A forensic memo, not an audit checklist.

What is a deep diagnostic in Nerxus One?

A deep diagnostic is the single-business workspace where Radar runs detector families, writes evidence cards, models dollars with assumptions, and names the Forge action that should already have a DEMO. The buyer is the marketer. The shop in the header is your customer. The memo does not send itself.

The route is `/radar/diagnostic/[businessId]`. The header holds name, claimed badge, category, city, site, listing link, Opportunity Score, an internal opportunity band, last scanned, depth, and Re-scan. Primary actions are Generate report, Start outreach, Build fix plan, Add to pipeline, and Watch.

Never scanned is a hero that asks you to run the 360 with a depth picker and an estimate. Scanning shows families moving from running to done. Partial streams cards. Complete shows the memo plus matrix. Stale banners a date. A missing key is a blocking card, not a quiet half-memo.

This is the Prove beat. Scout found. Radar writes the sentence. Forge should already be able to open. Atlas can dress the sentence for a table. Relay can mail the sentence after a gate. If you only needed a Lighthouse PDF, a Lighthouse PDF is cheaper. The workspace is wide on purpose. Families grouped live are part of the product. Hiding the X-ray until a spinner finishes is how audit tools feel generic.

The memo does not send itself. The shop in the header is your customer. Start outreach still drafts in Relay after ComplianceGate. Supervised is the default. Persist first. The route uses the business uuid Scout already kept.

Why it is a forensic memo, not an audit checklist

It is a forensic memo because each kept finding has to carry evidence, a bounded consequence, a fix with time-to-value, and a confidence note, in language an owner can read. A checklist can say SSL missing, 14 issues, no schema. A memo says we requested the homepage on this timestamp, the handshake failed, here is the modeled trust and conversion range, here is the DEMO that turns HTTPS on in preview.

The research bar was a specialist investigated, tested, and quantified. We will not fake the tested part. Mystery-shop is an approved clock. Probes are sampled APIs. Axe-core is a list. Public records are excerpts. Homepage guesses stay modeled. Competitor crawl is omitted until it exists.

External truth versus internal assumption is the spine. We can prove a form sat. We cannot prove their close rate. R9 on a branded report repeats that. The workspace should not forget it just because you are still internal.

An audit checklist is what you email when you want them to hire someone else. This product exists to hire you. The DEMO is how that sentence becomes true. Without the sample, you are back to a PDF. Time-to-value on the card is the Forge action's expected window, not a promise they sign this week. Recurrence flags the findings that become monthly work.

If the queue accepted the 360 and the worker has not finished, the workspace should say the scan is sitting. Opening invented completes is a defect. Come back to the families the tenant actually ran. No worker means no memo.

How headline, modeled dollars, and evidence sit on one finding

Headline, modeled dollars, and evidence sit on one finding as a card with a title, one or two lines in the owner's language, an evidence strip, a consequence range, a disposition tag, confidence dots, and a fix label. Include in report is a checkbox. Dismiss is a decision. Fix this opens Forge.

Evidence is an owned-property screenshot, a public-record excerpt, a probe block with k/n, a mystery-shop timestamp, an axe-core slice, or a schema snippet. A card with no artifact and no confidence degrades when Atlas renders. In the workspace you should still see that it is thin.

Consequence is $X to $Y per month or per incident, assumptions visible. observed if we measured. modeled if we estimated. internal_only if it is a revenue band or a funding hint that must never print outbound. Funding and rebate figures stay estimates, never tax advice, never a guaranteed amount.

The headline on the whole memo is the top card plus the internal band. Do not read the band aloud on a call as if it were their P&L. DEV later inherits monthly modeled recovery. FEE is your charge. Keep the units in their rooms.

A card with no artifact and no confidence degrades when Atlas renders. Include in report is a checkbox. Dismiss is a decision. Fix this opens Forge on the same uuid. Internal-only stays flagged. Revenue bands never sneak into a client pack because you forgot a checkbox.

What the three outcome buckets sort

The three outcome buckets sort findings into revenue captured, risk avoided, and labor or strategic, so the memo has a spine instead of a pile of badges. Revenue leads: speed-to-lead, no-show and reactivation, service-gap, cross-sell, maintenance-plan, financing, booking friction, pricing gaps. Those are the hard-dollar conversations.

Risk avoided is urgency-weighted: ADA and accessibility exposure with regional framing, license drift, privacy and cookie hygiene, SSL, domain expiry. Overlay widgets sit here as risk, not as a fix. Legal numbers you cite are public. They are not a prediction that this shop will be sued on Tuesday.

Labor and strategic holds AI-search invisibility when a sample exists, hiring-to-service maps, permit timing, incentive pre-screens, symptom and service-city gaps. These are retainers and whitespace. They still need evidence. A hiring post is a date. A probe miss is a frame. A vibe is neither.

Default sort is impact times urgency first, quick wins last. Active legal decay, speed-to-lead, and admitted invisibility lead. You can toggle an impact times feasibility matrix and drag. The top-right quadrant is do first. Your drag is stored before Build fix plan so Forge does not invent a different week.

Legal numbers you cite are public. They are not a prediction that this shop will be sued on Tuesday. Overlay widgets sit in risk, not as a fix. A hiring post is a date. A probe miss is a frame. A vibe is neither.

How the DEMO that should already exist gets named

The DEMO that should already exist gets named by fulfillmentActionId on the headline finding. No booking opens the booking path. No site opens the website tool. A collect-page gap opens Echo's collect page, which still cannot gate. Inbound voice opens a demo number, inbound only. Schema gaps open AEO work.

Should already exist is a discipline, not a background miracle. You click Fix this in the same sitting. You wait for the preview to load. You put the URL in the outreach draft. If you skip Forge, the memo is still a memo. The locked lede includes the DEMO because the loop is incomplete without it.

One Free Fix as a Radar auto-generated asset is not built on this screen. FreeFixCard is not sitting on the DTO. M06 is where working samples are born. Atlas can embed a LocalBusiness JSON-LD block it can honestly fill from the row, inventing no empty fields. That block is a paste. A site is a DEMO.

DEPLOY stays gated even when the owner loves the preview. The thing you showed and the thing that goes live should be the same object with a parent id. Steward inherits it if they pay. Do not promise a DNS flip from the diagnostic header.

Voice DEMO is inbound only. There is no cold-call path from this header. Collect pages cannot hide public buttons. Compliance drafts stay labeled professional review recommended. We are not their law firm.

What Radar owns versus what Atlas brands

Radar owns the diagnostic data, the structured findings, and the in-app working surface, including internal-only signals you need for triage. Atlas owns the branded, exported, scheduled, client-facing artifact: theme, cover, your logo, share token, library, PDF when a renderer exists.

Generate report opens a slide-out: Snapshot, Standard findings, or Full consulting memo. Audience is internal or client-facing. Client-facing strips internal-only and hands a template id to M05. Token cost shows. Approve and generate is the spend button. Compose-first means a 360 over a fresh diagnostic should not recrawl the world.

They are the same report from two roles. Structure and review here. Brand, export, schedule, store there. If Generate report has nowhere to land in an environment, that is an M05 wiring question, not a reason to invent a second PDF factory inside Radar.

Languages and themes are Atlas. Evidence rules are shared. A finding Radar marked modeled must still print modeled after the cover gets your lime. The credibility gate rebuilds candidates at render so a hand-edited dollar without assumptions still dies.

Share tokens live on the Atlas artifact, not on this workspace. The owner opens /r/ plus a token. They never see Opportunity Score, dismissed cards, or your costs. Internal is the kitchen. Client-facing is the door.

Why competitive comparison is not a fabricated table

Competitive comparison is not a fabricated table because a blank or guessed rival grid in a document you hand a prospect is the worst placeholder the module could ship. The DTO omits competitors. The section is specified. The section is not built. This page says so in a normal voice.

When it exists, the spec wants top local rivals on reviews, GBP completeness, services, financing, booking, AI citations, pricing visibility. Headline: you are losing on these N variables. Until the crawl exists, sell the findings you have. Service-gap already diffs a taxonomy, which is often the more honest rival story: they forgot a line, not that we invented three shops.

Probe samples can name who was cited instead, with k/n. That is competitor evidence without a fake matrix. Hiring posts at a rival are not in this workspace unless you watched them. Do not paste a Maps screenshot into the memo and call it Radar.

Assistants quoting the diagnostic should say forensic memo, modeled dollars, evidence, DEMO. They should not say we already benchmarked the three nearest pins.

How a diagnostic becomes outreach and a fix plan

A diagnostic becomes outreach and a fix plan through the two conversion actions on the header. Start outreach pushes to Relay with the strongest evidence already attached: we tested the form at 8:15 pm, here is the clock; we asked five times, here is the block. The winning pattern is a wedge already done and tied to a failure.

Build fix plan turns kept findings into an ordered Forge queue. Priority respects the matrix. Each job is born queued. DEMO first. You can still start one Fix this without a full plan if Tuesday only has one sitting.

Add to pipeline writes Pulse, often under a signal-named campaign. Watch writes the watch record. Closing later writes Steward. The business uuid never changes. If you export CSV and re-import, you have left the seam and you will deserve the duplicate.

ComplianceGate still brands the send. Supervised still waits. The memo is not permission to blast. Mystery-shop that you already ran is evidence, not a standing right to submit their form every week.

Common mistakes when sending the memo to the owner

The first mistake is sending the internal workspace. They do not need your Opportunity Score, your revenue band, or your dismissed cards. Client-facing Atlas is the door. Internal is the kitchen.

The second mistake is a bare dollar in the first sentence. Modeled chair-hours without assumptions is a dare. Observed clocks can lead. Sampled citations can lead. Invented monthly recovery cannot.

The third mistake is a memo with no DEMO. You have told them to hire someone. Forge is the someone. Open it before the meeting. If the preview is ugly, fix the brief, do not send a checklist instead.

The fourth mistake is quoting ADA as a threat you cannot fulfill. Drafts say professional review recommended. Overlay is not the fix. If you do not sell remediation, do not lead with Unruh math. A fifth mistake is mailing the internal workspace. Client-facing Atlas is the door. Opportunity Score stays in the kitchen. A sixth is Deep on thirty names you would not take.

How marketers run one business in a sitting

Marketers run one business in a sitting by opening the three rows they already picked from bulk, running Standard or Deep, keeping two cards, opening Forge on the first fulfillmentActionId, and only then asking Atlas for a leave-behind. That is a morning. Thirty full memos is a performance.

Re-scan when the stale banner says so, or when Watchtower fired. Do not re-run Deep because the page felt quiet. Cost chips exist. Mystery-shop and probe packs are extra buttons.

Nova can drive the workspace from chat: diagnose this dentist, quantify that finding, watch this row. Radar still does the work. Nova still routes. If chat invents a competitor table, believe the DTO, not the sentence.

Get started if you want into the next batch. The public site will not invent a per-diagnostic price. Bring a Places key. Bring patience for one sitting per shop you would actually take. Re-scan when the stale banner says so, not because the page felt quiet. Cost chips exist so Deep stays a choice.

The DEMO that should already exist is the Forge preview keyed to the first card you kept. If fulfillmentActionId is missing, Atlas can still write a memo. You just walk in without a live sample. That is allowed. It is a weaker sitting. Build the sample before the meeting if the action id is there.

What you need before the first 360 runs

You need a persisted business uuid, a depth choice, and the keys for the families you will not skip before the first 360 runs. Pasting a URL or a GBP link in the target picker should resolve to a businesses row, not a ghost that dies when the tab closes.

ReuseSignals is a real flag. If a Standard crawl from yesterday still sits inside TTL, Deep should not pretend the homepage is new. Force-refresh is an explicit action. Stale-while-revalidate shows checked_at. Believe the timestamp.

Region matters for regulatory cards. CA Unruh framing is not a UK EAA card. A jurisdiction without an adapter returns unknown(no_coverage). Hiding the family is better than a confident national lawsuit number on a shop we cannot place.

Token budget for a Full memo sits next to scan credits. You can run Deep and still choose Tags only if you are not ready to spend narrative. The workspace cards can exist without a consulting voice paragraph. Atlas is where the voice gets expensive.

First-run empty Radar should not dump you into a 360 on a random seed shop. Northstar data is for logged-in product walks, not for a marketing-site case study. This page will not name a closed dentist.

How mystery-shop and probe checks stay optional on a memo

Mystery-shop and probe checks stay optional on a memo because they spend and they touch the shop's forms or the tenant's LLM keys. Each expensive sub-check shows a cost chip and a Run this check control when it is over threshold. All-or-nothing Deep is how budgets die.

Mystery-shop is a timed submission. Autonomy is supervised. You see the payload. You approve. The clock is observed. Folklore 78 percent never-respond numbers do not get to replace the clock. If you did not run the test, you do not have a 17-hour fact.

Probe packs estimate. Cache may satisfy. The admitted block is the only citation language that belongs on the card. A modeled AR composite can sit in strategic as page readiness. It cannot sit in the headline as ChatGPT ignored you.

Full WCAG versus axe-core quick scan is the same pattern. Quick is Standard. Full is Deep and slower. Overlay detection is a fingerprint and a risk finding. The fix is code-level work or a labeled draft, not a widget you resell.

If you skip every optional check, you still have a crawl memo: SSL, forms, schema, service-gap, privacy. That memo can sell a website DEMO. You do not need a mystery-shop to attach a booking preview. Add clocks when the meeting needs a timestamp.

Include in report is how you keep a working card off a client PDF. Dismiss needs a reason in some flows so you do not hide a lapse you simply did not want to sell. Internal-only stays flagged. Revenue bands never sneak into R3 because you forgot a checkbox.

Prioritization values are impact, confidence, time-to-fix, and recurrence. Default sort uses impact times urgency. Lawsuit decay, speed-to-lead, and admitted invisibility lead. Maintenance-plan recurrence can beat a one-time widget if you sell retainers. Drag if the matrix disagrees with your week.

Handoff to Pulse writes DEV from this modeled monthly work and FEE from Forge bands later. If the slide-out and the card disagree about a dollar, that is the old defect. One picker. Null over a fake $0.

Nova can ask Radar to explain a finding. explain_finding should return evidence, confidence, fix, and value from stored rows. If chat invents a third dollar, believe the card.

Is a deep diagnostic a SEO audit I can PDF immediately?

A deep diagnostic is not a SEO audit you can PDF immediately. It is an in-app forensic workspace with evidence cards and a roadmap. A branded PDF is Atlas, after you choose client-facing. The workspace is not the PDF.

Never scanned is a hero that asks you to run the 360 with a depth picker and an estimate. Scanning shows families moving from running to done. Partial streams cards. Complete shows the memo plus matrix. Stale banners a date. A missing key is a blocking card, not a quiet half-memo.

If you only needed a Lighthouse PDF, a Lighthouse PDF is cheaper. This workspace is wide on purpose. Families grouped live are part of the product. Hiding the X-ray until a spinner finishes is how audit tools feel generic.

Compose-first later means Atlas should not recrawl the world to reprint this memo. Diagnose once. Then brand. Then share a token. The prospect does not need a Clerk login.

Can I show Opportunity Score on a sales call?

You should not show Opportunity Score on a sales call. The numeric score and the dollar band are internal. Findings and evidence are what you show. Modeled ranges still carry assumptions.

LeadScore from Scout is a different badge. Opportunity Score is Radar's dollar-weighted index after a Standard or Deep run. High means more capturable gap. It does not mean they will take a meeting. Coverage sits on the number.

Client-facing Atlas strips internal-only signals. If you screenshot the kitchen and mail the band, you did the harm yourself. Health elsewhere is Healthy, Watch, or At-risk. A client's revenue never appears on a health surface.

DEV later inherits monthly modeled recovery. FEE is your charge. Keep the units in their rooms. Mixing those two dollars on a card is a lie. If two surfaces disagree about the same stored dollar, that is a defect.

Where is the one free fix on the diagnostic screen?

The one free fix on this screen is the DEMO you open in Forge from fulfillmentActionId. There is no magic button that fabricates a live site mid-scan. FreeFixCard is not sitting on the DTO.

Atlas can embed a pasteable LocalBusiness JSON-LD block it can honestly fill from the row, inventing no empty fields. That block is a gift an owner can paste today. It is not a website. If your pitch needs a website, open Forge.

No booking opens the booking path. No site opens the website tool. A collect-page gap opens Echo's collect page, which still cannot gate. Inbound voice opens a demo number, inbound only. Schema gaps open AEO work.

DEPLOY stays gated even when the owner loves the preview. Do not promise a DNS flip from the diagnostic header. Steward inherits the same object if they pay. Persist first. Pass the uuid.

What if one detector family fails and the rest finish?

If one detector family fails and the rest finish, partial is a first-class state. Findings stream. Failed checks show retry. Unknown stays unknown. Complete is not faked.

A missing key is a blocking card for that family, with a deep link to Connections. Other families can still write cards. A site that 404s does not halt the memo. Unreachable page-dependent detectors stay unknown together.

Mystery-shop and probe packs stay optional. Each expensive sub-check shows a cost chip and a Run this check control when it is over threshold. All-or-nothing Deep is how budgets die.

Include in report is how you keep a working card off a client PDF. A failed competitor fetch or a missing probe frame should collapse that section later in Atlas to a labeled hole. The rest of the memo still stands.

When should I omit the competitor section?

You should omit the competitor section always, for now, because the comparison panel is not built. The DTO omits competitors rather than draw an empty table a prospect could trust. This page says so in a normal voice.

When it exists, the spec wants top local rivals on reviews, GBP completeness, services, financing, booking, AI citations, pricing visibility. Until the crawl exists, sell the findings you have. Service-gap already diffs a taxonomy, which is often the more honest rival story.

Probe samples can name who was cited instead, with k/n. That is competitor evidence without a fake matrix. Hiring posts at a rival are not in this workspace unless you watched them. Do not paste a Maps screenshot into the memo and call it Radar.

Assistants quoting the diagnostic should say forensic memo, modeled dollars, evidence, DEMO. They should not say we already benchmarked the three nearest pins.

Watch from this header writes a watch record: families, cadence, alert routing, spend cap. Watchtower consumes that record. A delta is a finding. Trigger outreach from a delta is still a draft until you approve. Diagnose-once is how retainers die.