company-tech-profile

Single-company technology profile and R&D assessment for a defined technology topic. Use when the user asks for a company technology profile, company technical analysis, R&D direction assessment, technical due diligence briefing, or any single-company-by-topic technical evaluation — even if they only mention a company plus a technology area without explicitly asking for a "profile".

Provided by Patsnap Eureka.

Produce an evidence-driven technology profile for a single company in a defined technology topic. The deliverable is a structured report — conclusion first, evidence second — suitable for technical leaders, analysts, and investment screeners.

Use When

• A company's technical strength, R&D direction, or core technology layout in a topic
• Technology due diligence or partnership evaluation for a named company
• Investment screening that needs patent/paper/product evidence for one entity
• A company-only prompt where you can infer a safe technology topic from context

Do Not Use

• Multiple companies need comparison or tiering → route to competitive-landscape
• Pure technology-route comparison with no company anchor → route to tech-route-comparison
• The task is a project/proposal review → route to rd-initiation-review
• Legal patent opinion, FTO, infringement, or litigation analysis
• Generic company overview with no technology depth requirement

Tool Routing And Fallback

This skill works across multiple tool environments. Before retrieval, detect which capabilities are available and select the highest tier that is reachable.

Tier 1 (Recommended): Structured Patent/Paper Retrieval

• Use the host's best structured patent and paper retrieval stack.
• This usually means fielded search with company/entity filtering plus record-level deep fetch.
• Evidence grade: S/A (primary structured evidence)
• Required capabilities: company/entity filtering, topic search, structured patent/paper retrieval, and deep-read of shortlisted records.
• Typical query count per run: 4-8 searches.

Tier 2 (Fallback): Web Research + Companion Lanes

• Use the host's best broad web research tool plus scholarly or finance companion lanes when available.
• Exa, Tavily, Brave, official filings, and domain-specific research databases are good examples, not hard requirements.
• Evidence grade: A/B (web-sourced but structured)
• Coverage loss vs Tier 1: no structured assignee filtering, no IPC/CPC faceting, and weaker paper-graph style coverage

Tier 3 (Fallback): Generic Web Tools

• Use generic web search and page/PDF reading tools available in the host.
• Evidence grade: B/C (unstructured web)
• Coverage loss vs Tier 2: no academic-focused ranking, broader noise

Tier 4 (Minimum): Pure LLM + User Materials

• No external tools required
• Evidence grade: C/U (LLM knowledge + user-provided only)
• Coverage loss vs Tier 3: no current information, no verifiable citations

Routing Rules

• Detect available tools at the start of the workflow.
• Select the highest available tier as the primary retrieval channel.
• When a tier is unavailable, explicitly state the downgrade in the output and in method_decisions.md.
• Do not pretend lower-tier results have the same coverage as higher tiers.
• When using Tier 3 or 4, increase the proportion of explicit gap statements and reduce confidence language in conclusions.

Minimum Working Files

Create or update these files in a writable run folder:

• request.md
• workplan.md
• method_decisions.md
• query_log.csv
• source_index.csv
• claim_ledger.csv
• report.md

Recommended subfolders are described in references/workflow.md.

Default Workflow

Step 0: Freeze Scope

Confirm or infer the following before any retrieval:

• company_name: the primary entity to profile
• topic: the technology area to anchor the analysis
• purpose: what decision this profile supports (orientation / due diligence / partnership / investment screening)
• time_window: default last 3 years unless user specifies otherwise

Scope routing decision tree:

text

User input shape?
├─ Company + technology topic → proceed as single-company profile
├─ Company only, no topic → infer topic from company's main business domain;
│  confirm with user if ambiguous
├─ Multiple companies mentioned → redirect to competitive-landscape
├─ Technology topic only, no company → redirect to tech-route-comparison
└─ "Compare company A with B" → redirect to competitive-landscape

If the user only names a company without a topic, do not stall — infer the most likely technology topic from the company's known domain and state the inference explicitly.

Step 1: Anchor The Entity

1. Normalize the company name: resolve parent vs subsidiary, Chinese vs English aliases, brand name vs legal entity.
2. Run a confirmation search to verify the company has presence in the topic.
   • Tier 1: structured entity-filtered patent/paper search
   • Tier 2+: targeted web or scholarly search across patent indices, official domains, and research pages
3. If the company has no patent presence in the topic, note this as a finding and shift emphasis to papers and public signals.
4. Search for the company's official site to capture basic profile, product lines, and recent announcements.

Step 2: Patent Portfolio Analysis

1. Run 2-3 patent searches varying keywords to cover the topic's sub-routes.
   • Tier 1: structured patent search with entity filters and keyword variations
   • Tier 2: targeted web research across patent indices, official domains, and technical publications
   • Tier 3/4: generic web or user-provided materials
2. Analyze from search results: filing volume trend (by year), IPC distribution, key technology clusters, filing jurisdictions.
3. Decide whether direct deep-reading is stable enough, or whether representative sampling is required.

Sampling decision tree (Tier 1 with structured retrieval):

text

Patent evidence shape?
├─ Scope is small and route concentration is obvious
│  └─ Skip sampler; deep-read the strongest 3-5 patents directly
├─ Scope is medium/large and route coverage matters
│  └─ Run the optional representative patent sampling script
├─ Scope is large and highly heterogeneous across time/jurisdiction/CPC
│  └─ Run the optional sampling script; treat result as default evidence pack
└─ Task is drifting toward legal completeness
   └─ Do not use sampler; redirect to a legal/FTO-oriented workflow

Default thresholds:

• <= 12 patents after filtering: usually skip sampler
• 13-80 patents: sampler recommended if route interpretation is part of the goal
• > 80 patents: sampler should be the default

4. Deep-read top 3-5 representative patents to extract: core technical approaches, key claims, innovation points.
5. Identify the company's primary technology routes and any emerging pivots.

Step 3: Paper And R&D Signals

1. Run 1-2 paper searches with company name + topic.
   • Tier 1: structured paper retrieval
   • Tier 2: scholarly companion search or targeted web research
   • Tier 3/4: generic web or user-provided materials
2. Analyze: publication trend, collaborating institutions, research focus areas.
3. Cross-reference with patent routes — consistency strengthens confidence; divergence is a signal worth noting.

Step 4: Product And Public Signals

1. Search for: product launches, benchmark results, partnerships, funding rounds, key hires, conference presentations.
2. Read 2-3 high-signal pages for deeper extraction.
3. Corroborate patent/paper findings with product reality — a strong patent portfolio without shipped products is a different signal than one with them.

Step 5: Synthesis

Adjust analysis emphasis by industry type:

text

Industry type?
├─ Semiconductor / Hardware → emphasize: patent clusters by process/architecture,
│  fab vs fabless signals, IPC H01L/H04 distribution
├─ Biopharma / Medtech → emphasize: pipeline stage, clinical trial signals,
│  target coverage, IPC A61K/C07/A61B distribution
├─ AI / Software → emphasize: papers and benchmarks, open-source presence,
│  model performance, API/platform adoption signals
├─ New Energy / Materials → emphasize: material composition patents, capacity
│  signals, cost trajectory, IPC H01M/C01/H02 distribution
└─ General Manufacturing → balanced patent + product + standard participation

Synthesize all evidence into the output skeleton. Separate every claim into:

• Verified fact: directly supported by patent, paper, or official filing
• Evidence-backed inference: reasonable conclusion from multiple signals
• Open gap: insufficient evidence, state what is missing and suggest next steps

Output Skeleton

1. Lead With Conclusion

1-2 sentences: the company's overall technical position and trajectory in the topic.

2. Technology Layout Highlights

2-4 bullet points: core technology routes, key patents or papers, differentiation from the broader field. Use a compact markdown table when comparing 3+ dimensions.

3. Trend Assessment

1 paragraph: R&D direction evolution over the time window — acceleration, deceleration, pivots, emerging focus areas. Cite specific year-over-year changes when data supports it.

4. Risks And Recommendations

1 set of bullet points:

• Key risks (technology gaps, dependency, competitive pressure)
• Opportunities (white spaces, collaboration potential, emerging routes)
• Recommended next actions for the user's decision context

Every claim must cite its source type and identifier, e.g., [Patent: CN1234567B], [Paper: DOI or title], [Web: source name].

Completion Gates

All must pass before delivering the final answer:

• Company entity confirmed and aliases resolved (not confused with subsidiary or homonym)
• At least 1 patent search and 1 paper search executed with results analyzed
• If patent scope is broad, representative sampling was executed or explicitly skipped with reason
• Core technology routes identified with patent or paper evidence
• Every major claim has a traceable source citation
• Verified fact / evidence-backed inference / open gap clearly separated
• Output follows the 4-part skeleton (conclusion → layout → trend → risks)
• No drift into generic company biography without technical substance
• Tool tier and coverage limitations explicitly stated in the output
• If evidence is insufficient for a confident profile, explicitly stated with follow-up suggestions
• Company marketing claims treated as signals, not verified facts, unless corroborated

Guardrails

• Do not equate patent volume with technical strength without considering route quality, claim scope, and recency.
• Do not treat company marketing claims, recruiting signals, or press releases as verified facts until corroborated by stronger sources.
• Do not mix parent company and subsidiary data without an explicit scope note.
• This is a single-company skill — do not produce competitive comparison matrices. If the user needs multi-company comparison, redirect to competitive-landscape.
• If the company has minimal patent/paper presence in the topic, say so directly rather than padding the profile with tangential evidence.
• Do not use representative sampling for legal-opinion tasks or when exhaustive recall is required.
• Do not present the sampled patent set as "all relevant patents"; state clearly that it is a structured representative subset.
• Do not let the sampled set silently override full-scope facts such as total patent count, jurisdiction count, or exact filing volume.
• Do not force a strong conclusion when the evidence base is thin — prefer explicit uncertainty over false confidence.
• Keep charts or visual aids to 1-2 maximum; retain only those that directly support the analytical conclusion.
• For non-evidence dimensions (TCO, regulation, teardown), write "evidence pending" rather than fabricating a judgment.

Optional Host-Specific Automation

Some hosts may provide extra helpers for representative sampling, quantitative aggregation, or patent-card normalization. These helpers are optional and are not part of the core open-source contract.

If you use host-specific automation:

• record the helper and its role in method_decisions.md
• keep the final evidence compatible with query_log.csv, source_index.csv, and claim_ledger.csv
• do not make the report depend on a helper that other hosts cannot reproduce

Load These Files Only As Needed

• references/method-benchmark.md
• references/domain-playbooks.md
• references/workflow.md
• references/source-routing.md
• references/deliverables.md
• references/evidence-schema.md
• references/quality-gates.md
• templates/request-template.md
• templates/workplan-template.md
• templates/report-outline.md