15 KiB
Employee Onboarding Manual — Digital Operations Partner (v1.1)
Audience: Any employee, contractor, or operator joining the company
Purpose: Give you everything you need to work inside DOP without guessing
Source of truth: The Gitea repository Tony_tech/digital-operations-partner
Status: Active operating guide (reflects system as of 2026-08-02)
Read README “Now vs Later” first. Strategy docs (strategic-frame.md, whitepaper.md) are horizon, not day-to-day SOP.
1. What this company does
Working definition
We continuously prevent Silent Customer Loss for local businesses by making sure customers can find, trust, and engage the business — and that search engines and AI systems can find, understand, and accurately represent it — often before the owner would notice something was wrong.
What Silent Customer Loss means
Problems that cost the business customers without the owner clearly seeing the loss:
- Customer Path Leakage — interest exists, but digital friction blocks contact, booking, visit, or engagement (broken links, no booking button, wrong phone, dead CTAs).
- Discovery Failure — the business is hard to find or is misrepresented in search, maps, or AI answers.
Three service domains
- Customer Path Integrity — can a real person complete the path from interest → contact/book/visit?
- AI Visibility Integrity / AEO — can search engines and AI systems find and accurately represent the business?
- Local Competitive Awareness — what is happening in the local competitive set (visibility, reviews, positioning)?
What we are not (V1)
Do not sell or deliver as core offering unless separately quoted:
- Social media content production
- Paid advertising management
- Website redesign
- Branding projects
- Content marketing campaigns
- Full SEO campaigns
- CRM / email marketing
We are a digital operations oversight business, not a general marketing agency.
2. Non-negotiable operating rules
Memorize these. They govern every task.
-
Agents detect and draft. Humans approve before publish.
Nothing client-facing goes live without a human sign-off in Version 1. -
Evidence Model only
- Tier 1 (Verified): concrete, timestamped, observable, or instrument-backed.
- Tier 2 (Indicative): sampled, directional, probabilistic, or risk-based.
Ambiguous findings default to Tier 2. Agents propose; humans approve final tiers.
-
No invented numbers, threats, or client facts.
If you did not observe it or the owner did not provide it, do not write it as fact. -
Fixed order of work
Users → Problems → Requirements → Workflows → Agents → Tools
Do not design agents or tools before requirements and workflows exist. -
Draft → Human Review → Locked
Every lasting artifact has a status. Never treat a Draft as Locked. -
Prefer the stricter, more evidence-bound position when uncertain.
-
Gitea is the system of record.
Client work, templates, playbooks, and decisions live in the repo under the documented paths. Do not keep the “real” version only in chat or email. -
Competitor claims about other businesses are Tier 2 by default, never the primary threat claim, and require snapshot date + source. Wrong third-party claims are a higher risk class — prefer omit.
3. Where everything lives
| What | Path |
|---|---|
| Repo root / mission | README.md |
| Path to PoC (master sequence) | docs/architecture/path-to-poc-sequencing.md |
| Economics guide | docs/architecture/economics-pricing-execution-guide-v1.md |
| Client lifecycle | docs/business-model/client-lifecycle.md |
| Agent Charter | docs/agents/agent-charter-v1.md |
| Audit Playbook | docs/agents/audit-playbook-v1.md |
| Leonard GBP execution card | docs/agents/leonard-gbp-execution-card.md |
| Templates | docs/templates/ |
| Client instances | docs/clients/<client-name>/ |
| GBP Snapshot HTML form | tools/gbp-snapshot-form.html |
| Task reliability ledger | docs/operations/task-reliability-ledger.md |
| This manual | docs/operations/employee-onboarding-manual-v1.md |
| Strategy (LATER, not SOP) | docs/strategic-frame.md, docs/whitepaper.md |
Client artifact naming (standard):
data-inventory-v1.mdthreat-register-v1.mdeconomics-v1.mdrequirements-v1.mdworkflows-v1.md
Raw snapshots (when stored): data/clients/<client>/raw/...
4. The Path to PoC (how every engagement is sequenced)
Every real client engagement follows these layers in order. Do not skip.
| Layer | Name | What you produce | Human gate |
|---|---|---|---|
| 1a | Public Account Intelligence | Data Inventory (public / Tier 2 ceiling) | Inventory matches what was publicly observable |
| 1b | Owner-access Intelligence | Upgraded inventory (Tier 1 eligible) | Owner-gated fields confirmed |
| 2 | Threat Diagnosis | Threat Register | Each threat verified/downgraded/rejected |
| 3 | Economics & Pricing | Economics doc + pricing bands | Owner conversation required |
| 4 | Requirements | Outcome-based requirements | Scoped to pricing band |
| 5 | Workflow Engineering | Operational workflows | Cadence matches real business hours |
| 6 | Agent Build & Test | Working agents on real data | Human reviews full run |
| 7 | POC Delivery | Demo + gap doc + retainer framing | Owner has seen live draft alert |
Cold audits complete at Layer 1a (or Draft Benchmark). Do not claim full Layer 1 engagement-ready without 1b or explicit Tier 2 ceiling acceptance.
Layer 3 cannot be skipped for paid path. Without owner economics, downstream pricing and scope are speculative and must be labeled as such.
Full detail: docs/architecture/path-to-poc-sequencing.md
5. Day-one cold audit workflow (what you actually do)
This is the current production procedure for a new prospect / first assessment — Layer 1a path.
Step A — Open a client folder
- Create
docs/clients/<client-slug>/if it does not exist. - Add a short
README.mdwith location, phone, website, layer status table, and known public sources. - Copy templates into the client folder as Draft documents:
- Data Inventory
- (Threat Register only after Inventory is ready for human review)
Step B — Public digital baseline (no owner access yet)
Collect only what exists:
- Website (live paths, broken links, contact/booking surface, NAP)
- Public social links from the site
- Review/directory presence (Yelp, etc.) at a factual level
- GBP via the GBP Snapshot Form (see Step C)
Rules:
- Public observation = Tier 2 until owner access upgrades it.
- Record Snapshot Date for every source.
- Missing sources are gaps, not silence.
Step C — GBP snapshot (required human sensor for cold audits)
- Open
tools/gbp-snapshot-form.htmlin a browser. - While looking at the Google Maps / GBP listing, fill every field (2–5 minutes).
- Click Generate for Leonard → Copy or Download .txt.
- Paste the full block to Leonard (or the designated audit agent) with instruction to follow:
docs/agents/leonard-gbp-execution-card.mdexactly.
Leonard must:
- Map intake fields into the Inventory GBP row only
- Fire only threat rules R1–R3 if conditions match
- Not invent fields or write solutions
You (human) then review and approve or correct. Record scored runs in docs/operations/task-reliability-ledger.md.
Step D — Data Inventory (Layer 1a)
Use docs/templates/data-inventory-template.md and the Audit Playbook.
Required minimum sources:
- GBP (or explicit “missing/unclear”)
- Website
- Primary social channel(s)
- Booking/scheduling (or “not visible”)
- Review surfaces if observable
Exit only when a human has signed the review gate → Locked (1a) or Draft Benchmark with blockers.
Step E — Threat Register (Layer 2)
Only after Layer 1a is human-confirmed (or explicitly accepted as Draft Benchmark with documented blockers).
Rules from the Audit Playbook:
- One primary fact per threat
- Evidence Snippet must be concrete
- Data Recency must reference inventory snapshot dates
- No solution language (“fix the 404s” is forbidden here)
- Severity: Critical / Major / Minor — when unsure, choose lower
- Competitor facts: secondary only, Tier 2, snapshot + source
Step F — Stop at the gate
Do not invent economics. Do not write requirements or workflows until Layers 2 and 3 allow it.
If GBP is still incomplete, either:
- Complete the form and update, or
- Explicitly document “GBP unresolved” as a blocker and freeze as Draft Benchmark for training only
6. How to work with agents (Leonard and others)
Roles
- Human operator (you): eyes on Maps/owner access, final decisions, locks, client communication.
- Director / lead (architecture): sequence, standards, repo integrity.
- Leonard (audit agent): form-fill and template-faithful drafting under Playbook + execution cards.
- Aggregator (when used): merge sequential agent outputs; prefer stricter evidence-bound position; no scope creep.
Hard limits for agents
- Stay inside the assigned layer.
- Never self-lock a document.
- Never invent client revenue, WTP, or threat counts.
- GBP ingest without the execution card is out of process.
Scoring Leonard (training)
Use the rubric in docs/agents/audit-playbook-v1.md (0–2 on evidence, tier discipline, scope, recency, template fidelity). Pass ≥ 7/10 with no zeros. Log every scored run in the task reliability ledger.
7. Client lifecycle (commercial path)
High-level path (see full doc for detail):
- Prospect
- Qualification
- Initial Assessment (Layers 1a–2 style work)
- Evidence classification
- Proof-of-Concept fix (low-risk, scoped, human-approved before publish)
- Before/After report
- Retainer offer
- Ongoing monitoring
- Periodic reporting
- Renewal / expansion
- Cancellation / offboarding
GTM motion: proof before retainer. We do not lead with a cold subscription pitch.
Good V1 fit: local service, owner- or small-team operated, depends on local discovery/calls/bookings, meaningful lifetime value. Priority verticals include home services, auto repair, salon/spa (non-medical), boutique fitness, and similar.
8. Economics conversation (Layer 3) — human only
No agent can close Layer 3 alone. You need a direct conversation with the owner covering:
- Current revenue baseline (owner-provided)
- Ad spend if any (owner-provided)
- Owner estimate of what a problem “like this” costs monthly
- Willingness to pay ranges for:
- Band A: detection only
- Band B: detection + draft response
- Band C: fuller escalation with recommended next action
If the owner declines numbers, document Tier 2 speculative and flag risk to any retainer discussion. Do not invent figures.
Guide: docs/architecture/economics-pricing-execution-guide-v1.md
9. Publish / change control
Anything that changes a client-facing surface requires human approval first, including:
- GBP edits
- Review replies
- Q&A answers
- Hours, phone, booking links
- Website or directory updates
- Business/service descriptions
Until a specific task type has a documented reliability run of ≥ 100 error-free executions in docs/operations/task-reliability-ledger.md, the human gate stays on for that task type. Empty ledger = never auto-publish.
10. Communication standards
Internal docs
- Status header: Draft / Human Review / Locked (and 1a vs 1b where relevant)
- Last Updated date
- Evidence tier on factual claims
- Blockers listed explicitly
Client-facing language
- Separate Verified vs Indicative clearly
- No scare tactics from Tier 2 alone
- No guaranteed rankings, AI citations, or “we’ll get you #1”
- Reports answer: what we watched, what changed, what we found, what we fixed/recommended, what remains at risk, what happens next
- Competitor comparisons: provisional, relative, never prosecution from thin Tier 2 alone
11. First-week checklist for a new hire
Day 1 — Orientation
- Read this manual end to end
- Read
README.md(including Now vs Later), Agent Charter, Audit Playbook - Skim Path to PoC sequencing and Client Lifecycle
- Open
tools/gbp-snapshot-form.htmland click through once (empty is fine)
Day 2 — Shadow the system
- Read Overcome Fitness client folder end to end (
docs/clients/overcome-fitness/) - Note what is Draft vs what is still blocked; identify 1a vs full Layer 1
- Identify the open GBP gate and why Layer 3 is blocked
Day 3 — Practice cold audit mechanics
- Fill GBP form against a practice listing (or Overcome Fitness)
- Generate Leonard paste block
- Draft or revise a Data Inventory row from the intake without adding solutions
Day 4 — Threat discipline
- Write 2–3 practice threats that each contain exactly one primary fact + evidence snippet + recency + tier
- Have a lead score them against the playbook rubric and log the score in the ledger
Day 5 — Gate fluency
- Explain aloud when Layer 1a can lock, when 1b is needed, when Layer 2 can lock, and why Layer 3 needs an owner
- List V1 exclusions from memory
12. Current system state (so you know what is real vs aspirational)
Real and usable now
- Mission, evidence model, operating rule
- Full Path to PoC sequence with Layer 1a / 1b
- Templates for Layers 1–2–4–5
- Audit Playbook + Leonard GBP execution card + competitor evidence rules
- GBP HTML form
- Task reliability ledger (counters start at zero)
- First benchmark client instance (Overcome Fitness) in Draft on 1a path
Not built yet / do not pretend exists
- Locked production client package with full economics
- Automated Places API pipeline (optional future)
- Fully automated Leonard → Gitea commit loop without human facilitation
- Standard client-facing report template (still to productize)
- Implementation code under
implementation/(scaffold only) - Public Assessment AI / fixed-fee product stack (strategy horizon only)
When in doubt: if it is not in the repo as a Locked or Active operating doc, it is not process.
13. Escalation
Escalate to the lead/director when:
- A client asks for out-of-scope work (ads, redesign, social production)
- Evidence is ambiguous and a Tier 1 claim is being requested
- Owner access is required to progress and is delayed
- An agent output invents numbers or solutions inside Layer 1/2
- You are unsure whether something is publish-gated
Default action when stuck: document the blocker, do not invent past it.
14. Glossary (short)
| Term | Meaning |
|---|---|
| GBP | Google Business Profile (Maps / local panel listing) |
| DOP | Digital Operations Partner |
| Tier 1 / Tier 2 | Verified vs Indicative evidence |
| Layer 1a | Public baseline inventory (cold audit) |
| Layer 1b | Owner-access inventory (engagement) |
| Locked | Human-approved; safe to build the next layer on |
| Draft Benchmark | Training/reference package; not client-locked |
| WTP | Willingness to pay |
| PoC | Proof of Concept |
| NAP | Name, Address, Phone consistency |
Version: v1.1
Maintainer: Lead / Director
Update rule: When Path to PoC, Playbook, or cold-audit tooling changes, this manual must be updated in the same change set or immediately after.