ScoutResearch Intelligence
Read the company before you write a word to it.
Scout takes one domain and returns a dossier: what the company is, what changed in the last two quarters, how it buys, what it runs, who decides, and where the evidence says you fit. Every claim carries a numbered source and the moment that source was retrieved — and the last section is a list of the things it could not establish.
- Pipeline
- Stage 08 — after qualification
- Input
- One domain
- Output
- 7 sections, every claim sourced
Subject
Halverson Freight Systems
Identity and shape
verifiedA regional freight brokerage running four depots, with 210 employees1 on its last confirmation statement. Revenue is not filed at this size and no credible public estimate exists, so none is given.
It sells to shippers directly rather than through a marketplace, which is visible in the quote request flow2 on its own site: a named account manager, not an instant rate.
- Entity
- HF-2291
- Incorporated
- 2009
- Depots
- 4
- Employees
- 210 (filed)
What changed in the last two quarters
verifiedFour revenue roles opened inside eleven days: three SDR positions and a Head of Revenue Operations, the last of which is new to the company3. Two depot leases were registered in the last twelve months.
The pricing page has been rewritten twice4 this quarter, moving from per-load rates toward contracted volume. Companies rewrite pricing when they are changing who they sell to.
- Window
- Last 180 days
- Material changes
- 6
How they buy
inferredA public tender notice names a Q4 renewal5 for the transport management contract. That sets the shape of the cycle. It does not set the date of the decision, and nothing published narrows it further.
No procurement function appears on any public listing, so the committee is reconstructed from role titles alone: COO, the new revenue operations lead, and finance sign-off above a threshold that is not published. Treat this paragraph as a reconstruction, not a record.
What they run
verifiedNo CRM is detectable on the public estate. Contact forms post to a shared mailbox, the analytics tag was replaced 34 days ago, and there is no marketing automation of any kind on the site.
The absence is the finding. A revenue team being built against a renewal, with no system of record6, is a specific situation with a short shelf life.
- CRM
- none detected
- Marketing automation
- none detected
- Analytics
- replaced 34d ago
- Form target
- shared mailbox
Who decides
verifiedTwo directors are listed on the register and one of them signs the filings. The COO has been in post four years and is named on the tender notice as the contact for the category.
The Head of Revenue Operations started nine days ago with no public history in this category. A function being defined rather than inherited behaves differently in a first conversation, so it is worth knowing which one you are walking into.
- Directors
- 2 (1 signatory)
- Named on tender
- COO
- RevOps tenure
- 9 days
Where you fit
inferredThe evidenced gap is between a renewal decision landing in Q4 and a revenue function that currently has no system of record to make it from. That is a claim about their situation, drawn from sections 02 and 04.
It is not a claim that you solve it. Scout does not model your product, and a research tool that tells you that you are a perfect fit for everyone it reads is telling you nothing.
What we could not establish
unresolvedRevenue. Not filed at this size; no credible public estimate. Reported as unknown rather than modelled.
The incumbent vendor. The tender notice names the category, not the supplier.
Budget owner. No public evidence of who signs above the approval threshold, or what the threshold is.
Decision date. Q4 is a renewal window, not a decision date. Anyone who gives you a date here has invented it.
Illustrative record. Halverson Freight Systems is fictional, as is every company named on this site — Arvyn has no customers to name. The dossier structure, the citation discipline and the seventh section are exactly as shipped.
Position
Research is the expensive step, so it goes last.
Most teams research first and qualify by feel afterwards, which is how a week disappears into a beautifully briefed account that was never going to buy this year.
Nexus does the cheap work first: thousands of companies assessed on public, dated signals until twenty are left. Only then does Scout open one, and by that point the account has already survived an argument about whether it deserved the attention.
That ordering is why the dossier can afford to be deep. It is also why Scout will not run on an unqualified domain from inside the product — the API will do it on request, and it will tell you it is expensive.
Where it goes next
The dossier is not an export. It sits on the account record beside the score ledger that shortlisted it, and Sonar reads the same record to build the rehearsal persona — no copy-paste, no second source of truth.
- 01Understand the businessYour site, your deck, your own wordsNexus
- 02Define the ICPSegments, sizes, regions, buying committeeNexus
- 03Build the candidate universeThousands of companies, none contacted yetNexus
- 04QualifyShould we contact them — not can weNexus
- 05Score the buying windowObservable evidence only, each one citedNexus
- 06PrioritiseThe shortlist, capped on purposeNexus
- 07Find the decision-makerDiscovered, then verified before useNexus
- 08Research the accountThe cited dossierScout
- 09Write and sendPer-account brief, sending windows, repliesNexus
- 10Take the conversationRehearsal first, live coaching secondSonar
- 11Learn from the outcomeEvery reply re-weights the scoringNexus
Research the accountfour separated passes
Retrieve, extract, verify, then assemble.
Most research tools collapse these four into one prompt and hand you a fluent paragraph with no way to tell which sentence came from a filing and which came from the model's imagination.
RETR08
Retrieval
Scout reads the company’s own estate, the archived versions of it, the public registers it appears in, the tender and filing portals that carry dates, its job listings, and trade coverage. Pages that need JavaScript are rendered rather than guessed at from raw HTML, because half of what matters on a modern site is not in the source.
Everything is fetched fresh, stamped with the time it was fetched, and diffed against the archived copy so that a rewrite shows up as a rewrite rather than as the current state of the world.
- Respects
- robots directives, rate limits, and the login wall
- Never
- Purchased exports of gated data, or anything requiring credentials
- Stamps
- Retrieval time, per document, stored with the claim
- The company's own estaterendered
Positioning, pricing, stack, form targets
- Archived copiesdiffed
What changed, and when it changed
- Company registerpublic filing
Entity, directors, filed headcount, charges
- Tender and contract portalspublic notice
Renewal windows, categories, named contacts
- Job boards and careers pagefetched
Roles opened, roles refilled, function shape
- Press and trade coveragefetched
Public commitments with dates attached
EXTR08
Extraction: claims, not summaries
The unit of extraction is a claim, and a claim is only ever created attached to the document span it came from. Nothing is summarised into a paragraph and then cited afterwards, because a citation applied to a paragraph after the fact tells you where the writer was reading, not where the fact came from.
This is the difference you can feel when you interrogate a dossier. You can point at any specific sentence in Scout’s output and ask which document produced it, and there is exactly one answer, with a timestamp. A claim that lost its source during assembly is dropped rather than softened into a hedge.
- Unit
- One claim, one span, one source, one retrieval time
- Ordering
- Extraction happens before any prose is written, never after
- Drop rule
- A claim that cannot name its span does not reach the dossier
VERI08
Verification, and what happens when sources disagree
Material claims need two independent sources. Where only one exists, the claim is marked inferred rather than deleted — the reader is better placed than the engine to decide how much weight a single-sourced fact deserves.
Where sources conflict, filed records outrank profile pages, which outrank mentions. The losing figure stays visible in the record, because knowing that two published numbers disagreed is itself worth carrying into a first conversation.
Assembly rules
- Two sourcesRULE-01
A material claim needs two independent sources, or it is marked inferred
StrongEngine rulealways on - ConflictRULE-02
When sources disagree, the filed record wins and the disagreement is shown
StrongEngine rulealways on - FreshnessRULE-03
Every claim carries the moment it was retrieved, not the moment it was written
ModerateEngine ruleevery run - AbsenceRULE-04
Absence is recorded as absence, never as a low value
StrongEngine rulealways on
LEDG08
The source ledger
Every superscript in the dossier resolves here: the document, how it was obtained, and the exact minute it was read. The ledger travels with the dossier into the PDF export and into the API record, so a claim never arrives anywhere without the means to check it.
Sources that go dark between runs keep their timestamp and are flagged on the next pass. A dossier is a photograph of a company at a moment, and it should be obvious which moment.
- [1]Company register — confirmation statement, employee countregister.example.gov/companypublic register2026-08-15 04:12 UTC
- [2]Own site — quote request flow and account-manager routinghalverson-freight.example/quoterendered2026-08-15 04:14 UTC
- [3]Careers page, diffed against the archived copy from 2026-05-02halverson-freight.example/careersfetched + archive diff2026-08-15 04:15 UTC
- [4]Pricing page, two revisions this quarterweb.archive.example/halverson-fre…archive2026-08-15 04:16 UTC
- [5]Tender portal — notice HF-40912, transport management renewaltenders.example.gov/noticepublic register2026-08-15 04:18 UTC
- [6]Stack detection across the public estate — headers, tags, form targetshalverson-freight.example/contactrendered2026-08-15 04:19 UTC
Retrieval times are stored with the claim. A source that goes dark keeps its timestamp and is flagged on the next run.
The seventh section is the one that earns the other six.
Research tools are rewarded for looking complete. Ask one for revenue and it will produce a number; ask it for the decision date and it will produce a quarter. Neither was published anywhere. Both will be repeated back to a buyer who knows the real figure, and the conversation is over.
Scout is built the other way round. Unknown is a valid output and it is rendered as prominently as anything the engine did establish, which is the only way a reader can tell the difference between what was found and what was filled in.
Revenue
Not filed at this size. No credible public estimate exists, so none is given.
Incumbent vendor
The tender notice names the category, not the supplier.
Budget owner
No public evidence of who signs above the approval threshold.
Decision date
Q4 is a renewal window. Anyone who gives you a date has invented it.
Hand-off to stage 10
The dossier is not something you skim in the car park.
A brief you read once and half-remember is worth very little at minute eleven of a call. So the dossier is also the material Sonar builds the rehearsal buyer from, which means the objections you practise against are the ones this company actually has grounds to raise.
Halverson Freight Systems
- Q4 renewal named in a public tender notice
- Head of Revenue Operations, nine days in post
- No CRM detectable anywhere on the public estate
- Four things the research could not establish
7 sections · 23 sources · retrieved 2026-08-15 04:21 UTC
Priya Nandakumar, Head of Revenue Operations
- Defends the incumbent, because switching lands on her
- Knows the renewal date and will not confirm it to you
- Will not accept a claim about her stack without evidence
- Refuses to answer the four questions the dossier flagged
Built from DOS-4471. No re-entry, no copy-paste
The persona inherits the gaps as well as the findings. Because the dossier recorded that the budget owner was never established, the rehearsal buyer deflects that question — and you find out in a practice room rather than on the call that you had no plan for it.
Specification
What goes in, what comes out.
Including the parts that are inconvenient to publish.
On cost
A dossier is the most expensive operation in the system, which is the honest reason it runs after qualification rather than before it. The ordering is a budget decision as much as a quality one, and pretending otherwise would be the first unsourced claim on this page.
Input and output
- input
- One domain. Optionally a name and a country when the domain is ambiguous.
- output
- Seven sections, a source ledger, and a machine-readable record over the API.
- citation unit
- The claim. Not the section, not the document — every claim resolves to one source.
- gap reporting
- Section 07 always renders, even when it is empty. An empty gap list is itself a finding.
Retrieval
- surfaces
- The company's own estate, public registers, tender and filing portals, job boards, archives.
- rendering
- JavaScript-rendered pages are rendered, not guessed at from raw HTML.
- change detection
- Pages are diffed against archived copies, so a rewrite is visible as a rewrite.
- respect
- robots directives and rate limits are honoured. Nothing behind a login is touched.
Assembly
- verification
- Two independent sources for a material claim, or the claim is marked inferred.
- conflicts
- Filed records outrank profile pages outrank mentions. The conflict stays in the record.
- recency
- Retrieval time is stored per claim and shown on every source line.
- refresh
- Re-run on demand; a stored dossier shows its own age rather than looking current.
Delivery
- in product
- On the account record in the console, beside the score ledger that shortlisted it.
- export
- PDF and JSON. The PDF keeps the superscripts and the source ledger.
- hand-off
- Sonar reads the same record to build the rehearsal persona. No re-entry, no copy-paste.
Limits
What Scout refuses to do.
Research tools are judged on how complete they look, which is precisely the incentive that makes them unreliable. These refusals cost Scout the appearance of completeness on purpose.
It will not quote a source it could not retrieve
If a page did not load, the claim does not appear. There is no cached summary standing in for a document nobody could open, because that is how a research tool starts inventing quietly.
It will not touch anything behind a login
No scraped private profiles, no purchased exports of gated data, no credential sharing. If a claim cannot be reached by anyone with a browser, it is not in the dossier.
It will not profile a person beyond their public professional record
Role, tenure, and what they have said publicly about their work. Not their personal accounts, not their location, not their family, not anything a stranger has no business knowing before a first call.
It will not estimate a number that nobody published
Unfiled revenue is reported as unknown. A modelled figure presented next to filed figures is indistinguishable from a fact on the page, and that is exactly the confusion that gets repeated back to a buyer.
It will not resolve an ambiguous company for you
Two entities with the same trading name produce a question, not a merged dossier. Merging them silently is worse than returning nothing.
It will not tell you that you are a good fit
Section 06 states the gap the evidence supports. Whether your product closes that gap is your judgement, and a research tool that flatters every account is not researching.
It will not present a stored dossier as current
Every dossier displays its own retrieval time, and a stale one says so before you read it. Research ages faster than anyone expects it to.
It will not run before qualification
Deep research on an unqualified account is the most expensive way to be wrong. Scout opens accounts that have already argued their way onto the shortlist.
Either side of the dossier
Scout is the middle step. One product decides the company is worth reading; the other turns what was read into a conversation you have already had once.
Decides which twenty companies are worth this much research, and shows the ledger and the argument that put each one on the list.
Read the Nexus page
Builds the rehearsal buyer from this dossier — findings and gaps alike — then sits on the real call and flags what you have not asked.
Read the Sonar page
Walk in already knowing, and knowing what you don't.
Fifteen days. Connect a domain, let the shortlist build, and open the first dossier the system produces. Read section 07 first.