ms-brief

ms-brief — Mercury Strategy briefing stage

Collection stage 1 of 3. Produces a structured evidence manifest covering company identity, benchmark data, situational awareness, and a high-level website inventory. No analysis. No findings.

---

Before you begin

Read this file in full before starting. Then:

1. Use the company name provided in the slash command argument. Look up the primary domain yourself via web search — do not ask the consultant to confirm it unless ambiguous (e.g. multiple companies with the same name, or a recent rebrand/merger).
2. Assume full pipeline scope unless the consultant says otherwise. Do not ask for scope confirmation — just begin collection immediately.
3. Do not open an input dialog or ask a question before starting work. The consultant has already told you what to do by running the command.

---

Step 1 — Company identity

Establish and record the following. All fields are required before proceeding.

Field	Source
Legal name	Company website, regulatory filings
Trading name (if different)	Website, news coverage
Stock ticker and exchange	IR landing page, Bloomberg, LSE/NYSE
Listing status	Listed / private / recently listed / delisted
Sector and sub-sector	Company description, SIC code
Headquarters	About or contact page — verify against a second source (e.g. Companies House, Bloomberg, annual report) before recording. HQ is the most trust-sensitive field; an error here undermines the entire brief.
Primary domain	Confirmed live URL
Subdomains (if any)	Note any careers / IR / sustainability subdomains
Geographic scope	Global / regional / single-market
Revenue scale (approximate)	Annual report, press coverage — record if available, note if not

Record in the evidence manifest as company_identity.

---

Step 2 — Benchmark check

Path A — Company is in the BigQuery benchmark dataset

Query the Connect.IQ benchmark dataset for this company. The dataset covers 747 companies including the FTSE 100, FTSE 250, S&P 500, and Euro STOXX 50.

Run three queries:

Query 1 — Company scores

SELECT company, overall, company_narrative, content_mix, channel_mix,
       optimization, reach, about_us, ir, media, csr, careers,
       reputational_resilience, index_name, dataset_year
FROM sector_intelligence.iq_benchmarks
WHERE LOWER(company) LIKE LOWER('%{company}%')
LIMIT 5

Query 2 — Index statistics

SELECT
  AVG(overall) AS mean_score,
  APPROX_QUANTILES(overall, 4)[OFFSET(2)] AS median_score,
  APPROX_QUANTILES(overall, 4)[OFFSET(3)] AS p75_score,
  COUNT(*) AS company_count
FROM sector_intelligence.iq_benchmarks
WHERE index_name = '{index_from_query_1}'

Query 3 — Rank within index

SELECT company, overall,
  RANK() OVER (ORDER BY overall DESC) AS rank,
  COUNT(*) OVER () AS total
FROM sector_intelligence.iq_benchmarks
WHERE index_name = '{index_from_query_1}'
ORDER BY overall DESC

Record in the evidence manifest as benchmark:

{
  "overall": 0,
  "index_name": "",
  "sector_median": 0,
  "sector_p75": 0,
  "rank": 0,
  "total_in_index": 0,
  "dataset_year": ""
}

Path B — Company is not in the dataset

Record in the evidence manifest:

{
  "benchmark": {
    "status": "not_in_dataset",
    "note": "Company not found in Connect.IQ benchmark dataset. No comparative scoring available."
  }
}

Do not attempt to estimate a score. Do not search for alternative benchmark sources at this stage.

Fallback context: When a company is not in the dataset, query general index statistics (median, P25, P75) for the company's listing index (e.g. FTSE 100, FTSE 250) and record them in benchmark.index_context. This gives the consultant sector framing even without a company-specific score.

---

Step 3 — Situational awareness

Gather recent intelligence on the company. Scope: last six months only. Do not surface older material unless it is a structural fact (founding date, listing history).

3a — Web search

Run web searches covering:

• Recent corporate news and announcements
• Strategy updates or leadership changes
• Financial results or trading statements
• ESG commitments or sustainability announcements
• Brand or positioning changes
• Regulatory matters (if public)

Do not search for: share price, analyst ratings, or market cap data. These are not used in the findings stage.

3b — Material events check

Apply references/MATERIAL_EVENTS_CHECKLIST.md to the search results. For each event type in the checklist, record:

• Whether the event occurred in the last 12 months
• Date and source URL
• Relevance to the digital estate (e.g. a new CEO often signals an upcoming site refresh; an M&A event may mean IR content is in transition)

Record in the evidence manifest as situational_awareness:

{
  "search_date": "",
  "material_events": [],
  "recent_news_summary": "",
  "sources": []
}

3c — Direct fetch of company news section

Fetch the company's news or newsroom section directly — do not rely on search results alone for recent news. Use firecrawl_scrape on the newsroom landing page and note:

• Date of most recent item
• Nature of recent announcements (results, strategy, people, product)
• Publishing cadence (estimated from visible archive)

---

Step 4 — Site structure discovery

Establish the primary domain's structure using firecrawl_map. Check references/CRAWL_CONFIG.md for any domain-specific configuration before running the map.

Default call:

firecrawl_map(url: "{primary_domain}")

If a CRAWL_CONFIG entry exists for this domain, use the specified parameters instead.

From the map output:

• Count total URLs returned
• Identify top-level sections (from first path segment after locale stripping — see references/CLASSIFICATION_RULES.md Priority 2)
• Note subdomains present
• Check robots.txt at {domain}/robots.txt and record any significant Disallow rules

If firecrawl_map returns fewer than 10 URLs, the site may be blocking the crawler. Record firecrawl_map: blocked and fall back to sitemap.xml.

Sitemap fallback: Fetch {domain}/sitemap.xml. If it returns a sitemap index, fetch each child sitemap. Extract <loc> URLs and count them.

Record in the evidence manifest as site_discovery:

{
  "firecrawl_map_status": "complete | partial | blocked",
  "sitemap_status": "present | present_incomplete | not_found",
  "total_urls_discovered": 0,
  "top_level_sections": [],
  "subdomains_found": [],
  "robots_disallow": []
}

---

Step 5 — Homepage and key section quick audit

Scrape the homepage and four to six key section landing pages. Priority order:

1. Homepage
2. IR or investor relations landing
3. Strategy or about/at-a-glance
4. Sustainability or ESG landing
5. Careers landing
6. Newsroom or media centre

For each page, use firecrawl_scrape with onlyMainContent: true (or CRAWL_CONFIG override if applicable).

Apply URL classification from references/CLASSIFICATION_RULES.md to assign each page a section_key and playbook_page_type.

Apply presence quality classification from references/CLASSIFICATION_RULES.md (Presence quality classification section) to each scraped page:

Quality	Criteria
present	400+ words, structured headings, content addresses the expected concept
present_thin	Page exists but fewer than 200 words or generic boilerplate
present_stale	Content not updated in 18+ months
present_documents_only	Page consists only of PDF download links
present_external	Content served via external platform
present_generic	Page exists but not configured for expected audience or purpose

Record in the evidence manifest as section_inventory — one entry per page:

{
  "url": "",
  "section_key": "",
  "playbook_page_type": "",
  "classification_rule": "priority_1_exact | priority_2_segment | priority_3_deep | ambiguous",
  "classification_confidence": "high | medium",
  "presence_quality": "",
  "word_count": 0,
  "page_title": "",
  "content_summary": ""
}

---

Step 6 — Document inventory

6a — Shallow probe

Probe for downloadable documents linked from the pages crawled in Step 5. Look for PDF, PPTX, XLSX links in the scraped content. Do not fetch document contents at this stage — record URLs and apparent document types only.

Classify by document type using references/CLASSIFICATION_RULES.md Document sub-classification table:

• Annual report
• Results presentation
• Sustainability report
• ESG data pack
• Investor presentation / CMD deck
• Governance report
• Modern slavery statement
• Gender pay report

Priority document types (most valuable for the findings stage):

1. Capital markets day / investor day deck
2. Annual report
3. Results presentation (most recent)
4. ESG / sustainability report

6b — Consultant approval gate

Present the document list to the consultant before extracting any documents. Show:

• Document type
• URL
• Approximate file size (if visible from link context)
• Priority (High / Medium / Low)

Wait for confirmation before proceeding. The consultant may choose to:

• Extract all documents
• Extract priority documents only
• Skip document extraction entirely

Do not extract document contents without explicit confirmation.

6c — Document extraction (if approved)

For approved documents, use firecrawl_scrape on the document URL. Record:

• Key claims or strategy statements found
• Financial data points cited
• ESG targets or commitments stated
• Any content that materially affects the situational awareness picture

---

Step 7 — Peer research

Read references/PEER_RESEARCH_GUIDE.md before beginning this step.

7a — Peer selection

Propose 4–5 sector peers using the three primary filters from PEER_RESEARCH_GUIDE.md:

1. Sector alignment — same GICS sector or sub-industry, or sufficiently close that the same investor and stakeholder audiences would evaluate both
2. Index membership — same index or one tier adjacent; default to the client's own index
3. Comparable market capitalisation — broadly comparable in size (3–5x range acceptable); do not compare a £500m company against a £50bn company

After applying the primary filters, use the secondary considerations to choose between eligible candidates (direct competitive overlap, acknowledged benchmarks, digital maturity contrast, avoid related parties).

Present the proposed peer set to the consultant before running any research. The peer table must be the last thing you output before stopping — do not append menus, follow-up options, or other text after the table. This ensures the table is visible on screen when the consultant reads it.

Proposed peer set for [Client]:

| # | Company | Sector | Index | Market cap | Rationale |
|---|---------|--------|-------|------------|-----------|
| 1 | [Name]  | [sector] | [index] | [cap] | [1 sentence] |
| 2 | [Name]  | [sector] | [index] | [cap] | [1 sentence] |
...

Reply "go" to confirm, or suggest replacements.

Wait for confirmation. Do not begin peer research until the peer set is locked. Do not show a choice menu or any other UI element alongside the peer table — the table itself is the prompt.

---

7b — Feature matrix pass

For each confirmed peer, run the research scope defined in PEER_RESEARCH_GUIDE.md:

Per peer (credit budget: ~7 credits)

1. Run firecrawl_map on the root domain. Record total URL count and subdomains. Classify sections present using references/CLASSIFICATION_RULES.md.

2. Scrape 6–10 targeted pages: homepage, IR landing, sustainability landing, careers landing, newsroom, and any sections relevant to the specific archetypes being investigated (check which archetypes are most likely given the company profile from Step 1).

3. Run the abbreviated document check: annual report, sustainability report, results presentation, TCFD report, and any documents directly relevant to the engagement focus. Do not run the full DOCUMENT_CHECKLIST at this stage — that is the client's domain (ms-crawl).

4. Note any features or capabilities that are notably stronger or weaker than the client.

For each peer, complete the feature matrix using the standard feature set from PEER_RESEARCH_GUIDE.md (F01–F39). Record each feature as present, present_thin, present_external, absent, or not_assessed.

Feature categories:

• F01–F10 — IR and investor content
• F11–F18 — Sustainability and ESG
• F19–F25 — Careers and employer brand
• F26–F33 — Digital and content capability
• F34–F39 — Navigation and audience experience

If the total peer research budget would exceed 35 credits, reduce pages scraped per peer rather than reducing peer count. Minimum viable pass per peer: homepage + IR landing + sustainability landing + careers landing (3 targeted pages, ~4 credits).

If a peer site blocks the crawl, record all feature matrix entries as not_assessed and note the limitation.

---

7c — Connect.IQ integration

Where Connect.IQ benchmark data is available for peers, pull it to supplement the feature matrix:

SELECT company, overall, ir, csr, careers, media, about_us
FROM sector_intelligence.iq_benchmarks
WHERE LOWER(company) IN ([peer names lower-cased])
AND dataset_year = 2024

If Connect.IQ data is not available for a peer (company not in the benchmark universe), note this and rely solely on the feature matrix. Do not estimate IQ scores for peers.

---

7d — Compile peer_context block

Record peer research findings in the evidence manifest as peer_context:

{
  "peer_context": {
    "confirmed_peers": [
      {
        "name": "",
        "domain": "",
        "index": "",
        "sector": "",
        "market_cap_approx": "",
        "pages_discovered": 0,
        "subdomains": [],
        "iq_scores": {},
        "notable_features": [],
        "notable_absences": [],
        "research_limitations": []
      }
    ],
    "feature_matrix": {
      "client": "",
      "peers": [],
      "research_date": "",
      "features": [
        {
          "feature_id": "F01",
          "category": "",
          "feature": "",
          "client": "present | present_thin | present_external | absent | not_assessed",
          "peers": {},
          "notes": ""
        }
      ]
    },
    "peer_patterns": [
      {
        "pattern": "",
        "client_position": "aligned | ahead | behind | not_assessed"
      }
    ]
  }
}

peer_patterns: after completing the feature matrix, identify patterns across the peer set — features present across most or all peers, features the client has that peers lack, and gaps shared across the entire peer set. These patterns feed directly into benchmark framing in ms-findings.

---

Step 8 — Save the evidence manifest

Save the complete evidence manifest as {company}-ms-brief-evidence.json.

Manifest structure:

{
  "stage": "ms-brief",
  "company": "",
  "domain": "",
  "collected_at": "",
  "company_identity": {},
  "benchmark": {},
  "situational_awareness": {},
  "site_discovery": {},
  "section_inventory": [],
  "document_inventory": [],
  "document_extraction": [],
  "peer_context": {},
  "evidence_gaps": [],
  "consultant_notes": []
}

evidence_gaps: record any sections that could not be accessed, pages that returned errors, or data points that could not be confirmed. These are inputs to the ms-findings limitations section.

consultant_notes: record any observations that require consultant judgement — ambiguous company structures, recently changed domains, apparent site migrations in progress, etc.

---

Persist to BigQuery (optional)

After saving the evidence manifest, store the brief results in BigQuery if the bigquery connector is available (mcp__bigquery__run_query in Cowork). This is a best-effort step — if the tool is unavailable, skip silently.

SELECT GENERATE_UUID() AS analysis_id

Then insert:

INSERT INTO sector_intelligence.ms_analyses
  (analysis_id, company, domain, generated_at, analysis_type,
   executive_summary, coverage_confidence, iq_score, index_name,
   sector, listing_status, pages_loaded, sections_assessed,
   evidence_gaps, limitations, loaded_at)
VALUES (
  '{analysis_id}',
  '{company}',
  '{domain}',
  CURRENT_TIMESTAMP(),
  'ms_brief',
  NULL,
  NULL,
  {benchmark.iq_score or NULL},
  '{benchmark.index_name or NULL}',
  '{sector or NULL}',
  '{listing_status}',
  0,
  ['{top_level_section_1}', '{top_level_section_2}', ...],
  ['{evidence_gap_1}', ...],
  [],
  CURRENT_TIMESTAMP()
)

The analysis_type is 'ms_brief'. Fields like executive_summary and coverage_confidence are NULL — they are populated when ms-findings runs and updates this row. The artefact_json column can optionally hold the full manifest JSON.

Do not block: If the INSERT fails, proceed to Stage completion.

---

Stage completion

After saving the manifest, show a clean summary to the consultant:

Show:

• Company confirmed (name, domain, listing status)
• Benchmark position (or note if not in dataset)
• Top-level sections identified (count and names)
• Material events flagged (if any)
• Documents found (count by type) — and whether extraction is pending, approved, or skipped
• Peer set confirmed and researched (peer names, key feature matrix patterns)
• Evidence gaps (if any)

Do not show: raw JSON, criterion observations, findings, or recommendations.

Offer:

• Continue to /ms-crawl (recommended next step)
• Pause here (manifest is saved — crawl can run later)

If document extraction is still pending consultant approval, surface the document list now and wait for instruction before offering to continue.

---

What this stage does not do

• Produce findings, evaluations, or recommendations
• Assess the quality of individual pages beyond presence quality classification
• Run the full four-source crawl on the client site (that is ms-crawl's responsibility)
• Run the full DOCUMENT_CHECKLIST against the client site (that is ms-crawl's responsibility)
• Deep-crawl peer sites — peer research is a feature matrix pass only (~7 credits per peer)
• Apply archetype criteria
• Assess audience journeys