prompt with file15.pdf through file20.pdf ...
""")

Step 3: Main agent synthesizes results from all agents: After all agents return, the main agent:

Merges per-PDF summaries
Identifies cross-agent themes (patterns Agent 1 found + patterns Agent 2 found)
Picks top 3 content angles across all PDFs
Selects 2-5 PDFs for Step 1b deep extraction

Output: Identify Which PDFs Matter

After scanning, you'll know:

Which reports have the best data
Cross-document connections (e.g., "3 reports confirm memory shortage")
Thesis recommendations (2-3 angles to explore)
Which to deep-extract with RLM

⚠️ WARNING: Explore agents can hallucinate specific numbers. Treat all numbers from Explore summaries as "unverified claims" until RLM grep confirms them. Component counts, percentages, and market sizing are especially prone to errors.

Capacity (tested 2026-02-07): Single Explore agent (haiku) handled 19 PDFs at medium thoroughness in 83 seconds, using 125k tokens (~4k tokens/PDF for pages 1-5). 3 agents in parallel = ~30-40s for the same batch.

Step 1b: Deep Extract with RLM

Use RLM for deep extraction from specific PDFs you've identified in Step 1a.

MANDATORY for any number you'll publish. Explore agents summarize; RLM verifies.

When to Use

You know which 2-5 PDFs matter most
Need specific numbers, charts, tables
Building cross-document verification tables
Extracting technical details (fabs, yields, WPM)

Single PDF

cd ~/.claude/skills/rlm-repl/scripts
python3 rlm_repl.py init "/Users/Shared/ksvc/pdfs/YYYYMMDD/file.pdf" --extract-images
python3 rlm_repl.py exec -c "print(grep('revenue|growth|target|price', max_matches=20, window=200))"

Multiple PDFs (synthesis)

cd ~/.claude/skills/rlm-repl-multi/scripts
python3 rlm_repl.py init "/path/to/report1.pdf" --name report1 --extract-images
python3 rlm_repl.py init "/path/to/report2.pdf" --name report2 --extract-images
python3 rlm_repl.py exec -c "results = grep_all('keyword', max_matches_per_context=20)"

View Extracted Charts/Images

# List images from a context
python3 rlm_repl.py exec --name report1 -c "print(list_images())"

# Get image path, then use Read tool to view
python3 rlm_repl.py exec --name report1 -c "print(get_image(0))"

Charts often contain key data (P/B trends, margin history, capacity timelines) that text extraction misses.

Extraction Validation (MANDATORY)

⚠️ After EVERY rlm_repl.py init, validate the extraction actually worked.

RLM reports chars_extracted after init. A multi-page analyst report should yield thousands of chars. If you get suspiciously few, the PDF is likely image-based and RLM only extracted metadata/headers.

Validation rule:

Chars Extracted	Expected Report Type	Action
> 5,000	Multi-page report	✅ Proceed with grep
1,000 - 5,000	Short note / partial	⚠️ Check `list_images()` — if many images, trigger fallback
< 1,000	Image-based PDF	❌ MUST use Read tool fallback

The threshold is context-dependent. A 20-page Goldman Sachs report yielding 666 chars is obviously broken. A 1-page pricing table yielding 800 chars might be fine. Use judgment, but when in doubt, fallback.

Mandatory Fallback when RLM extraction is low:

# Step 1: RLM init (always try first)
python3 rlm_repl.py init "/path/to/report.pdf" --extract-images
# Output: "Extracted 666 chars from 15 pages, saved 9 images"

# Step 2: Check - is 666 chars enough for a 15-page report? NO.
# → Trigger fallback

# Step 3: Check extracted images first (they may contain the data)
python3 rlm_repl.py exec -c "print(list_images())"
# View extracted images with Read tool
# Read(file_path="/path/to/extracted/image-0.png")

# Step 4: Read the PDF directly (use symlinked path for subagents)
# Read(file_path=".claude/pdfs-scan/report.pdf", pages="1-10")
# Read(file_path=".claude/pdfs-scan/report.pdf", pages="11-20")

⚠️ Path rule: Subagents must Read PDFs via the symlinked project path (.claude/pdfs-scan/), NOT from /Users/Shared/. See "Subagent Permissions" section above.

Why this exists (Case Study — ABF Substrate Shortage, 2026-02-07):

Goldman Sachs published two reports: a main ABF upcycle report (71K chars, extracted fine) and a Kinsus upgrade report (15 pages, but only 666 chars extracted). We skipped the Kinsus PDF because "the main report had everything we needed." It didn't. The Kinsus report had unique data (company-specific capacity plans, margin guidance, order book details) that would have strengthened the thread. Skipping it was lazy — the Read tool fallback takes 30 seconds and would have recovered the data.

Rules:

Never skip a relevant PDF just because RLM extraction was low. Use the fallback.
Check extracted images. RLM with --extract-images often saves chart/table images even when text extraction fails. View them with Read tool.
Log the fallback. In the extraction cache, note "extraction_method": "read_fallback" so audit knows the data source.
If fallback also fails (corrupted PDF, DRM), document it and move on. But you must TRY.

RLM Cache: Include Visual Data

When extracting, capture all data types for potential chart generation later:

Source Type	What to Extract	Cache Format
Text numbers	Exact quotes with page refs	`{"value": 5.3, "unit": "B", "source": "p.3", "quote": "規模約53億美元"}`
Tables	Full table as structured JSON	`{"columns": [...], "rows": [...], "source": "p.20"}`
Charts	Data points + source image path	`{"data": {...}, "source_image": "pdf-3-1.png", "page": 3}`

Why cache visual data? Step 6 (chart generation) needs this. If you only cache text, you'll lose table structures and chart data points that make great visualizations.

Cross-Document Reasoning

Build thesis by triangulating claims across multiple reports:

# Find where multiple reports discuss the same topic
python3 rlm_repl.py exec -c "results = grep_all('DRAM.*price|ASP', max_matches_per_context=5)"

# Compare forecasts across sources
python3 rlm_repl.py exec -c "results = grep_all('2026|2027|growth|demand', max_matches_per_context=5)"

Use cross-doc to verify:

Do multiple sources agree on price forecasts?
Are supply constraint timelines consistent?
Any contradictions between reports?

Step 1b.5: Build Extraction Cache (MANDATORY)

⚠️ Why this step exists: RLM creates state.pkl during extraction, but the writing phase (Step 3) doesn't access it. Without a persistent cache, writers rely on memory, leading to errors like wrong product types, missing time periods, or source attribution mistakes.

What this does: Extracts from state.pkl (RLM's internal format) into structured JSON with context labels that the writing phase can reference.

When to Run

After Step 1b (RLM extraction) and before Step 3 (writing).

Workflow	When to Cache
Single PDF (rlm-repl)	After `rlm_repl.py init` completes
Multiple PDFs (rlm-repl-multi)	After all `init` commands complete

How to Build Cache

New in v2: Auto-generates source tags and attribution map from PDF filenames!

Single PDF (rlm-repl):

cd ~/.claude/skills/kirk-content-pipeline/scripts

# Auto-extracts from default rlm-repl state location
python3 build_extraction_cache.py \
  --output /path/to/draft-assets/rlm-extraction-cache.json

Multiple PDFs (rlm-repl-multi):

cd ~/.claude/skills/kirk-content-pipeline/scripts

# Use --multi flag to load from rlm-repl-multi state
python3 build_extraction_cache.py \
  --multi \
  --output /path/to/draft-assets/rlm-extraction-cache.json

With Cross-Doc Synthesis (Optional):

# Add manual synthesis descriptions for cross-doc insights
python3 build_extraction_cache.py \
  --multi \
  --output /path/to/draft-assets/rlm-extraction-cache.json \
  --synthesis /path/to/cross-doc-synthesis.json

Synthesis format (optional, for complex multi-source threads):

{
  "dual_squeeze_thesis": {
    "description": "Memory shortage (1Q26) + ABF substrate shortage (2H26) = compounding AI server bottleneck",
    "components": [
      {"topic": "Memory Pricing", "source": "gfhk_memory", "timeframe": "1Q26"},
      {"topic": "Abf Shortage", "source": "goldman_abf", "timeframe": "2H26-2028"}
    ]
  }
}

What auto-generates:

✅ Source tags from PDF filenames ("GFHK - Memory.pdf" → tag: "GFHK")
✅ Topics with primary_source, key_metrics, source_context
✅ Extraction entries with full context labels (product_type, time_period, units, scope)

Cache Format

The cache includes context labels and attribution map to prevent common errors:

{
  "cache_version": "1.0",
  "generated_at": "2026-02-05T14:00:00",
  "sources": [
    {
      "source_id": "gfhk_memory",
      "pdf_path": "/Users/Shared/ksvc/pdfs/20260204/GFHK - Memory.pdf",
      "pdf_name": "GFHK - Memory price impact.pdf",
      "tag": "GFHK",
      "chars_extracted": 13199
    },
    {
      "source_id": "goldman_abf",
      "pdf_path": "/Users/Shared/ksvc/pdfs/20260204/Goldman ABF shortage.pdf",
      "pdf_name": "Goldman Sachs ABF shortage report.pdf",
      "tag": "Goldman Sachs",
      "chars_extracted": 25000
    }
  ],
  "extractions": [
    {
      "entry_id": "mem_001",
      "source_id": "gfhk_memory",
      "figure": "Figure 2",
      "page": 3,
      "metric": "Total BOM",
      "product_type": "HGX B300 8-GPU server",
      "time_period": "3Q25 → 1Q26E",
      "units": "dollars per server",
      "scope": "single HGX B300 8-GPU server",
      "values": {
        "before": "$369k",
        "after": "$408k",
        "change": "+$39k"
      },
      "context": "Memory price impact on AI server BOM",
      "source_quote": "Figure 2: HGX B300 8-GPU server BOM...",
      "verification": "RLM grep + visual inspection"
    }
  ],
  "source_attribution_map": {
    "topics": {
      "Memory Pricing": {
        "primary_source": "gfhk_memory",
        "tag": "GFHK",
        "key_metrics": ["HBM3e ASP", "DDR5-6400 (128GB)", "NVMe SSD (3.84TB)", "Total BOM"],
        "source_context": "Figures: Figure 2; Time periods: 3Q25 → 1Q26E",
        "notes": "4 extractions from this source"
      },
      "Abf Shortage": {
        "primary_source": "goldman_abf",
        "tag": "Goldman Sachs",
        "key_metrics": ["ABF shortage ratio", "Kinsus PT", "NYPCB PT", "Unimicron PT"],
        "source_context": "Time periods: 2H26, 2027, 2028",
        "notes": "5 extractions from this source"
      }
    },
    "cross_doc_synthesis": {
      "dual_squeeze_thesis": {
        "description": "Memory shortage (1Q26) + ABF substrate shortage (2H26) = compounding AI server bottleneck",
        "components": [
          {"topic": "Memory Pricing", "source": "gfhk_memory", "timeframe": "1Q26"},
          {"topic": "Abf Shortage", "source": "goldman_abf", "timeframe": "2H26-2028"}
        ]
      }
    }
  }
}

Key fields that prevent errors:

product_type: Prevents "GB300 rack" when source says "HGX B300 server"
time_period: Prevents missing "3Q25 → 1Q26E" context
source_id: Prevents "Goldman's BOM" when data is from GFHK
tag: Auto-extracted from PDF filename for quick attribution
units: Prevents "22.5B racks" when source means "22.5bn dollars"
scope: Prevents "per rack" when source means "per server"

Attribution map benefits:

topics: Topic-level mapping showing which source is primary authority
key_metrics: Quick lookup of what each source covers
source_context: Summary of figures, time periods covered
cross_doc_synthesis: Manual insights connecting multiple sources

Integration with Step 3 (Writing)

MANDATORY: Reference the cache when writing.

Step 3a: Load cache and attribution map:

cache = load_json('rlm-extraction-cache.json')
attr_map = cache['source_attribution_map']

# Get topic attribution
topic = "Memory Pricing"
source_tag = attr_map['topics'][topic]['tag']  # "GFHK"
key_metrics = attr_map['topics'][topic]['key_metrics']

Step 3b: Write using cache labels and attribution:

## Content

3/ Memory squeeze is already here. GFHK's BOM breakdown (3Q25 → 1Q26E):
- HBM3e ASP: $3,756 → $4,378 (+17%)
- DDR5-6400 (128GB): $563 → $1,920 (+241%)
- HGX B300 8-GPU server BOM: $369k → $408k

Source: rlm-extraction-cache.json, entry mem_001, mem_002, mem_003

Context labels from cache:

Product type: HGX B300 8-GPU server (not GB300 rack)
Time period: 3Q25 → 1Q26E (quarterly change)
Source: GFHK Figure 2 (via attribution map tag)

Attribution map usage:

Used topics["Memory Pricing"]["tag"] → "GFHK"
Verified metrics against key_metrics list
Cross-doc synthesis: See dual_squeeze_thesis for memory + ABF connection

Enforcement

Before saving draft (Step 5), verify:

Every published number has a cache entry
Product types match cache labels
Time periods included from cache
Source attributions match cache source_id and attribution map tag
Units match cache (dollars vs racks, per server vs per datacenter)
Cross-doc claims reference cross_doc_synthesis if applicable

Red flags - stop if you notice:

Writing numbers from memory instead of cache
Product type differs from cache (product_type field)
Missing time period when cache has time_period
Attributing to wrong source vs cache source_id
Using wrong tag (e.g., "Goldman" for GFHK data)
Missing cross-doc synthesis when connecting multiple sources

Manual Cache Building

If automatic extraction fails, manually create cache entries:

{
  "entry_id": "manual_001",
  "source_id": "report_name",
  "metric": "Component count",
  "product_type": "Humanoid robot (dexterous hand)",
  "values": {"count": 22},
  "units": "DOF (degrees of freedom)",
  "context": "Dexterous hand articulation",
  "source_quote": "22自由度靈巧手",
  "verification": "Manual extraction from p.15",
  "notes": "Summed from finger joints (20) + wrist (2)"
}

See: ~/.claude/skills/kirk-content-pipeline/scripts/README-extraction-cache.md for full documentation.

Step 1c: Cross-Doc Synthesis (RECOMMENDED)

Why this step exists: Steps 1a and 1b produce per-document facts. Without explicit synthesis, the pipeline gravitates toward single-source claims ("KHGEARS P/E is 20x") rather than cross-doc insights ("Taiwan brokers are more bullish than Western analysts on humanoid robotics").

When to Use

Scenario	Use 1c?
Multiple PDFs on same topic	Yes
Comparing broker views	Yes
Finding consensus/disagreement	Yes
Single PDF deep dive	No (skip to Step 2)
Breaking news (speed matters)	No (skip to Step 2)

What 1c Produces

Output Type	Example	Audit Requirement
Consensus claim	"3 of 4 brokers see DRAM ASP rising in 2H26"	Cross-doc (rlm-multi)
Comparative insight	"HIWIN at 38x vs KHGEARS at 20x - market pricing in certainty"	Cross-doc (rlm-multi)
Disagreement flag	"MS says neutral, local brokers say buy - who's right?"	Cross-doc (rlm-multi)
Synthesized thesis	"Taiwan supply chain undervalued vs China peers"	Cross-doc (rlm-multi)

How to Run Cross-Doc Synthesis

cd ~/.claude/skills/rlm-repl-multi/scripts

# Initialize all relevant PDFs
python3 rlm_repl.py init "/path/to/broker1.pdf" --name broker1
python3 rlm_repl.py init "/path/to/broker2.pdf" --name broker2
python3 rlm_repl.py init "/path/to/broker3.pdf" --name broker3

# Ask synthesis questions (not just extraction)
python3 rlm_repl.py exec -c "
# Question 1: Do they agree on market sizing?
market_data = grep_all('market size|TAM|規模|billion|億', max_matches_per_context=10)
print('=== MARKET SIZE ACROSS SOURCES ===')
print(market_data)
"

python3 rlm_repl.py exec -c "
# Question 2: Compare recommendations
ratings = grep_all('BUY|SELL|NEUTRAL|買進|賣出|中立|rating|recommendation', max_matches_per_context=10)
print('=== RATINGS COMPARISON ===')
print(ratings)
"

python3 rlm_repl.py exec -c "
# Question 3: Find disagreements
pe_data = grep_all('P/E|PE|本益比|target price|目標價', max_matches_per_context=10)
print('=== VALUATION COMPARISON ===')
print(pe_data)
"

Synthesis Questions to Ask

Category	Questions
Consensus	Do sources agree on [market size / timeline / key risk]?
Comparison	How does [broker A] view differ from [broker B]?
Valuation	Are local vs foreign analysts pricing the same?
Timeline	Do sources agree on [catalyst / inflection point]?
Risk	What risks does one source mention that others miss?

Output Format: Synthesis Cache

After running 1c, document synthesized insights for Step 3 (writing):

## Cross-Doc Synthesis (Step 1c)

**Sources:** broker1 (永豐), broker2 (MS), broker3 (Citi)

### Consensus
- Market size: All 3 agree on $5-6B (2025) → $30-35B (2029)
- CAGR: 55-60% range across all sources

### Disagreements
- HIWIN: MS says NEUTRAL (38x too rich), 永豐 silent, Citi no coverage
- Timeline: 永豐 more bullish on 2026 ramp, MS cautious until 2027

### Comparative Insights (use in thread)
- "Taiwan brokers (永豐) bullish on KHGEARS; Western analysts (MS) more cautious on HIWIN"
- "Local coverage sees 2026 inflection; foreign houses waiting for 2027 proof points"

### Audit Flag
These synthesized claims require cross-doc verification in Step 4b:
- [ ] "3 sources agree on market size" → verify all 3 sources
- [ ] "Local vs foreign view divergence" → verify specific ratings from each

Integration with Audit (Step 4a)

⚠️ CRITICAL: Synthesized claims from Step 1c MUST be flagged for cross-doc audit in Step 4a.

In the audit manifest, mark these claims with cross-doc: true:

## Claims to Verify

| # | Claim | Type | Source ID | Cross-Doc? |
|---|-------|------|-----------|------------|
| 1 | KHGEARS P/E 20x | P/E | src1 | No |
| 2 | Market consensus $5.3B | Consensus | src1, src2, src3 | **Yes** |
| 3 | Local vs foreign view divergence | Synthesis | src1, src2 | **Yes** |

Cross-doc claims use rlm-repl-multi for verification, not parallel single-doc agents.

Extract with Technical Specificity

Go beyond surface numbers. Extract:

Wafer capacity (WPM)
Fab names (M15X, P4L, X2)
Yield percentages
Process nodes (1b, 1c)
Component counts per unit

Question	Extract
What	One-sentence summary
Why	Why readers should care
Who	Companies/tickers affected
When	Timeline (specific quarters)
Where	Fab locations, geography
How	Mechanism with technical detail

Step 2: Check KSVC Holdings (Initial)

⚠️ CRITICAL: This is a preliminary check. You MUST run Step 4c (Final Holdings Verification) after writing content to catch any tickers discovered during extraction.

All Models (7 Total)

US Models: usa-model1 ~ usa-model5 (5 models)
Taiwan (TWSE) Models: twse-model1 ~ twse-model2 (2 models)

Step 2a: Identify All Possible Tickers

Before querying the API, identify ALL possible identifiers for the company:

# Example: Global Unichip Corp
# Identifiers to search:
# - US ticker: N/A (not US-listed)
# - Taiwan ticker: 3443
# - Chinese name: 創意 or 全球晶圓科技
# - English name: Global Unichip, GUC
# - Stock code: 3443 TW (TWSE format)

# For Taiwan stocks, verify ticker via TWSE API first:
curl -s "https://www.twse.com.tw/en/api/codeQuery?query=3443"
# Returns: {"query":"3443","suggestions":["3443\tGUC"]}

Rules:

US stocks: Search by ticker only (e.g., "MU", "AMD", "NVDA")
Taiwan stocks: Search by stock code (e.g., "3443") - may appear as "3443 創意" in API
If unsure: Check both US and TWSE models

Step 2b: Query All 7 Models

NEVER assume a stock isn't held without checking ALL 7 models.

RECOMMENDED: Use tradebook for accurate entry prices and current status

# FASTEST METHOD: Check tradebook for entry price + status
# (Works for all models - US and TWSE)
curl -s "https://kicksvc.online/api/twse-model2" | \
  jq '.tradebook[] | select(.ticker == 6285 or .ticker == 3491) |
  {ticker, enterDate, enterPrice, todayPrice, profitPercent, exitDate}'

# Returns:
# {
#   "ticker": 6285,
#   "enterDate": "Wed, 28 Jan 2026 00:00:00 GMT",
#   "enterPrice": 162.0,
#   "todayPrice": 207.5,  # ⚠️ May be stale! Use Yahoo Finance for current
#   "profitPercent": 28.09,  # ⚠️ Based on stale todayPrice
#   "exitDate": null  # null = still holding
# }

⚠️ CRITICAL: API's todayPrice and profitPercent can be STALE (hours or days old). Always verify current price with Yahoo Finance API (Step 2d).

FALLBACK: Check equitySeries (slower, less data)

# Check ALL 5 US models
for i in 1 2 3 4 5; do
  echo "=== USA-Model $i ==="
  curl -s "https://kicksvc.online/api/usa-model$i" | \
    jq --arg t "MU" '.equitySeries[0].series[] | select(.Ticker == $t) |
    {ticker: .Ticker, return: .data[-1].value}'
done

# Check ALL 2 TWSE models (search by stock code)
for i in 1 2; do
  echo "=== TWSE-Model $i ==="
  curl -s "https://kicksvc.online/api/twse-model$i" | \
    jq '.equitySeries[0].series[] | select(.Ticker | contains("3443")) |
    {ticker: .Ticker, return: .data[-1].value}'
done

Why still use equitySeries?

Historical tracking: Shows return % evolution over time (.data[] array)
Verification: Confirms position is still active
Fallback: If tradebook is unavailable or empty
Entry date discovery: First data point (return ≈ 0) indicates entry date

Example: Finding entry date from equitySeries

# Get all data points to find entry date
curl -s "https://kicksvc.online/api/twse-model2" | \
  jq '.equitySeries[0].series[] | select(.Ticker | contains("6285")) | .data[0]'
# Returns: {"date": "2026-01-28 00:00:00", "value": 0}
# Entry date: Jan 28, 2026

Step 2c: Verification and Fallback Strategy

Use all three data sources for robustness:

Data Source	When to Use	What It Shows	Limitation
tradebook	Primary	Entry date, entry price, exit status	`todayPrice` may be stale
equitySeries	Verification	Return % over time, position status	No entry price/date
filledOrders	Fallback	Actual trade orders, prices	Empty if model didn't reset recently

Recommended workflow:

# 1. PRIMARY: Get entry details from tradebook
TRADEBOOK=$(curl -s "https://kicksvc.online/api/twse-model2" | \
  jq '.tradebook[] | select(.ticker == 6285)')

# 2. VERIFY: Cross-check with equitySeries
EQUITY=$(curl -s "https://kicksvc.online/api/twse-model2" | \
  jq '.equitySeries[0].series[] | select(.Ticker | contains("6285"))')

# 3. FALLBACK: If tradebook empty, check filledOrders
if [ -z "$TRADEBOOK" ]; then
  curl -s "https://kicksvc.online/api/twse-model2" | \
    jq '.filledOrders[] | select(.ticker | contains("6285"))'
fi

Cross-verification example:

# Check if tradebook and equitySeries agree on position status
TRADEBOOK_HELD=$(curl -s "https://kicksvc.online/api/twse-model2" | \
  jq '.tradebook[] | select(.ticker == 6285 and .exitDate == null) | .ticker')

EQUITY_HELD=$(curl -s "https://kicksvc.online/api/twse-model2" | \
  jq '.equitySeries[0].series[] | select(.Ticker | contains("6285")) | .Ticker')

# If both show position, high confidence
# If only one shows position, investigate discrepancy

Fallback: Check filledOrders (if tradebook empty)

If equitySeries is empty OR tradebook is empty (rare, but possible after model reset):

# Check ALL US models - filledOrders
for i in 1 2 3 4 5; do
  echo "=== USA-Model $i filledOrders ==="
  curl -s "https://kicksvc.online/api/usa-model$i" | \
    jq '.filledOrders[] | select(.ticker == "MU") | {ticker, price, quantity}'
done

# Check ALL TWSE models - filledOrders
for i in 1 2; do
  echo "=== TWSE-Model $i filledOrders ==="
  curl -s "https://kicksvc.online/api/twse-model$i" | \
    jq '.filledOrders[] | select(.ticker | contains("3443")) | {ticker, price, quantity}'
done

When data sources disagree:

Scenario	Action
tradebook shows position, equitySeries doesn't	Trust tradebook (equitySeries may lag)
equitySeries shows position, tradebook doesn't	Investigate - check filledOrders
filledOrders shows buy but no current position	Position was closed - check tradebook.exitDate
All three empty	Position not held in this model

Step 2e: Document Holdings with Accurate Returns

CRITICAL: Always calculate actual returns using:

Entry price from tradebook.enterPrice
Current price from Yahoo Finance API (NOT KSVC API's stale todayPrice)

Output format (with accurate data):

**KSVC Holdings Check:**
- ✅ WNC (6285.TW) - Held in TWSE Model 2
  - Entry: Jan 28, 2026 @ NT$162
  - Current: NT$187 (Yahoo Finance)
  - Gain: +15.4% (actual, not API's stale 28%)
- ✅ UMT (3491.TWO) - Held in TWSE Model 2
  - Entry: Jan 28, 2026 @ NT$1,120
  - Current: NT$1,280 (Yahoo Finance)
  - Gain: +14.3% (actual, not API's stale 23%)
- ❌ Not held in TWSE Model 1 or USA Models 1-5

**Note:** API's equitySeries and tradebook.todayPrice can lag hours/days behind market.
Always use Yahoo Finance for current prices.

If NOT held in any model:

**KSVC Holdings Check:**
- ❌ Not held in any of 7 models (checked USA 1-5, TWSE 1-2)
- Content angle: Industry analysis / Market observation

Integration Strategies

Situation	Approach	Example
Held (US)	Call out position	"KSVC Model1 holds $MU at $412 entry"
Held (TW)	Call out position	"KSVC台股Model1持有台積電 (2330)"
Not held	Industry framing	"Memory cycle benefits $MU, SK Hynix"
Win	Victory lap	"$MU +15% since Model1 added it"

Step 2d: Current Price Check (Yahoo Finance API - REQUIRED)

⚠️ CRITICAL: ALWAYS use Yahoo Finance for current prices. KSVC API's todayPrice can be stale.

US stocks:

# Get current price
TICKER="MU"
curl -s -A "Mozilla/5.0" "https://query1.finance.yahoo.com/v8/finance/chart/$TICKER?interval=1d&range=1d" | \
  jq '.chart.result[0].meta.regularMarketPrice'

# Get full market data
curl -s -A "Mozilla/5.0" "https://query1.finance.yahoo.com/v8/finance/chart/$TICKER?interval=1d&range=1d" | \
  jq '.chart.result[0].meta | {symbol, regularMarketPrice, currency, regularMarketTime}'

Taiwan stocks (use .TW or .TWO suffix):

# WNC (6285.TW)
curl -s -A "Mozilla/5.0" "https://query1.finance.yahoo.com/v8/finance/chart/6285.TW?interval=1d&range=1d" | \
  jq '.chart.result[0].meta | {symbol, regularMarketPrice, currency, regularMarketTime}'

# UMT (3491.TWO - OTC stocks use .TWO)
curl -s -A "Mozilla/5.0" "https://query1.finance.yahoo.com/v8/finance/chart/3491.TWO?interval=1d&range=1d" | \
  jq '.chart.result[0].meta | {symbol, regularMarketPrice, currency, regularMarketTime}'

Taiwan ticker suffixes:

.TW - Listed on Taiwan Stock Exchange (TWSE)
.TWO - Listed on Taipei Exchange (TPEx/OTC)

Calculate actual gain (not API's stale profit%):

# Example: WNC
TICKER="6285.TW"
ENTRY=162  # From tradebook.enterPrice
CURRENT=$(curl -s -A "Mozilla/5.0" "https://query1.finance.yahoo.com/v8/finance/chart/$TICKER?interval=1d&range=1d" | jq '.chart.result[0].meta.regularMarketPrice')
echo "$TICKER: NT\$$CURRENT | Entry: NT\$$ENTRY | Gain: $(awk "BEGIN {printf \"%.1f\", ($CURRENT - $ENTRY) / $ENTRY * 100}")%"

# Output: 6285.TW: NT$187 | Entry: NT$162 | Gain: +15.4%

Complete workflow (tradebook + Yahoo Finance):

# 1. Get entry price from tradebook
ENTRY=$(curl -s "https://kicksvc.online/api/twse-model2" | jq '.tradebook[] | select(.ticker == 6285) | .enterPrice')

# 2. Get current price from Yahoo Finance
CURRENT=$(curl -s -A "Mozilla/5.0" "https://query1.finance.yahoo.com/v8/finance/chart/6285.TW?interval=1d&range=1d" | jq '.chart.result[0].meta.regularMarketPrice')

# 3. Calculate actual gain
echo "Entry: NT\$$ENTRY | Current: NT\$$CURRENT | Gain: $(awk "BEGIN {printf \"%.1f\", ($CURRENT - $ENTRY) / $ENTRY * 100}")%"

Step 3: Write Content

See references/kirk-voice.md for full templates and examples.

Thread Numbering Convention

Format	When to Use
No number on Tweet 1	Recommended - cleaner hook, stands alone if quoted/shared
`2/`, `3/`, etc.	Standard thread format - signals "2 of N"
`1/` on first tweet	Optional - explicit "thread incoming" signal

Why skip number on first tweet:

Hook tweet often gets shared standalone
"1/" makes it look incomplete out of context
Cleaner visual presentation

Format preference: Use / not ) - it's the established Twitter thread convention.

✅ Recommended:
Humanoid robots going from science fair to factory floor. Taiwan supply chain getting interesting.

2/ TLDR:
- Market: $5.3B (2025) to $32.4B (2029)...

❌ Avoid:
1/ Humanoid robots going from science fair...

Pick Content Type

What kind of content? (Thread / Quick Take / Breaking / Shitpost / Commentary / Victory Lap)
Look up the formula in kirk-voice.md
Apply the blend

Technical Specificity

❌ Vague: "NAND supply is tight"

✅ Specific: "YMTC adding 135k WPM at Wuhan Fab 3. Still won't close the gap - Samsung X2 conversion delayed to Q2."

❌ Vague: "HBM margins are good"

✅ Specific: "SK Hynix HBM yields at 80-90%. Samsung stuck at 60% on 1c DRAM."

Always include: specific numbers, time frames, fab names, comparisons.

Referential Clarity (Learned 2026-02-08)

Never use vague pronouns or shorthand when the referent hasn't been introduced.

In thread format, each tweet may be read semi-independently. If earlier tweets discuss a concept as a category (e.g., "ASIC revenue"), don't suddenly refer to it as "the project" in a later tweet — the reader has no antecedent for "the project."

❌ Vague: "MS thinks the project is the 3nm Google TPU" (What project? The thread never introduced "a project.")

✅ Clear: "MS thinks the main client/program is the 3nm Google TPU" (Names what MS is identifying — who's buying and what they're building.)

Rule: When a shorthand ("the project", "this deal", "the play") saves words but costs clarity, it's not saving anything. Name the thing directly. A few extra words that prevent the reader from pausing to re-read are always worth it.

When shifting from category to specific: If the thread discusses an abstract category (ASIC revenue, memory supply) and then pivots to a specific entity (Google TPU, Samsung fab), bridge the transition. Don't assume the reader already knows which specific thing drives the category.

Step 4a: Audit (MANDATORY — MUST USE SUBAGENTS)

⚠️ WHY THIS STEP EXISTS: We learned that RLM extraction (Step 1b) is not the same as verification. Explore agents hallucinate numbers. Writers make inferences. This step catches errors BEFORE publishing.

⚠️ STRUCTURAL GATE: You (the main agent) are the WRITER. You cannot also be the AUDITOR. You MUST delegate audit to fresh-context subagents. See the "WARM STATE TRAP" section in the audit-content skill for why.

Step 4a Process (3 actions, in order)

Action 1: Generate audit manifest

Write audit-manifest.md with all claims, sources, and search hints.
This is the handoff document for the audit agents.

Action 2: Spawn Explore agents (MANDATORY — do NOT skip this)

Spawn 1 Explore agent per source PDF via Task tool.
Each agent gets: the manifest + its assigned PDF path + claim list.
Each agent returns: JSON with PASS/FAIL/UNSOURCED per claim.

⚠️ WARM STATE TRAP: If RLM is already loaded from Step 1b, you WILL be tempted to "just grep it yourself." DO NOT. The audit-content skill explains why: you wrote the draft, so you already "know" the answers. Self-auditing is confirmation bias, not verification.

Self-check: If you are about to type rlm_repl.py exec during Step 4a, STOP. You are skipping the gate.

Action 3: Collect results and write audit report

Aggregate agent results into audit-report.md.
MUST include audit_agent_ids from the Task tool responses.
If audit_agent_ids is empty, the audit is invalid.

Invoke the audit-content skill for full process details:

/audit-content

What Gets Verified

Claim Type	Example	How to Verify
Company names	"KHGEARS"	RLM grep + TWSE API
Ticker formats	"4571 TW"	TWSE API
Numbers	"62 harmonic reducers"	RLM grep exact count
Percentages	"19% cost"	RLM grep in source
P/E ratios	"20x"	RLM grep analyst target
Ratings	"BUY"	RLM grep recommendation
Timelines	"2H27"	RLM grep + verify context
Attributions	"shipping to X"	Must be explicit in source, not inferred

When to Proceed

All PASS: Save draft (Step 5)
Any FAIL: Fix the claim, re-audit
UNSOURCED: Either remove, add caveat ("reportedly"), or find source

Do NOT save draft with FAIL status. UNSOURCED claims need explicit decision.

Step 4a.5: Gemini Web Cross-Validation (RECOMMENDED)

⚠️ WHY THIS STEP EXISTS: RLM audit (Step 4a) is source-locked — it only checks claims against the cited PDF. This over-flags reasonable inferences that go beyond one report but are well-documented publicly. Step 4a.5 gives flagged claims a second chance via web-grounded search.

Case Study (Old Memory Squeeze, 2026-02-07):

Draft said "capacity getting cannibalized for HBM and DDR5"
RLM audit FAIL: MS report says "exiting DDR4" / "cannibalization" but doesn't name HBM/DDR5 as destination
Gemini confirmed: TrendForce, DigiTimes, The Elec all document the DDR4→HBM/DDR5 shift
Result: Claim restored with dual attribution (MS + public sources)

Same thread: "Samsung, Kioxia, Micron all reducing MLC NAND" — MS only confirmed Samsung, said Kioxia/Micron "could" reduce. Gemini confirmed all three are actively reducing per TrendForce (41.7% YoY MLC NAND capacity decrease).

When to Use

RLM Audit Result	Use Gemini?	Why
FAIL — wrong number	No	Number errors need source correction, not web search
FAIL — inference beyond source	Yes	Inference may be valid per public sources
FAIL — misattribution	Maybe	Check if correct attribution exists publicly
UNSOURCED — claim not in cited PDF	Yes	Claim may be common industry knowledge
PASS	No	Already verified

How to Run

# For each FAIL/UNSOURCED claim that looks like a reasonable inference:
gemini -p "Search the web: [specific factual question about the inference].
I need external industry sources (TrendForce, DigiTimes, The Elec, Reuters,
company earnings calls) from 2025-2026 confirming or denying this."

Key: Ask Gemini to search the web explicitly. Without "search the web", Gemini may read local files instead.

Decision Matrix

Gemini Result	Action
Confirmed with public sources	Restore claim, add dual attribution (`source + Gemini web`)
Partially confirmed	Soften language to match what's confirmed
Not confirmed / contradicted	Keep as FAIL, fix or remove claim
Gemini unsure / no sources found	Keep as FAIL (conservative default)

Audit Report Format

Update claims restored via Gemini:

| # | Claim | Source | Status |
|---|-------|--------|--------|
| 5 | DDR4 capacity → HBM/DDR5 | ms_old_memory + Gemini web | PASS (restored) |
| 8 | All three reducing MLC NAND | ms_old_memory + Gemini web | PASS (restored) |

Include Gemini's sources in the audit resolution log:

### Claim 5 (RESTORED via Gemini)
- **RLM flagged:** Source doesn't name HBM/DDR5 as destination
- **Gemini confirmed:** TrendForce, DigiTimes, The Elec document capacity conversion
- **Resolution:** Restored with corrected framing

Guidelines

Only use for inferences and industry knowledge claims, NOT for number verification
Gemini's web search is a second opinion, not the final word — present both perspectives
If Gemini contradicts both the source and common sense, flag for human review
Keep Gemini queries specific and focused — one claim per query

Step 4b: Final Holdings Verification (MANDATORY)

⚠️ CRITICAL: This is the FINAL holdings check. You MUST run this after writing content because:

Tickers/stock codes may be discovered during extraction (Step 1b)
Company names may be clarified during audit (Step 4b)
Step 2 was a preliminary check with limited information

Why This Step Exists

Problem: You might learn the correct ticker late in the pipeline.

Example - GUC Case:

Step 1b: Learn company is "Global Unichip (GUC)"
Step 1b: Extract ticker "3443 TW" from report
Step 2: ❌ Assumed "not held" without actually checking TWSE models
Step 3: Wrote "I don't have a position here"
Step 4c: ✅ Discovered GUC IS held in TWSE Model 1 (+2.22%)
Result: Had to rewrite content to reflect actual position

Step 4c Process

1. Extract ALL tickers/identifiers from the draft:

# Read the draft and extract tickers
grep -E "[0-9]{4}|\\$[A-Z]{2,5}" draft.md

# Example output:
# - 3443 TW (Taiwan stock)
# - $MU (US stock)
# - AMD, NVDA (US stocks)

2. For EACH ticker, check ALL 7 models:

# Taiwan stock example: 3443
for i in 1 2; do
  echo "=== TWSE-Model $i ==="
  curl -s "https://kicksvc.online/api/twse-model$i" | \
    jq '.equitySeries[0].series[] | select(.Ticker | contains("3443")) |
    {ticker: .Ticker, return: .data[-1].value}'
done

# US stock example: MU
for i in 1 2 3 4 5; do
  echo "=== USA-Model $i ==="
  curl -s "https://kicksvc.online/api/usa-model$i" | \
    jq --arg t "MU" '.equitySeries[0].series[] | select(.Ticker == $t) |
    {ticker: .Ticker, return: .data[-1].value}'
done

3. Compare Step 2 vs Step 4c results:

**Holdings Verification:**

Step 2 (Initial): Claimed "Not held"
Step 4c (Final): ✅ Found in TWSE Model 1 (+2.22%)

**Action Required:** Update draft to reflect actual position

4. If holdings status changed, update draft:

# Before (Step 2):
"I don't have a position here, but watching..."

# After (Step 4c):
"KSVC holds GUC in TWSE Model 1 (+2.22% since entry). Watching..."

Decision Matrix

Step 2	Step 4c	Action
Not held	Not held	✅ No change needed
Not held	HELD	❌ UPDATE DRAFT - change content angle
Held	Held	✅ Verify return % is current
Held	Not held	❌ UPDATE DRAFT - position was closed

Output Format

**Step 4c: Final Holdings Verification**

✅ Verified ALL 7 models (USA 1-5, TWSE 1-2)

**Tickers checked:**
- 3443 (GUC): ✅ Found in TWSE Model 1 (+2.22%)
- $MU: ❌ Not held
- $AMD: ✅ Found in USA Model 3 (+12.5%)

**Changes required:**
- Update draft line 32: Add KSVC position note for GUC
- Update draft line 45: Add KSVC position note for AMD

Step 4c: Stylize (MANDATORY)

Why this step exists: The data backbone (Step 3) is Serenity-heavy - precise, comprehensive, verified facts. Step 4c transforms it into Kirk's authentic voice with emotional range and character.

Invoke the kirk-mode skill:

/kirk-mode

What Kirk Mode Does

Transforms verified data into Kirk's voice by:

Mode selection - Matches Kirk's emotional mode to situation (Analytical, Sarcastic, Emo, Shitpost, Degen, GIF Master)
Voice elements - Adds discovery moments ("Wait though"), reactions ("wayyy bigger"), first-person thesis
Meme culture - Integrates fintwit slang (ngmi, wagmi, brother, probably nothing) strategically
Anti-formula - Rotates structure to prevent templating (varies TLDR → "ok so" → question)
Credibility balance - Online enough to relate, credible enough to trust

When to Use Each Mode

Situation	Kirk Mode	Example
Deep fundamental dive	Analytical	"ok so", "Wait though", data-heavy with reactions
Market absurdity	Sarcastic	"brother Elon literally applied for 1M satellites"
Positions down	Emo	"honestly getting wrecked", vulnerable lowercase
Quick reaction/meme	Shitpost	me/also me format, nobody: format
High-conviction risky play	Degen	"sir this is a casino", YOLO energy
Victory lap	GIF Master + Analytical	Perfect GIFs + receipts

Most natural: Mix modes in single post (Analytical + Sarcastic + maybe GIF)

Workflow

Assess situation - What's happening? (Deep dive, absurd market, position down, quick reaction)
Select mode(s) - Use kirk-mode decision tree or mix modes naturally
Apply voice toolkit - Discovery moments, strategic "wayyy", emphasis markers
Check meme integration - Would slang/GIF enhance or distract from analysis?
Verify authenticity - Read aloud: sounds like intern at bar or ChatGPT report?

Output: Transformed content with Kirk's character voice - ready for humanizer pass.

See kirk-mode skill for:

Complete mode descriptions with examples
Meme vocabulary and format templates
Anti-formula principles
Credibility boundaries

Step 4d: Humanize (MANDATORY)

Note: Humanizer runs AFTER stylize to remove any AI patterns that slipped through during transformation.

Invoke the humanizer skill:

/humanizer

Patterns to Remove

Pattern	Fix
"Full stop."	"Simple as." or just delete
Em-dashes (—)	Periods, commas
"It's not X. It's Y."	"The play is Y, not X."
Perfect parallelism	Vary structure
Rule of three	Break the pattern
Over-confidence	Add skepticism phrase

AI Words to Remove

Additionally, crucial, delve, emphasize, testament, enhance, foster, landscape, showcase, tapestry, underscore, vibrant, pivotal, key (adj), interplay

Soul to Add

Skepticism: "I might be wrong" / "Not sure about this"
Reactions: "That number is wild" / "Interesting"
First person: "I keep thinking about..."
Mixed feelings: "Impressive but also kind of unsettling"
Questions: Ask the audience

Step 5: Save Draft

File Organization Convention

CRITICAL: Use assets folder structure for all drafts.

content-pipeline/draft/
└── YYYY-MM-DD-topic-assets/
    ├── README.md                           # Inventory, traceability, verification log
    ├── YYYY-MM-DD-topic.md                 # Original draft
    ├── YYYY-MM-DD-topic-citrini7.md        # Tone rewrites (if applicable)
    ├── YYYY-MM-DD-topic-audit-manifest.md  # Audit claims list
    ├── YYYY-MM-DD-topic-audit-report.md    # Audit verification results
    ├── YYYY-MM-DD-topic-audit-final.md     # Final audit with corrections
    ├── chart1_*.png                        # Generated charts
    ├── chart2_*.png
    └── source_*.png                        # Source images for traceability

Example: 2026-02-05-guc-valuation-debate-assets/

Draft Content Format

Save main content as: YYYY-MM-DD-topic-assets/YYYY-MM-DD-topic.md

# [Topic] [Type] Draft

**Date:** YYYY-MM-DD
**Source:** [Report name, date]
**Type:** Thread | Quick Take | Reaction
**Status:** PENDING APPROVAL
**Process:** RLM extraction → KSVC check → Humanizer pass

---

## Content

[Content here]

---

## Source Citations
- [List sources]

## Notes
- [KSVC holdings: $TICKER at $PRICE entry]
- [Technical details verified via RLM]
- [Any caveats or uncertainties]

README.md Template

Create README.md in the assets folder to document the work:

# [Topic] Assets

**Date:** YYYY-MM-DD
**Type:** [Thread/Quick Take/etc]
**Topic:** [Brief description]

---

## Content Files

| File | Description | Status |
|------|-------------|--------|
| YYYY-MM-DD-topic.md | Original draft | ✅ APPROVED |
| YYYY-MM-DD-topic-citrini7.md | Citrini7 rewrite | ✅ APPROVED |

---

## Charts (Original Work - OK to Publish)

| File | Description | Data Source |
|------|-------------|-------------|
| chart1_*.png | [Description] | [Source PDF + page] |
| chart2_*.png | [Description] | [Source PDF + page] |

**Theme:** ocean_depths

---

## Data Verification Log

### [Claim Category 1]
\```
[Claim]: [Value]
- Source: [PDF name, page]
- RLM verified: [grep results or calculation]
\```

### [Claim Category 2]
\```
[Claim]: [Value]
- Source: [PDF name, page]
- Verified: [evidence]
\```

---

## Audit Reports

| File | Purpose |
|------|---------|
| YYYY-MM-DD-topic-audit-manifest.md | Claims to verify |
| YYYY-MM-DD-topic-audit-report.md | Initial audit results |
| YYYY-MM-DD-topic-audit-final.md | Final audit with corrections |

**Audit result:** X/Y claims verified

---

## KSVC Holdings

\```bash
# Verification command
curl -s "https://kicksvc.online/api/[model]" | jq '...'

Result: [Holdings status]
\```

---

## Source Documents

| Source | Path | Used For |
|--------|------|----------|
| [Report name] | /Users/Shared/ksvc/pdfs/YYYYMMDD/file.pdf | [What data] |

---

## Corrections Made

1. [Correction 1]
2. [Correction 2]

---

## Lessons Learned

1. [Lesson 1]
2. [Lesson 2]

Step 6: Chart Decision & Generation

Timing: After draft is complete. The draft crystallizes the thesis - then you see which claims benefit from visualization.

When to Make Charts

Content Type	Chart Likely?	Why
Long Thread	Yes	Multiple data points, trends
Quick Take	Maybe	One key number might not need visual
Breaking News	Rarely	Speed > polish
Victory Lap	Maybe	Entry vs Now comparison

Chart-Tweet Pairing

Principle: Put the most eye-catching visual early (Tweet 1-3) to hook engagement.

Chart Type	Best Tweet Position	Why
Market size / growth bar	Tweet 2 (TLDR)	Pairs with market numbers, shows scale
Component breakdown pie	Tweet 3-4	Pairs with component discussion
Company comparison table	Tweet 5-6	Pairs with company analysis
Timeline / roadmap	Tweet 7-8	Pairs with forward-looking content

Pairing logic:

Match chart to the tweet that contains the same data
Hook tweet (Tweet 1) can go either way:
- Text-only: Clean, curiosity-driven, lets words land first
- With chart: Visual stop, data-forward, shows you have receipts
Visuals work best on data-heavy tweets, not opinion tweets
Final tweet (watchlist/conclusion) usually doesn't need a chart

Example pairing (humanoid robotics thread):

Tweet 1: Hook (optional: market_size_bar.png for visual hook)
Tweet 2: TLDR + market_size_bar.png ← $5.3B→$32.4B numbers
Tweet 3: Component counts (optional: component_pie.png)
Tweet 5: Taiwan names + taiwan_companies_table.png ← KHGEARS/HIWIN/AIRTAC

Decision Process

Review draft - identify "chartable moments"
- Time series data (market growth, price trends)
- Component breakdowns (pie charts)
- Company comparisons (tables)
Check RLM cache - do we have the data?
- Text numbers → bar/line charts
- Tables → comparison tables
- Source charts → reference or recreate

DECLARE SOURCE (MANDATORY) - before any chart generation

"I am charting [METRIC] from [SOURCE] page [X]"
"Source contains these exact values: [list them]"

Generate with chart-factory
```
/chart-factory
```

Chart Generation Workflow

Draft complete → identify chartable claims
                        ↓
              Pull data from RLM cache (NOT from draft text)
                        ↓
              ⚠️ DECLARE SOURCE (state metric + page + exact values)
                        ↓
              Save source image FIRST (before generating)
                        ↓
              Generate with chart-factory (use theme-factory)
                        ↓
              Verify with verification agent
                        ↓
              Save to assets folder

Source Declaration (LEARNED FROM MISTAKE)

⚠️ Why this exists: We once created a "component count" chart but saved a "cost %" source image. The metrics didn't match, making the source invalid for verification.

Before generating ANY chart, you MUST:

Step	Action	Example
1. State	"I am charting [METRIC] from [SOURCE]"	"I am charting hardware cost % from 永豐 p.20"
2. Show	Screenshot the exact source table/chart	Save as `source_hardware_cost_p20.png`
3. Confirm	"Source contains: [exact values]"	"19%, 16%, 13%, 52%"
4. Flag	If transforming data, justify it	"I am NOT transforming - using values as-is"

Red flags - STOP if you notice:

Source shows % but you're charting counts (metric mismatch)
Source has 15 items but chart has 5 (cherry-picking)
Source image doesn't contain your chart's numbers (wrong source)
Company name romanized/guessed from Chinese (fabricated data)
Ticker suffix assumed without checking (TT vs TW)

Company & Ticker Verification

⚠️ LEARNED FROM MISTAKE: We fabricated "Chuing" for 祺驊 (4571). Official name is "KHGEARS".

# Always verify Taiwan company names via TWSE API
curl -s "https://www.twse.com.tw/en/api/codeQuery?query=4571"
# Returns: {"query":"4571","suggestions":["4571\tKHGEARS"]}

Never romanize Chinese names (祺驊 ≠ "Chuing")
Use TW suffix for general audience (TT = Bloomberg only)

Using chart-factory

from chart_factory import create_bar_chart, create_pie_chart, create_table_chart

# Market size bar chart
create_bar_chart(
    data={'2025': 5.3, '2026': 8.3, '2027': 13.0},
    title="Global Humanoid Robot Market",
    theme="ocean_depths",
    annotations={"type": "cagr", "value": "57%"}
)

# Component pie chart
create_pie_chart(
    data={'Reducers': 62, 'Motors': 30, 'Screws': 48},
    title="Component Breakdown",
    theme="ocean_depths",
    explode_largest=True
)

# Company comparison table
create_table_chart(
    columns=['Company', 'Ticker', 'P/E', 'Rating'],
    data=[['Chuing', '4571', '24x', 'BUY'], ...],
    title="Taiwan Supply Chain",
    theme="ocean_depths"
)

Verification (MANDATORY)

After generating, spawn Explore agent with thoroughness: quick for focused verification:

Task(subagent_type="Explore", prompt="""
**THOROUGHNESS: quick**

**CONTEXT ISOLATION: You have NO external conversation history. Work ONLY from this prompt.**

CHART VERIFICATION TASK

Chart: /path/to/chart.png
Type: bar

Source Data (expected):
{"2025": 5.3, "2026": 8.3, "2027": 13.0}

Source Context:
永豐 p.3 - "2025年全球人型機器人規模約53億美元"

Task:
1. Read the chart image
2. Extract numbers from visual
3. Compare to expected data
4. Check for unit consistency (B vs M, % formatting)

Return ONLY JSON:
{
  "verified": true/false,
  "numbers_in_chart": [...],
  "numbers_in_source": [...],
  "discrepancies": [...],
  "notes": "..."
}
""")

Verification checks data → chart integrity. Source accuracy is RLM's responsibility (Step 4a).

Thoroughness = quick: Single-pass verification, focused on specific data points. Fast visual-to-data check.

Save Charts

Save to: draft/YYYY-MM-DD-topic-assets/

Include:

Generated charts (chart1_.png, chart2_.png)
Source images from PDF (for traceability)
generate_charts.py script (reproducibility)

Step 7: Publish to Final Folder

After approval, publish clean version to /Users/Shared/ksvc/threads/.

File Organization Convention

CRITICAL: Flat folder structure, one folder per post.

/Users/Shared/ksvc/threads/
├── 2026-02-03-humanoid-robotics/
│   ├── thread.md                         # Clean content (ready to post)
│   ├── _metadata.md                      # Internal reference (not for posting)
│   ├── chart1_market_size.png
│   ├── chart2_component_breakdown.png
│   └── chart3_taiwan_companies.png
└── 2026-02-05-guc-valuation-debate/
    ├── thread.md                         # Clean content (ready to post)
    ├── _metadata.md                      # Internal reference (not for posting)
    ├── guc-eps-comparison.png
    └── guc-pt-comparison.png

Rules:

✅ Flat structure: YYYY-MM-DD-topic/ at root level (not nested in 2026-02/)
✅ Charts directly in folder (not in charts/ subfolder)
✅ thread.md = clean content only (no metadata header)
✅ _metadata.md = internal reference (sources, audit, not for posting)

thread.md Format

Clean version with just the tweets - no metadata header:

# [Topic Title]

1/ [First tweet]

2/ [Second tweet]
- bullet point
- bullet point

3/ [Third tweet]

...

_metadata.md Format

Internal reference file (prefixed with _ to indicate not for posting):

# Metadata (not for posting)

**Date:** YYYY-MM-DD
**Type:** Long Thread (10 tweets) | Quick Take | etc.
**Status:** READY TO POST

## Sources
- [Source 1 PDF name] ([Date])
- [Source 2 PDF name] ([Date])

## KSVC Holdings Check
- ✅ Held in [Model name] (+X.X% since entry) OR
- ❌ Not held in any of 7 models (checked USA 1-5, TWSE 1-2)
- Integration strategy: [Personal stakes | Industry framing | Victory lap]

## Audit Log
- [Key claim verified via RLM grep]
- [Correction made: old → new]
- [Methodology improvement discovered]

## Charts
- chart1_*.png - [Description] ([Data source])
- chart2_*.png - [Description] ([Data source])

## Notes
- [Special handling notes]
- [Lessons learned]

Example: See /Users/Shared/ksvc/threads/2026-02-05-guc-valuation-debate/_metadata.md

Publish Workflow

# 1. Create publish folder (flat structure)
mkdir -p /Users/Shared/ksvc/threads/YYYY-MM-DD-topic

# 2. Copy clean content as thread.md
cp draft/YYYY-MM-DD-topic-assets/YYYY-MM-DD-topic-citrini7.md \
   /Users/Shared/ksvc/threads/YYYY-MM-DD-topic/thread.md

# 3. Copy charts directly into folder (not subfolder)
cp draft/YYYY-MM-DD-topic-assets/chart*.png \
   /Users/Shared/ksvc/threads/YYYY-MM-DD-topic/

# 4. Create _metadata.md from draft notes
# (Document sources, audit log, holdings, charts)

Result:

/Users/Shared/ksvc/threads/YYYY-MM-DD-topic/
├── thread.md               # Ready to post
├── _metadata.md            # Internal reference
├── chart1_*.png
└── chart2_*.png

When to Publish

Status	Action
Draft approved	Publish to /Users/Shared/ksvc/threads/
Needs revision	Stay in content-pipeline/draft/
Posted to X	Move to /Users/Shared/ksvc/threads/archive/ (optional)

Quality Checklist

Extraction (Step 1a/1b):

⚠️ Checked published threads (/Users/Shared/ksvc/threads/) before topic selection
Topic does NOT duplicate a recently published thread (same source + same angle = reject)
Scanned recent PDF folders (at least 3) with Explore agents
Identified cross-document connections
Deep extracted key reports with RLM
Charts/images extracted and reviewed (use --extract-images)
⚠️ Extraction validation: Every PDF's chars_extracted checked against expected size
⚠️ Read tool fallback used for any PDF with < 1000 chars (or suspiciously low for page count)
Key numbers verified via RLM grep (not just Explore summary)

Cross-Doc Synthesis (Step 1c):

Used rlm-repl-multi to compare across sources (if multiple PDFs)
Asked synthesis questions (consensus, comparison, disagreement)
Documented synthesized insights in cache
Flagged cross-doc claims for audit in Step 4b
Identified unique insights that single-source extraction would miss

Content:

All published numbers have RLM grep confirmation
Technical specifics included (fabs, yields, WPM)
Time frames clear (Q1 2026, 2027e)
Sources cited (multiple reports for cross-doc)
Cross-doc reasoning: claims triangulated across multiple reports
Unique insight that connects dots others miss

KSVC:

equitySeries checked (all 5 US + 2 TWSE models)
filledOrders fallback checked (all 7 models) if equitySeries shows 0%
Entry prices noted for victory lap potential
Integration strategy clear (held vs industry framing)

Voice:

Appropriate type (thread vs quick take vs reaction)
Skepticism included where uncertain
Energy for high-conviction points
Not over-polished

Humanizer (Step 4d):

No AI patterns (em-dashes, "Full stop", etc.)
Has personality/voice
Shows thinking process, not just conclusions

Audit (Step 4a):

All factual claims extracted from draft
Each claim verified via RLM grep against source
Taiwan company names verified via TWSE API
No FAIL status claims remain
UNSOURCED claims either removed, caveated, or sourced
Audit report generated and attached to draft

Charts (Step 6):

Identified chartable claims in draft
Data pulled from RLM cache (not draft text)
⚠️ SOURCE DECLARED before generating (metric + page + exact values)
⚠️ Source image saved FIRST (before chart generation)
⚠️ Source image contains same metric as chart (not transformed)
If data transformed, transformation documented and justified
Used chart-factory with theme-factory theme
Verification agent confirmed data→chart integrity
generate_charts.py script included (reproducibility)

Publish (Step 7):

Draft approved for posting
Created folder in /Users/Shared/ksvc/threads/YYYY-MM-DD-topic/
thread.md contains clean tweets only (no metadata header)
_metadata.md contains sources, audit log, chart descriptions
Charts copied to final folder
Verified all files present before announcing ready to post

PDF Location

Research PDFs: /Users/Shared/ksvc/pdfs/

ls -la /Users/Shared/ksvc/pdfs/ | tail -5

References

references/kirk-voice.md - PRIMARY - Unified voice guide with all content types, formulas, and templates
references/serenity-style.md - Deep dive: data-heavy thread patterns
references/citrini7-style.md - Deep dive: punchy quick take patterns
Full creator studies: ksvc-intern/content-pipeline/creator-studies/

kirk-content-pipeline

Kirk Content Pipeline

Pipeline Steps (MANDATORY)