МАТЧАСТЬ / MVP SCOPE · FINAL ROADMAP / DOCUMENT 43 / 03.09.2026
MVP Scope / Final Roadmap
Финальная сборка исследовательской фазы проекта «Матчасть». Документ сводит решения документов 9–42 в один исполнимый продуктовый roadmap и отвечает на главный вопрос: что именно нужно построить до первой продажи, что разрешено отложить, какие функции добавляются после 10, 30 и 100 платных клиентских пространств, какие технологические и юридические gates нельзя обходить, и в каком порядке собирать систему так, чтобы ранняя версия уже демонстрировала главный moat: verified entity → publication → Search Proof → AI Visibility → Before/After → Next Best Action.
P0 = полный циклне набор экранов, а реальная компания, платный заказ, публикация, измерение и следующий шаг
Human-gatedAI ускоряет precheck, extraction и analysis; финальная публикация проходит human moderation
Transactional first4 понятных SKU, без forced subscription до доказанного repeat
Research completeпосле Doc 43 следующий этап — implementation specification и код
1. Финальная формула продукта
Матчасть = Verified Business Knowledge + Publishing + Distribution + Proof + AI Visibility + Next Best Action.
IDENTITY
company / expert / brand / verified facts
↓
PUBLISH
article / case / research / news
↓
DISTRIBUTE
home / topics / related / digest / Telegram
↓
PROVE
HTTP / canonical / crawler / search / impressions / clicks
↓
MEASURE AI
mentions / recommendations / citations / sources / competitors
↓
COMPARE
baseline / +7 / +30 / +60
↓
ACT
publish / update / verify / distribute / earn external source / wait
↓
MEASURE AGAIN.
2. Что в действительности строит «Матчасть»
Не каталог компаний, не блог, не SEO-биржу и не отдельный GEO dashboard. Ядро — evidence-driven publishing platform, где company identity, source-backed content и measurement принадлежат одному lifecycle.
3. Главный MVP-принцип
Один end-to-end vertical slice раньше широкого набора функций.
REAL COMPANY
↓
BASIC VERIFICATION
↓
ORDER
↓
ARTICLE / CASE
↓
AI PRECHECK
↓
HUMAN MODERATION
↓
PUBLIC SSR URL
↓
PUBLICATION HEALTH
↓
SEARCH PROOF
↓
AI BASELINE / POST
↓
+30 REPORT
↓
NEXT BEST ACTION.
4. Что считается настоящим MVP
MVP exists when one external client can:
1. claim/verify a real company
2. pay for a real SKU
3. submit or create a real material
4. pass moderation
5. receive a public permanent URL
6. see Publication Health
7. receive Search/AI observations
8. receive a methodology-bound report
9. understand the next evidence-backed action.
5. Что НЕ считается MVP
| Не является MVP | Почему |
|---|---|
| Красивый каталог компаний | Нет коммерческого workflow |
| AI Visibility dashboard без publishing | Становимся ещё одним tracker |
| CMS без реального клиента | Infrastructure demo |
| 1000 auto-generated company pages | SEO/site-reputation risk |
| Agency Workspace раньше direct flow | Масштабируем непроверенный процесс |
| Reviews/rating раньше trust system | Высокий fraud/moderation burden |
| Автопубликация LLM | Противоречит CMS/editorial policy |
| Mock dashboard с invented metrics | Не проверяет measurement layer |
6. Непереговорные продуктовые инварианты
NO paid editorial approval guarantee
NO dofollow sale
NO indexing guarantee
NO AI citation guarantee
NO invented source
NO AI as source of truth
NO private cross-client data leak
NO missing provider data represented as zero
NO causal claim from simple Before/After
NO hidden commercial disclosure
NO autonomous final publication decision by AI.
7. 13 модулей: финальный phase map
| Модуль | P0 | P1 | P2+ |
|---|---|---|---|
| Identity Graph | Core entities/relations | Products/services/cases | Advanced graph/benchmarks |
| Verification & Trust | Company/domain/basic facts | Expert/counterparty/credentials | Advanced verification network |
| Publishing Engine | Core | More formats/workflows | API/bulk controlled |
| Editorial Quality | Human moderation + AI precheck | More automation/evals | Advanced risk learning |
| Distribution Engine | Lite: topics/related/home | Digest/Telegram/ranking | Personalization/portfolio |
| Publication Health/Search Proof | Core | Advanced anomalies | Portfolio benchmarks |
| Lead Generation | Basic CTA only | Referrals/forms | Advanced attribution |
| Reputation Layer | Case evidence only | References/credentials/media | Independent reviews/rating later |
| AI Visibility | Core fixed cohort | Sentiment/source categories | Fact Accuracy/advanced datasets |
| Next Best Action | Rules lite | Full action taxonomy | Portfolio optimization/LTR |
| Commerce & Packages | 4 SKU + order/payment | Credits/recurring | Enterprise billing |
| Agency Workspace | Data model ready | Agency Lite | API/SSO/advanced portfolio |
| Editorial & Research | Seed corpus | Research engine | Data products/benchmarks |
8. P0 public surface
/
/articles/{slug}
/cases/{slug}
/research/{slug}
/companies/{slug}
/experts/{slug}
/topics/{slug}
/for-business
/pricing
/methodology
/methodology/ai-visibility
/methodology/before-after
/policies/*
/login.
9. Что не публиковать массово
Не создавать сотни /industries, /cities, /best-X, /compare-X-Y и query-variant pages до появления плотного полезного corpus.
10. P0 client workspace
Overview
Company
Publication
Order / Payment
Publication Health
Search Proof
AI Visibility
Report
Next Action
Team basic.
11. P0 internal moderation/admin
Entities
Verification
Publications
Moderation queue
Orders
Reports
Providers
Exceptions
Audit
Incidents/health links.
12. P0 content types
CUSTOMER:
ARTICLE
CASE
EDITORIAL:
ARTICLE
CASE
RESEARCH
NEWS selectively
Do not block launch on:
interview/opinion/comparison
as separate complex workflows.
13. Почему Article + Case первыми
Они закрывают основной commercial job: экспертная информация и доказанный опыт. Case особенно полезен для future Recommendation Rate/source-gap loop.
14. Исследования на P0
Research нужен собственной редакции как credibility/data moat, но custom paid research не является launch requirement.
15. Коммерческий P0: четыре SKU
| SKU | Pilot price | Workflow |
|---|---|---|
| Publish | 7 900 ₽ | готовый материал → moderation → publication |
| Edit + Publish | 12 900 ₽ | черновик → substantive edit → moderation → publication |
| Create + Publish | 19 900 ₽ | brief/source collection → draft → client fact-check → moderation → publication |
| Publish + Visibility 30d | 14 900 ₽ | publish → measurement window → report |
16. Важное техническое упрощение SKU
В интерфейсе четыре продукта, но backend не строит четыре разных системы. Все используют один Order + Publication lifecycle; отличаются включённые production/measurement services.
17. Не subscription-first
Launch:
one-time transaction
After repeat:
monthly monitoring
After agency evidence:
shared credits / committed volume
After scale:
enterprise/data/API.
18. P0 pricing review gate
After 20–30 paid orders:
measure:
conversion
editor minutes
support
AI cost
refund
margin
repeat
capacity
Only then:
consider 9.9 / 14.9 / 24.9 / 19.9k direction.
19. Contribution targets
Publish:
≥65% target
Edit:
≥60%
Create:
≥50–55%
Automated Visibility:
≥70%
Treat:
targets, not assumed reality.
20. Free acquisition layer
Company search
Public profile
Basic claim/verification
Methodology/examples
Free layer:
acquisition + data quality
Paid:
publishing / production / measurement.
21. P0 Entity Graph depth
ENTITY:
Organization
Person/Expert
Brand
Publication
Topic
Source
RELATIONS:
WORKS_AT
AUTHORED
ABOUT
EXPERT_IN
OFFICIAL_DOMAIN
SOURCE_FOR
PUBLISHED_BY.
22. P1 Entity depth
Product
Service
Client/Vendor relation
Credential
Award
Media Publisher
Case relation
Review relation.
23. P0 verification
company legal identifiers
official domain
representative/domain email
authoritative/public source
basic conflict resolution
verification history.
24. P1 verification
expert relation
counterparty confirmation
case claim
credential
advanced ownership disputes.
25. Source-of-truth split
Legal identifier:
registry / authoritative source
Company description:
submitted claim
Editorial summary:
Mathchast editorial
Publication:
version store
Payment:
commerce
AI observation:
measurement raw record
Verification:
verification subsystem.
26. CMS P0
structured blocks
versions
sources
claims
entities
links
preview
submission
moderation
revision
publish
correction/history.
27. CMS is workflow engine
Publication содержит independent editorial, legal, verification, commercial, publication and measurement states. Оплата не превращает material directly into PUBLISHED.
28. P0 publication flow
ORDER / EDITORIAL INITIATIVE
↓
DRAFT
↓
SUBMIT
↓
AI PRECHECK
↓
HUMAN MODERATION
├─ NEEDS CHANGES
├─ REJECT
└─ APPROVE
↓
LEGAL / COMMERCIAL GATES
↓
READY
↓
PUBLISH
↓
HEALTH / DISCOVERY / MEASUREMENT.
29. Human gate — окончательное решение
Финальный publish decision на P0 остаётся у человека. Это прямое решение Docs 24–25. Автоматизация ускоряет reviewer, но не заменяет editorial accountability.
30. AI Precheck P0
required fields
claim extraction
source coverage
entity candidates
links
duplicate similarity
topic fit
commercial signals
restricted/legal keywords
PII flags
title/style issues
editor summary.
31. AI не делает на MVP
NO final publish
NO final legal verdict
NO synthetic interview
NO invented expert quote
NO source invention
NO autonomous mass page generation
NO silent fact rewrite.
32. После 50–100 материалов
Add carefully:
claim-source matcher
moderation message assistant
question generator
draft from verified brief
semantic claim diff
Context Bridge suggestions
risk/eval automation.
33. P0 distribution
Homepage
Topic hubs
Related materials
Company/expert links
Internal graph
No:
engagement-maximizing endless feed.
34. P1 distribution
Email digest
Telegram
follow topic
save
better related ranking
reader history optional
editorial collections.
35. Publication Health P0
HTTP status
canonical
robots/noindex
schema
sitemap
assets
disclosure
commercial link rel
crawler observation.
36. Search Proof P0
Published
→ Technically Discoverable
→ Crawler Observed
→ Indexed/Searchable Observed
→ Impressions
→ Clicks
→ downstream action where measurable.
37. Search Proof rule
«Отправили IndexNow/sitemap» не равно «проиндексировано».
38. P0 first-party analytics
page view/read
company profile click
official site click
source click
related click
CTA click
referrer taxonomy
AI referral category.
39. P0 AI Visibility cohort
30–50 strategic prompts
2–3 permitted/reliable platform paths
3–5 fixed competitors
one language/region first
repeated baseline
locked versions
scheduled post observations.
40. P0 AI Visibility metrics
Mention Rate
Recommendation Rate
Share of Voice
Citation Rate
Mathchast Citation Rate
Top Sources
Top URLs
Watched URLs
Platform breakdown
Topic breakdown
Prompt Explorer
Coverage/data quality.
41. Не входит в P0 AI Visibility
magic AI score
fictional AI impressions
all private user conversations
hundreds of platforms
persona universe
broad Fact Accuracy
sentiment as headline KPI.
42. Before/After P0
Measurement Plan before intervention
Baseline:
3–5 repeated observations
over ~3–7 days when practical
Post:
+7 early
+30 primary
+60 persistence later
Report:
counts
denominators
percentage points
quality state
confounders.
43. Causal language
MVP reports measured change and temporal association. It does not claim the Mathchast publication «caused +X% AI visibility» without stronger design.
44. New URL Search reporting
Do not show:
"SEO grew from 0 to 400"
Show:
Published
Crawler observed
Searchable observed
First impression
First click
+7/+30/+60 trajectory.
45. Next Best Action P0
Rules first, LLM explanation second.
INPUT:
persistent gap
prompt intent
competitor evidence
source patterns
existing content
entity quality
search health
OUTPUT:
PUBLISH_CASE
PUBLISH_EXPLAINER
UPDATE_EXISTING
VERIFY_ENTITY
FIX_MEASUREMENT
REDISTRIBUTE
WAIT / DO_NOTHING.
46. Next Best Action must be allowed to say «не публиковать»
If:
measurement partial
→ FIX_MEASUREMENT
If:
existing page sufficient
→ UPDATE/DISTRIBUTE
If:
official fact missing
→ VERIFY / client docs
If:
external review/media source needed
→ external action
Only then:
new publication.
47. Recommendation evidence card
Action
Topic
Priority
Evidence quality
Why
Observed gap
Existing coverage
Required sources
Expected hypothesis
Measurement plan
Alternative action.
48. No predicted uplift in MVP
Не показывать «78% chance ChatGPT will mention you». Исторических calibration data для такого обещания пока нет.
49. Commerce P0 objects
order
price snapshot
SKU
payer
client
advertiser
payment
refund state
publication link
measurement entitlement.
50. Billing P0
B2B invoice / bank transfer primary
SBP/card secondary where integrated
Need:
idempotent payment event
order state
refund/manual exception
closing docs.
51. Credits
Credit ledger можно заложить архитектурно в P0, но client-facing package/agency credits становятся приоритетом после первых direct orders или первого agency pilot.
52. Reputation on P0
Basic:
verified company facts
case evidence
external source relations
Not:
public reputation score
open review marketplace.
53. Reputation P1
Partner Reference
Case Confirmation
Credentials
External Media Portfolio
Company Reply
Dispute workflow.
54. Independent Reviews
Запускать позже P1/P2, после антифрода и moderation processes. Public rating — P3, not launch.
55. Agency architecture readiness in P0
Even before UI:
organization_id
workspace_id
membership
delegation
client_entity_id
payer/advertiser separation
tenant tests
Reason:
do not retrofit tenancy later.
56. Agency Workspace P1
agency org
3–10+ clients
workspace switcher
shared credits
basic roles
client viewer/approver
co-branded report
pitch workspace lite
portfolio attention queue.
57. Agency P2
advanced white-label
API/webhooks
SSO
portfolio analytics
bulk controlled workflows
custom reporting
committed volume.
58. Agency never owns client entity
Access remains delegated, scope-based, temporal and revocable.
59. P0 brand/design
Master:
Матчасть / Mathchast
Direction:
technical editorialism
Typography:
IBM Plex Sans + IBM Plex Mono
Palette:
Paper / Ink / Signal Orange
Public:
light-first
Workspace:
light/dark capable
Core motif:
source / evidence / relation.
60. P0 screens
Homepage
Article
Case
Company
Expert basic
Topic
For Business
Pricing
Methodology
Policies
Login
Workspace Overview
Company Claim
Publication Editor
Moderation Status
Order
Publication Health
AI Visibility
Report
Next Action.
61. Не нужны до client #1
full agency portfolio
review center
public rating
industry rankings
advanced semantic explorer
custom dashboards
mobile app
white-label builder
partner directory.
62. Technical architecture P0
Next.js
Fastify / TypeScript API
PostgreSQL
pg_trgm
pgvector available
Redis
BullMQ
MinIO / S3-compatible
Authentik OIDC
nginx
Docker Compose
OpenTelemetry
Prometheus / Grafana / Loki
pgBackRest.
63. Important correction: no FastAPI/Redis Streams requirement
Canonical architecture from Doc 40 is TypeScript/Fastify + BullMQ. Python remains optional for specialized AI/data workers, not the core API.
64. Deployment topology
nginx
├─ Next.js Web
├─ Fastify API
├─ Authentik
│
├─ PostgreSQL
├─ Redis / BullMQ
├─ MinIO
│
└─ Workers
├─ core
├─ crawler/search
├─ AI/provider
├─ report
├─ media
└─ local GPU optional.
65. Modular monolith
13 product domains remain one transactional codebase with explicit module boundaries. Extract services only from measured scaling/failure needs.
66. Deferred infrastructure
NO Kubernetes
NO Kafka
NO Neo4j
NO Elasticsearch/OpenSearch
NO ClickHouse
NO full event sourcing
NO GraphQL requirement.
67. PostgreSQL first
Relational truth
Entity edges
FTS
trigram
JSONB
pgvector candidate similarity
audit/history
measurements early scale.
68. Async by default
Never block HTTP request on:
AI provider
crawler
Search Console
Yandex
PDF
bulk email
media transformation.
Use:
Postgres transactional outbox
→ BullMQ
→ idempotent workers.
69. Local RTX 4080 role
Useful:
embeddings
classification
duplicate detection
precheck
draft assistance
source classification
Not critical dependency:
public site
payment
publishing state
external AI visibility.
70. Security P0 launch gates
MFA privileged
server-side authorization
selected RLS
no public DB/Redis
no Docker socket
service-scoped secrets
SSRF protection
upload validation/quarantine
private/public object separation
CSP/sanitization
audit
off-host backups
restore drill
privacy data map.
71. Russian personal-data launch gate
До production проверить фактическое размещение баз и весь data flow на соответствие требованиям 152-ФЗ, включая localization and external processor/cross-border flows.
72. External AI privacy gate
Private evidence, contracts, unpublished confidential drafts and personal documents не уходят во внешний LLM by default.
73. Backup P0
PostgreSQL:
continuous WAL + base backup / PITR
pgBackRest preferred
Objects:
versioning + off-host copy
Authentik:
DB/config recovery
Secrets:
separate encrypted recovery copy
Restore:
tested.
74. Recovery targets
Initial hypothesis:
DB RPO ≤15 min
core RTO ≤4h
Not SLA:
until restore drills prove them.
75. Monitoring P0
OTel Collector
Prometheus
Grafana
Loki
Alertmanager
Node Exporter
DB/queue/provider metrics
external black-box probe
dead-man heartbeat
status page
backup freshness.
76. Internal reliability targets
Public read:
99.5% rolling 30d target
Workspace:
99.0% interactive success target
No contractual uptime SLA:
until 60–90d evidence
+ restore/incidents
+ support model.
77. Failure isolation requirement
Can fail without public site outage:
AI provider
local GPU
crawler
IndexNow
report worker
email
analytics
Critical:
network/nginx
web/API
Postgres
essential object assets
auth for workspace.
78. P0 GTM readiness
Pricing
Sample publication
Sample report
Methodology
Editorial policy
Commercial policy
Refund/payment
FAQ
For Business
5–20 target accounts
first design partners.
79. Founding customer profile
B2B SaaS
tech/digital services
professional services
PR/content/SEO/GEO agencies
companies with real experts/cases
Avoid:
pure link buyers
mass guest posting
fake reputation
guaranteed ChatGPT buyers.
80. First 100 roadmap
| Stage | Paid workspaces | Главная задача |
|---|---|---|
| P0 | 1–10 | доказать полный loop и willingness to pay |
| P1 | 10–30 | repeat, monitoring, Agency Lite, reputation lite |
| P2 | 30–100 | agency scale, advanced visibility/reputation/API |
| P3 | 100+ | benchmarks, ratings, data products, ecosystem |
81. Client #1 gate
Paid
↓
Moderated
↓
Published
↓
Health valid
↓
Search/AI observed
↓
Report delivered
↓
Next action understandable.
82. Client #1 не проверяет масштаб
Он проверяет, что цепочка технически, редакционно и коммерчески замыкается.
83. Clients #2–5 gate
Same workflow
without custom DB surgery
and without unique code
for every company.
Track:
time
editor workload
support
errors
client confusion.
84. Главный вопрос к client #5
Какие операции всё ещё требуют «магии основателя», и какие из них действительно повторяются?
85. Client #10 gate
Need:
10 paid workspaces
3+ meaningful repeat/expansion actions
AI/report perceived useful
delivery stable
unit-cost data
clear objections
at least one agency signal.
86. Agency signal
Старый roadmap требовал двух агентств, добавивших второго клиента уже к #10. Это слишком жёсткий universal gate. Финальная версия: agency motion считается подтверждённым только когда хотя бы несколько агентств добавляют 2nd/3rd client; если agency пока не основной channel, P1 может стартовать с direct evidence.
87. Если после 10 клиентов нет repeat
STOP broad feature expansion. Не строить ratings/API/enterprise. Диагностировать offer, proof value, pricing and ICP.
88. P1 starts after validated core
Recurring monitoring
Agency Workspace Lite
Shared credits
Pitch Workspace Lite
Partner Reference
Case Confirmation
External Media Portfolio
Scheduled reports
Report share links
Email/Telegram distribution
Better reader search/related
Expanded Next Best Action.
89. Client #30 gate
≥30 paid workspaces
repeat visible
known contribution margin
stable moderation
monitoring attach understood
agency second-client evidence
few severe workflow surprises
security/restore proven
seed corpus credible.
90. P2 after #30
Agency scale
API/webhooks
SSO
advanced reports
Fact Accuracy
advanced source graph
external source auto-discovery
advanced reputation
more regions/platforms
scaling infrastructure only if measured.
91. Client #100 state
100 paid workspaces
10–20 agencies possible
direct channel still meaningful
repeat revenue
monitoring recurring revenue
strong case library
known CAC/ARPO/margin
stable editorial capacity
measured automation/reliability.
92. Bad version of #100
100 one-off backlink orders
no repeat
no monitoring
no agency expansion
huge manual editing backlog
quality incidents
discount dependence
no reader corpus.
93. P3 after proven market
Public ratings
industry benchmarks
rankings
partner directory
advanced portfolio optimization
large-scale anonymized benchmarks
research/data/API products
enterprise HA/SLA where demanded.
94. Build sequence — canonical
FOUNDATION
↓
AUTH / TENANCY
↓
ENTITY / SOURCE / CLAIM
↓
PUBLIC COMPANY
↓
PUBLICATION / VERSION
↓
MODERATION
↓
PUBLIC SSR
↓
COMMERCE
↓
OUTBOX / WORKERS
↓
PUBLICATION HEALTH
↓
SEARCH PROOF
↓
AI VISIBILITY
↓
BEFORE / AFTER REPORT
↓
NEXT BEST ACTION
↓
SECURITY / MONITORING FINAL GATES
↓
CLIENT #1.
95. Почему tenancy раньше Agency UI
workspace_id, membership and server auth are data-model foundations. Agency screen может появиться позже; tenant-safe storage — нет.
96. Почему moderation раньше AI automation
Сначала существует correct state machine and human decision; AI ускоряет уже понятную работу.
97. Почему AI Visibility после real publication
Так measurement строится вокруг настоящего intervention/asset, а не абстрактного dashboard.
98. Почему Next Best Action последним в P0
Без verified entity, inventory and observations recommendation превращается в generic advice.
99. Sprint 0 — Foundation
Hypothesis: 1 week
Deliver:
repo
monorepo rules
Docker dev/staging/prod skeleton
CI
configuration
DB migration framework
logging/request IDs
design tokens
ADRs
basic health.
100. Sprint 1 — Auth / Tenancy / Audit
1–2 weeks
Authentik OIDC
sessions
organizations/workspaces
memberships
roles
audit
negative isolation tests
admin shell.
101. Sprint 2 — Entity / Verification
1–2 weeks
company
expert
brand
aliases
domains
sources
claims
relations
claim company
basic verification
public profile.
102. Sprint 3 — Publishing Core
1–2 weeks
structured editor
publication
versioning
sources
claims/entities
preview
states
SSR
canonical
JSON-LD
sitemap.
103. Sprint 4 — Moderation + AI Precheck
1–2 weeks
precheck schema
duplicate/source/entity signals
moderation queue
reason codes
client revisions
human approve/reject
legal/commercial gates
audit.
104. Sprint 5 — Commerce
~1 week
4 SKU config
order
price snapshot
payer/client/advertiser
invoice/payment
webhook
refund exception
order→publication relation.
105. Sprint 6 — Async foundation
~1 week
transactional outbox
BullMQ
worker classes
scheduler
idempotency
retry/dead jobs
job metrics.
106. Sprint 7 — Publication Health / Search Proof
1–2 weeks
HTTP/canonical/robots/schema
sitemap/IndexNow
crawler observations
GSC
Yandex
timeline
freshness/data quality
first-party events basic.
107. Sprint 8 — AI Visibility
2 weeks
prompt set/version
competitor set
2–3 providers
raw observations
mention resolver
citations
sources
metrics
coverage
platform/topic UI.
108. Sprint 9 — Before/After + Report
1 week
measurement plan
baseline/post windows
paired deltas
counts/denominators
quality state
confounder annotations
report snapshot
HTML/PDF.
109. Sprint 10 — Next Best Action
~1 week
gap taxonomy
preconditions
dedup
rules
priority components
explanation
brief
measurement plan
WAIT/DO_NOTHING state.
110. Sprint 11 — Security / Monitoring / Launch hardening
1–2 weeks
SSRF/upload tests
MFA privileged
RLS selected
secrets
backup/PITR
off-host copy
restore drill
OTel/Prometheus/Grafana/Loki
external probe
alerts
status/runbooks
privacy launch gates.
111. Timeline hypothesis
| Scenario | Functional P0 | Commercially safer beta |
|---|---|---|
| Very focused / AI-assisted | 8–10 weeks | 10–12 weeks |
| Realistic solo/small team | 10–14 weeks | 12–16 weeks |
| Interruptions/integration rework | 14–20 weeks | 16–24 weeks |
Planning ranges, not deadline promises. Legal review, payment integration, AI provider methods and security findings can extend the schedule.
112. What can run in parallel
BUILD:
product
EDITORIAL:
seed corpus
GTM:
design partners / 100-list
LEGAL:
offer / ads / privacy / 152-FZ
BRAND:
wordmark / tokens / templates
OPS:
backups / monitoring / staging.
113. Parallel seed-content target
Launch does not need 100 articles. Target quality first.
Pre-launch minimum:
20–30 strong pieces
5–10 company/expert profiles
3–5 topic hubs
1–3 flagship research/data pieces
several real case/expert formats
all with clean source/provenance.
114. Stretch content target
If editorial capacity allows:
40–60 strong pieces
10+ topic hubs
20+ useful entities
Do not:
publish filler to hit number.
115. Seed themes
AI Visibility
GEO/AEO
B2B marketing
PR
SEO/Search
AI tools
automation
SaaS
business software
content operations.
116. Why these themes first
Они совпадают с ранним ICP и позволяют «Матчасти» dogfood собственные Entity/Source/AI measurement features.
117. Design partners
Before broad launch:
3–5 direct companies
+
3–5 agencies/interviews
Paid pilot preferred:
real invoice/payment
Free:
limited audit/profile
not full production by default.
118. Concierge is allowed
First 1–5 clients may involve:
founder sales
manual configuration
manual report explanation
manual moderation
manual legal escalation
Not allowed:
manual hidden metric invention
direct DB editing as normal workflow
fake verification
fake AI observations.
119. Manual work must create a future system rule
Repeated founder action?
↓
Document
↓
Template / form / rule / tool
↓
Automate where safe
↓
Keep human gate where accountability matters.
120. Automation philosophy
Automation removes preparation and analysis work; it does not erase editorial accountability.
AI:
extract
classify
summarize
draft
compare
recommend candidates
Deterministic system:
auth
payment
states
formulas
disclosure
links
retention
health gates
Human:
final publication
borderline legal/editorial
identity dispute
high-risk claims
security/privacy response.
121. Automation KPI
Measure:
AI precheck usefulness
editor time/material
revision count
false flags
missed issues
report generation time
Do not target:
95% auto-publish
because policy says human final gate.
122. Human scaling goal
Human sees:
compact review packet
critical claims
sources
risk flags
commercial/legal state
AI summary
Not:
14 tabs of routine data.
123. First editor staffing hypothesis
0–10 paid/month:
founder/product + 1 editor possible
20–30 paid/month:
dedicated editor + senior/founder
50–100 paid/month:
2–4 editors depending format mix
Measure:
minutes by SKU.
124. Technical Definition of Done
works
tests
authorization
audit
metrics
loading/error/empty
accessibility
responsive where relevant
migration
rollback/forward-fix
no secret leak
documentation.
125. AI feature DoD
prompt/version
structured schema
eval cases
failure state
tool scope
privacy classification
provider cost
audit
fallback
human-use context.
126. Background job DoD
idempotent
timeout
retry/backoff
failed/dead state
metrics
trace/job ID
recovery
runbook if critical.
127. Connector DoD
authentication
quota
rate limits
retry
raw response
normalized schema
freshness
provider-change state
tests
missing ≠ zero.
128. Publication page DoD
SSR
canonical
robots
JSON-LD
sources
disclosure
versions
responsive
accessible
OG
sitemap
health check.
129. Private workspace DoD
server auth
workspace isolation
no shared cache
sensitive audit
negative access test
error/loading/empty states
current client context visible.
130. Testing priority order
1 Tenant isolation
2 Payment/order invariants
3 Publication lifecycle
4 Source/claim integrity
5 AI measurement correctness
6 Provider missing/error states
7 Backup/restore
8 SSR/machine readability
9 UX regression.
131. Must-pass tests before client #1
Client A cannot read B
App DB role cannot migrate/drop
Duplicate payment webhook is safe
Duplicate job is safe
Publication version remains stable
Article renders without JS
Bad source/URL handling is safe
SSRF private targets blocked
Missing AI provider result ≠ zero
Backup restores.
132. Launch technical checklist
[ ] domain/TLS
[ ] SSR public pages
[ ] canonical
[ ] sitemap
[ ] robots
[ ] JSON-LD
[ ] OIDC/login
[ ] tenant authorization
[ ] moderation state machine
[ ] payment/order
[ ] outbox/BullMQ
[ ] Publication Health
[ ] AI provider policy
[ ] audit
[ ] backup/PITR
[ ] off-host copy
[ ] restore test
[ ] external monitor
[ ] alerts/status page.
133. Launch legal/editorial checklist
[ ] editorial policy
[ ] commercial policy
[ ] link policy
[ ] correction/takedown
[ ] moderation reasons
[ ] ad marking workflow
[ ] advertiser/payer model
[ ] offer/refund
[ ] privacy notice
[ ] personal-data map
[ ] processor register
[ ] 152-FZ location review
[ ] external AI data policy
[ ] security contact.
134. Launch content checklist
[ ] homepage not empty
[ ] 20–30 quality pieces
[ ] useful company profiles
[ ] topic structure
[ ] flagship research
[ ] sample commercial publication
[ ] sample report
[ ] methodology pages
[ ] source/provenance visible.
135. Launch sales checklist
[ ] 4 pilot SKU
[ ] pricing page
[ ] first 100 account list
[ ] 15+ discovery conversations underway
[ ] sample report
[ ] FAQ/no guarantees
[ ] procurement/invoice path
[ ] first 3–5 paid pilot candidates
[ ] agency one-pager basic.
136. Launch reliability checklist
[ ] Prometheus/Grafana/Loki
[ ] external probe
[ ] dead-man heartbeat
[ ] DB/queue metrics
[ ] backup age
[ ] WAL age
[ ] provider freshness
[ ] incident template
[ ] runbooks
[ ] severity model
[ ] restore drill completed.
137. Day 30 objective
Company/entity works
Publication lifecycle works
Human moderation works
Public SSR article works
Sources/schema visible
Basic workspace works
Own seed content published.
138. Day 60 objective
Commerce live
Publication Health live
Search Proof basic
AI baseline basic
first paid pilot
security/backup functional.
139. Day 90 objective
+30 report
Next Best Action
5–10 paid workspaces
first repeat/expansion
agency pilot signal
stable monitoring/runbooks.
140. 4–6 month objective
20–40 paid workspaces
recurring monitoring
Agency Lite
shared credits
reputation lite
better distribution
flagship research cadence
known best ICP.
141. 6–12 month objective
50–100 paid workspaces
healthy agency cohort
repeat revenue
monitoring recurring revenue
known contribution by SKU/channel
stable moderation staffing
API/enterprise only if demanded.
142. First 10 GTM playbook
0–10:
founder-led
warm introductions
researched outbound
paid pilot
high-touch learning
Track:
reply
qualified
paid
SKU
objection
delivery
repeat.
143. 10–30 GTM
Narrow ICP
stabilize offer
first cases
referrals
agency pilots
productized audit
monitoring attach.
144. 30–60 GTM
Agency channel
research-driven inbound
scheduled reports
repeat process
second/third client per agency
pricing lock.
145. 60–100 GTM
Document repeatable motion
delegate repeatable sales
scale healthy channels
first sales hire only if economics/capacity support.
146. Product metrics 1–10 clients
paid workspaces
time to publish
editor minutes
revision count
report delivery
measurement coverage
client comprehension
second-action intent.
147. Product metrics 10–30
repeat within 90d
Visibility attach
recommendation action rate
agency second-client
contribution margin
support cost
moderation capacity.
148. Product metrics 30–100
CAC by channel
ARPO
retention
expansion
monitoring MRR
agency client yield
editorial capacity
SLO/reliability
refund
margin.
149. Strong PMF signals
Client buys second/third action
without deep discount
Agency adds second/third client
+30 report leads to a real next decision
Publish + Visibility share rises
Clients reference Mathchast URL externally
Seed research earns links/citations/readers
Revenue grows without proportional founder labor.
150. Weak signals
lots of registrations
but no paid
traffic only from founder posts
clients only ask dofollow
Create+Publish dominates
because clients want outsourcing agency
AI report admired but never purchased again
agencies trial once and stop.
151. Kill / rethink criteria
| Observation | Action |
|---|---|
| Большинство лидов хочет только ссылку | Пересобрать ICP/message/channel |
| No repeat after report | Revisit core value/recommendation loop |
| Visibility attach ≈0 | Check whether AI layer is differentiator or packaging failure |
| Agency never adds second client | Do not overinvest in agency product |
| Create margin <50% | Raise price/narrow scope |
| Human moderation bottleneck severe | Improve precheck/forms or staffing; do not remove gate blindly |
| Security/privacy launch gate unresolved | Do not scale paid usage |
| Restore test fails | Block commercial scale until recovery works |
152. Reader-side success
search/discover
→ useful article
→ entity/topic
→ related
→ return
Metrics:
organic discovery
meaningful reads
entity navigation
repeat reader
source clicks.
153. Data moat loop
more quality clients
↓
more verified entities
↓
more claims/sources
↓
more publications
↓
more Search/AI observations
↓
more outcome history
↓
better recommendations
↓
higher repeat.
154. Cross-client learning restriction
Moat не строится на утечке client strategy. Cross-client learning только aggregated/de-identified и только в рамках terms/privacy.
155. Competitive moat at P0
Not:
domain authority
huge audience
cheap article
Moat seed:
verified entity model
source provenance
durable publication
Search Proof
AI measurement
transparent methodology
next action.
156. Competitive moat at scale
entity graph
verified relationships
quality corpus
historical Search/AI observations
intervention/outcome history
agency workflow
original research
distribution
reputation evidence.
157. Founder dashboard P0
Orders
Moderation queue
Blocked publications
Provider health
Reports due
Backup status
Incidents
Sales pipeline link
Unit economics snapshot.
158. Not an exception-only automation company
Старый draft roadmap чрезмерно смещал проект к «AI runs everything, human only exceptions». Финальная модель следует Docs 24–25: human editorial gate remains deliberate product quality control, while routine preparation/analysis is automated.
159. Recommended team for P0
Founder/Product/GTM
1 strong builder
1 editor or founder+editor hybrid
legal/privacy external
security review external/on-demand
design support
AI assists development/editorial.
160. What not to hire before evidence
SDR team
5 editors
full-time DevOps team
data-science team
community moderation team
enterprise customer success department.
161. First sales hire gate
Consider when:
30–50 paid workspaces
same ICP closes repeatedly
sales script known
founder overloaded
margin can fund role
editorial capacity exists.
162. First editor expansion gate
Trailing 4–6 weeks:
70–80% sustainable review capacity
+
growing queue
+
quality stable
Then:
add capacity.
163. Infrastructure scaling gate
External search engine:
only measured search bottleneck
ClickHouse:
only OLAP/event interference
Kafka:
only queue/event throughput need
Kubernetes:
only multi-host/service orchestration need
Second production node:
when SLO/SLA/revenue justifies HA.
164. Search implementation P0
exact company
aliases
prefix
pg_trgm
Postgres FTS
pgvector:
candidate similarity/dedup
advanced ranking later.
165. No paid ranking boost in public search
Платный tariff не делает company/entity более релевантным public search query.
166. Public methodology P0
AI Visibility
Before/After
Publication Health concepts
editorial/commercial disclosure
corrections
Explain:
sampling
limits
provider changes
no guarantees.
167. Trust-building rule
Mathchast should expose more methodology than competitors' opaque score dashboards, not less.
168. Public report examples
Use:
real permissioned
or clearly SAMPLE
Show:
denominators
period
method
limitations
what changed/not changed.
169. No «all-green» case library
Flat/mixed result can strengthen credibility if methodology and next action are useful.
170. P0 legal boundaries in sales
Can promise:
defined workflow
moderation
public URL policy
measurement scope
Cannot promise:
index
position
traffic
AI citation
leads
revenue
editorial approval.
171. P0 support
Email/support channel
order status
moderation reason codes
payment/refund path
report explanation
No:
24×7 enterprise SLA.
172. Internal SLO before SLA
Первые 60–90 дней reliability telemetry формируют будущий commercial SLA. До этого наружу не продаётся недоказанный uptime percentage.
173. Final P0 scope — compact
MUST:
Identity / basic verification
Article + Case publishing
Structured CMS/versioning
Human moderation
AI precheck
4 transactional SKU
Payment/order/refund basics
SSR public pages
Structured data/sitemap
Publication Health
Search Proof
First-party analytics basic
30–50 prompt AI Visibility
2–3 platforms
Before/After descriptive report
Next Best Action lite
Audit
Security/privacy launch gates
PITR/off-host backup
External monitoring.
174. Final P0 exclusions — compact
NOT P0:
Open comments
Public ratings/rankings
Full independent reviews
Full agency CRM
Enterprise SSO/SCIM
Public developer API
Mass content import
Autonomous LLM publishing
Neo4j
Elasticsearch
Kafka
ClickHouse
Kubernetes
Mobile app
100 industry pages
Large product catalog
Universal AI Score
Guaranteed GEO.
175. P1 compact
Recurring monitoring
Agency Lite
Pitch Workspace Lite
Shared credits
Scheduled/co-branded reports
Partner references
Case confirmations
Credentials/media portfolio
Reader follow/save/digest
Telegram
Expanded Next Best Actions
More automation/evals.
176. P2 compact
Agency scale
API/webhooks
SSO
advanced reputation
independent reviews mature
Fact Accuracy
source graph
regional/platform expansion
portfolio analytics
enterprise billing/support
infra scaling where measured.
177. P3 compact
Public ratings/rankings
industry benchmarks
partner directory
advanced data products
cross-market research
portfolio recommendation optimization
higher-availability architecture
contractual enterprise SLA.
178. Final roadmap table
| Phase | Product | Commercial proof | Gate to next |
|---|---|---|---|
| P0 | Full publish→measure→act loop | 1–10 paid workspaces | repeat/value + stable delivery |
| P1 | Recurring + Agency Lite + reputation lite | 10–30 | known margin + agency/repeat evidence |
| P2 | Scale/enterprise/advanced AI/reputation | 30–100 | repeatable GTM + operational scale |
| P3 | Benchmarks/rating/ecosystem/data products | 100+ | market depth and corpus maturity |
179. Главный roadmap risk
Строить платформу горизонтально. Сделать 20% Entity Graph, 20% CMS, 20% Agency, 20% reviews, 20% AI — и ни одного закрытого клиентского цикла.
180. Правильный способ реализации
Вертикальными slices.
Slice A:
Company → public profile
Slice B:
Draft → moderation → public article
Slice C:
Order → paid publication
Slice D:
Publication → Search Proof
Slice E:
Prompt Set → AI Visibility report
Slice F:
Report → Next Action
Then:
repeat / agency / reputation.
181. Финальный Definition of Product
Если «Матчасть» умеет только публиковать, это молодое медиа. Если умеет только измерять AI, это молодой tracker. Продукт становится «Матчастью», когда verified identity, publication, Search/AI evidence and next decision работают как одна цепочка.
182. Что заканчивается этим документом
Research phase:
COMPLETE
We have:
positioning
ICP
competitors
legal/editorial
commercial model
IA/entity
verification
CMS
AI editor
distribution
reader
client workspace
pricing/billing
Publication Health
AI Visibility
Before/After
Next Action
Reputation
Agency
GTM
Brand
Architecture
Security
Monitoring
Final MVP roadmap.
183. Что начинается дальше
Implementation phase. Следующий документ, если продолжать серию, должен быть не новым исследованием, а MASTER PLAN реализации: repo, migrations, API contracts, screens, background jobs, prompts, tests, milestones и atomic tasks для AI-assisted coding.
NEXT:
Mathchast_44
MASTER PLAN РЕАЛИЗАЦИИ
Then:
implementation documents / code.
184. Финальное решение
Исследовательскую архитектуру «Матчасти» можно считать завершённой. Строить нужно узкий P0, который уже содержит настоящий commercial/evidence loop: real company → basic verification → one of four transactional publication SKU → AI-assisted precheck → human moderation → durable SSR publication → Publication Health/Search Proof → fixed prompt AI Visibility → descriptive Before/After → evidence-backed Next Best Action. Identity, workspace tenancy, audit, security, privacy and recoverability закладываются сразу, но broad Agency, Reputation, Reader and Enterprise functionality появляются только после подтверждённого core value. Canonical technical stack: Next.js + TypeScript/Fastify + PostgreSQL + Redis/BullMQ + MinIO + Authentik + Docker Compose, with optional Python/GPU workers, transactional outbox, off-host PITR backup and external monitoring. AI accelerates extraction, drafting, review preparation and analytics, but final publication remains human-gated. Launch commercial model remains transactional-first with 7 900 / 12 900 / 19 900 / 14 900 ₽ pilot SKU; recurring monitoring and agency credits follow repeat evidence. P1 starts around 10 validated paid workspaces, P2 around 30, and broad ecosystem/rating/benchmark features only after roughly 100 proven workspaces. The next useful work is no longer research: convert this exact scope into an implementation master plan and begin the first vertical slice.
185. Внутренние источники синтеза
| Блок | Основные документы |
|---|---|
| Positioning / ICP / competition | 9–12 |
| Legal / editorial / commercial policy | 13–17 |
| IA / entity / verification / machine readability | 18–22 |
| Content / CMS / AI / distribution / reader / seed | 23–28 |
| Client / pricing / billing | 29–31 |
| Search / AI measurement / causality / recommendations | 32–35 |
| Reputation / agency / GTM | 36–38 |
| Brand / architecture / security / reliability | 39–42 |
Документ 43 является синтезом уже принятых проектных решений. Сроки, client-count gates, contribution targets и launch metrics остаются рабочими гипотезами и должны заменяться реальными данными после первых заказов. При конфликте старых черновиков с финальными domain-документами приоритет имеют последние специализированные решения: Docs 24–25 для human moderation, Doc 30 для 4 SKU, Doc 39 для IBM Plex visual system, Doc 40 для Fastify/BullMQ architecture, Docs 41–42 для security/recovery/SLO.