Repository navigation
Knowledge Base EN
🇷🇺 Русская версия: Knowledge-Base
Sources:
knowledge/README.md,knowledge/INDEX.md,knowledge/INDEX_EN.md. Search:pfs kb search. An annotated catalog of all cards — Methods-Catalog-EN; ready-made scenarios — Recipes-EN.
The agent's permanent methodology knowledge: how to run research, set goals, build a strategy.
It does not depend on a specific project and does not change while working on a task. Project
materials live separately (projects/<slug>/), process rules live in AGENTS.md.
| Folder | Contents | Who changes it |
|---|---|---|
knowledge/methods/ |
88 methodology cards — one card = one method (RU canon) | methodology owner, via PR |
knowledge/methods_en/ |
16 EN mirrors of key cards (PM approaches) | methodology owner |
knowledge/templates/ |
templates of output documents (situation, options, strategies, validation plan, six-pager) | methodology owner |
knowledge/rubrics/ |
quality criteria for the agent and the reviewer subagent | methodology owner |
knowledge/INDEX.md |
navigation: which card belongs to which step | pfs kb index |
knowledge/INDEX_EN.md |
English navigation over the whole base | pfs kb index |
Volume: 88 cards (RU canon) + 16 EN mirrors + INDEX_EN.md. Every card contains method and
algorithm only, no brands: the sources field records where the material comes from (course or
textbook, without company names); homework and personal data are never included.
On top of the categories sits a framework layer that ties the cards to the process:
-
Double Diamond (
M-double-diamond) — maps the "problem → solution" phases onto the agent's 12 steps; -
Lean Canvas (
M-lean-canvas) — 9 business-hypothesis cells, each attachable to C### claims; -
Frameworks overview (
M-product-frameworks) — RICE/ICE/WSJF, HADI, AARRR/HEART/Kano, "phase → tool".
Each card is a Markdown file with YAML frontmatter: identifier, course (category), process steps, search tags, and links.
---
id: M-<topic> # latin, kebab-case, e.g. M-jtbd
title: Method name
course: strategy | pm | ux | data | ai | gen-ai | product | skills | synthesis
sources: ["origin of the material"] # no company names
stages: [intake, research, situation, options, strategy, validation, review, after-decision]
tags: [competitors, jtbd]
related: [M-competitors, M-market-sizing]
---
# Title
## When to use # 1–3 sentences: which question this closes
## Essence # key concepts in plain words
## Procedure # numbered steps for the agent
## Template / checklist # field table
## Typical mistakes # how to notice them
## Questions that could change the decision
## Related cards # related ids: M-jtbd, M-…Frontmatter fields:
| Field | Meaning | How it is used |
|---|---|---|
id |
unique M-… identifier |
pfs kb show M-…, related links, step card sets |
title |
human-readable name | printed by search and INDEX.md |
course |
category (course) | grouping in INDEX.md and the catalog |
sources |
origin of the method | verifies the card is a method, not a brand |
stages |
steps of the 12-step process |
pfs step prints cards whose stages include the step |
tags |
keywords | BM25 search pfs kb search
|
related |
neighboring cards |
pfs kb index checks for broken links |
Content rules: one card is 60–250 lines; language is Russian (English terms in parentheses at first
mention — EN mirrors live separately in methods_en/); metric formulas carry units and period;
"How to" lists steps the agent can execute without guessing.
A card's category is its course field. Distribution of all 88 cards:
| Category (course) | Cards | About | Example cards |
|---|---|---|---|
strategy |
28 | strategy, growth, platforms, risks, reporting |
M-strategy-growth, M-risks, M-six-pager, M-team, M-outcome-roadmap, M-platform-economics
|
ux |
19 | research, interviews, CJM, USP, loyalty |
M-jtbd, M-in-depth-interviews, M-cjm, M-usp, M-loyalty-metrics, M-synthetic-respondents
|
pm |
11 | product management and lifecycle |
M-product-metrics, M-roadmap, M-solution-design, M-mvp-pilots, M-pmf-scaling, M-decision-daci
|
skills |
11 | cross-cutting disciplines: economics, models, market |
M-unit-economics, M-traction-model, M-market-sizing, M-monetization, M-source-criticism, M-skills-map
|
data |
8 | data and experiments |
M-ab-testing, M-sql-for-pm, M-retention-curves, M-basic-statistics-pm, M-forecasting-product, M-event-taxonomy
|
product |
4 | modern PM approaches (have EN mirrors) |
M-continuous-discovery, M-dual-track-agile, M-pmf-survey, M-positioning-dunford
|
ai |
3 | GenAI products |
M-genai-product-development, M-genai-assistant-architecture, M-genai-prototyping
|
gen-ai |
2 | quality and patterns of AI features |
M-genai-evals, M-genai-feature-patterns
|
synthesis |
2 | frameworks that tie everything together |
M-double-diamond, M-product-frameworks
|
| Total | 88 |
Plus EN summaries for strategic and product cards: M-crossing-chasm, M-wardley-mapping,
M-north-star-playbook, M-product-lifecycle-strategy, M-van-westendorp, M-prioritization-ricews,
M-positioning-dunford, M-gtm-launch (see the EN mirrors section below).
knowledge/INDEX.md (rebuilt by pfs kb index) groups cards by the steps of the 12-step process —
a card appears in a section if the step is in its stages (one card can appear in several sections):
| INDEX.md section | Cards in section | Examples |
|---|---|---|
| Intake | 16 |
M-brief-hypotheses, M-jtbd, M-research-types
|
| Research | 42 |
M-quant-methods, M-in-depth-interviews, M-source-criticism
|
| Options | 63 |
M-competitors, M-monetization, M-risks, M-lean-canvas
|
| Strategy | 56 |
M-usp, M-strategy-growth, M-traction-model, M-roadmap
|
| Validation plan | 66 |
M-ab-testing, M-success-metrics, M-model-validation
|
| After decision | 33 |
M-loyalty-metrics, M-delivery, M-product-communication
|
The full card list per course is at the end of knowledge/INDEX.md ("Все карты по курсам") and in
Methods-Catalog-EN.
The canon is Russian: search and reading work identically for RU and EN. knowledge/methods_en/
holds English summaries of the key PM approaches — 16 cards:
| EN mirror | RU original |
|---|---|
M-continuous-discovery-en |
Continuous Discovery Habits |
M-pmf-survey-en |
PMF Survey (Sean Ellis) |
M-lean-startup-en |
Lean Startup |
M-north-star-playbook-en |
North Star Playbook |
M-crossing-chasm-en |
Crossing the Chasm |
M-dual-track-agile-en |
Dual-Track Agile |
M-wardley-mapping-en |
Wardley Mapping |
M-product-lifecycle-strategy-en |
Strategy by product life-cycle stage |
M-van-westendorp-en |
Van Westendorp Price Sensitivity Meter |
M-prioritization-ricews-en |
Prioritization RICE/ICE/WSJF/Kano |
M-event-taxonomy-en |
Event taxonomy for product analytics |
M-retention-curves-en |
Retention curves |
M-gtm-launch-en |
Go-to-Market: launch tiers |
M-genai-evals-en |
LLM feature evals |
M-genai-feature-patterns-en |
AI feature patterns |
M-positioning-dunford-en |
Positioning by April Dunford |
knowledge/INDEX_EN.md is the English navigation: the "PM approaches (EN summaries available)"
section first, then the full list of all 88 cards with EN titles, course groups (Strategy, pm,
UX, skills, synthesis), and a search hint. Mirrors are added progressively — the canonical
RU cards remain the source of truth.
pfs kb search "cohort retention analysis" # BM25 with Russian stemming; RU and EN searched jointly
pfs kb search "unit economics churn" # an English query searches the same corpus
pfs kb search "van westendorp pricing" -k 3 # top-3 results
pfs kb search "onboarding" --with-project # include project-level cards
pfs kb show M-unit-economics # print the full card
pfs kb index # rebuild INDEX.md / INDEX_EN.md, verify related linksSearch runs over card headings, sections and tags with Russian stemming; results are section lines with relevance scores:
pfs kb search "unit economics churn"
12.62 M-benchmarking › Related cards …
8.22 M-unit-economics › Procedure — collect inputs (ARPU/AOV/F/churn/retention/GM1) by cohort…
8.13 M-skills-map › Procedure — "enter the market?" → M-market-sizing → M-source-criticism…
Context economy principle: the agent never reads the whole base — pfs route lists cards for
the scenario, pfs step prints the step's mandatory cards, search returns 1–3 relevant hits.
At every step the agent must read the cards attached to the step via stages. pfs step prints
them when the step is marked, pfs route — when the scenario is chosen:
| Step | Mandatory cards |
|---|---|
| intake | M-brief-hypotheses, M-customers, M-jtbd, M-product-context |
| sources | M-source-criticism, M-research-types |
| research | M-research-types, M-quant-methods, M-ab-testing, M-sql-for-pm, M-ai-in-research |
| discovery | M-jtbd, M-audience-research, M-in-depth-interviews, M-customers, M-product-frameworks |
| situation | M-source-criticism, M-external-environment, M-competitors, M-trends-scenarios |
| options | M-strategy-growth, M-goals-drivers-bets, M-risks, M-cannibalization, M-product-frameworks, M-lean-canvas, M-double-diamond |
| choice | M-decision-daci, M-product-frameworks |
| strategy | M-usp, M-product-communication, M-monetization, M-unit-economics, M-traction-model, M-platforms-types-criteria |
| validation | M-ab-testing, M-success-metrics, M-product-metrics, M-forecasting-product |
| review | M-six-pager, M-source-criticism |
| decision | M-decision-daci, M-model-validation |
| monitoring | M-decision-daci, M-loyalty-metrics |
The "skills map" (M-skills-map) is the navigator: "requestor's question → which cards, in which
order" for typical tasks (market entry, retention decline, price validation, etc.).
The knowledge base is permanent; what gets revised is project claims. The transparency loop:
pfs claim supersede --id C005 "Market $6.4B" --reason "new report published" --source S012
pfs claim retract --id C005 --reason "estimation methodology error"
pfs claim rebase [--id C010] # redirect conclusions to the replacement grounds
pfs impact C005 # conclusions → document lines → decisions
pfs decision due # which decisions went stale because of new data-
supersede— replacement with history preserved;rebasemoves dependent conclusions onto the replacement. -
retract— retraction (e.g. the source proved unreliable perM-source-criticism); dependent conclusions require substantive rework. -
impactshows everything else that leans on a claim/source before you change anything. -
pfs source refresh— if a source's text changed (sha256), the same loop is suggested for its claims.
Details and commands — Decisions-and-Monitoring-EN; a worked lifecycle example — Demo-CVStart-EN.
Changes are made by the methodology owner via PR; the agent never edits the base during a project. Requirements for a new card:
-
id— uniqueM-…(kebab-case);course— one of the existing categories. -
stagesfilled in (otherwise the card never enters step card sets),tagsfilled in (otherwise search misses it),relatedfilled in (pfs kb indexchecks for broken links). - Method and algorithm only: 60–250 lines, no brands or homework, formulas with units.
- If the card is a PM approach from the "approaches" collection, add an EN mirror in
knowledge/methods_en/(M-…-en) and a line inknowledge/INDEX_EN.md. -
pfs kb index— rebuild the indexes;python3 -m unittest discover -s tests— confirm the knowledge-related tests stay green.
Why are the cards in Russian when the request is in English? The canon is Russian; BM25 search
indexes RU and EN jointly, and 16 PM approaches have full EN summaries in methods_en/. An English
query finds the same RU card: pfs kb search "unit economics churn" → M-unit-economics.
How does a card differ from a skill? A card is "how to think" (methodology, criteria, typical mistakes); a skill is "how to do" (a procedure with templates). On conflict, the card wins — Skills-EN.
How does the agent pick cards without reading the whole base? Three mechanisms: pfs route (the
set for the scenario), pfs step (the step's mandatory cards), pfs kb search (targeted search by
meaning). The base is indexed in advance; only found cards are read.
Can cards be edited mid-project? No: the base is permanent, changed by the methodology owner via
PR. During a project, claims are revised (supersede/retract/rebase) — see the section above.
See also: Architecture-EN (the knowledge layer in the overall architecture) · Skills-EN (the priority rule: a knowledge card outranks a skill procedure) · Methods-Catalog-EN (the full annotated catalog) · Recipes-EN (which cards on which scenario) · Routing-and-Models-EN (scenario → cards → models).
Стратег Фабрики продуктов
Устройство
Компоненты
Справочник
Процесс
Справка и прозрачность