The Fourward Method
Moving forward, four layers at a time
Why "Fourward" ?
Naming is hard ... if you know me ... you'd know that I like a good play on word . Fourward is a play on forward , the idea that you move your architecture forward four layers at a time . Not one thing in isolation , not everything at once , but four . A number small enough to hold in your head , large enough to cover the landscape .
The constraint is the method . By forcing every assessment into a 4×4 structure , you avoid the two failure modes of architecture planning : doing too little (a single layer , a single concern) or drowning in a spreadsheet of hundreds of line items that nobody reads .
Besides , none of the existing industry standards allow to navigate around the full picture , some are about visualising the architecture only , others are about managing work , and none are easy to compose .
The 4×4 Principle
At the heart of the method is a matrix : 4 layers across 4 dimensions .
Layers
The horizontal slices of your architecture , the things you build and operate . Edge & Networking , Clients , Services & Platform , and Data .
Dimensions
The lenses you hold up to each layer , the questions you ask . Operational Load , Risk Exposure , Automation , and Supportability .
The default four layers
The method ships with four layers that cover a typical technology stack :
Edge & Networking
Infrastructure at the network boundary . CDN , WAF , DNS , load balancing , API gateways .
Clients
User-facing applications and interfaces . Web apps , mobile apps , CLIs , admin consoles .
Services & Platform
Core services and platform infrastructure . APIs , microservices , queues , orchestration .
Data
Storage , pipelines , and the data layer . Databases , data lakes , ETL , analytics .
These layers are a starting point , not a prescription . Every business has a different architecture . A hardware company might replace "Clients" with "Firmware & Embedded" . A data-heavy organisation might split "Data" into "Storage" and "Pipelines" . A platform team might use "Developer Experience" instead of "Edge" . The method works regardless of what the four layers are , as long as there are exactly four and they are collectively exhaustive for your landscape .
The four dimensions
Every layer is assessed through four lenses . These are the questions you hold up to each slice of your architecture .
Operational Load
How much effort does it take to keep this layer running day to day ? Operational load measures the human cost of operating the system .
- Incident frequency and mean time to resolve
- On-call burden , pages per week , escalation rate
- Toil ratio , manual vs automated operational tasks
- Deployment frequency and rollback rate
AI-driven runbooks , predictive incident detection , autonomous remediation . The signals here are still taking shape .
Risk Exposure
What is the blast radius when something goes wrong ? Risk exposure measures how vulnerable the layer is to failure , security threats , and compliance gaps .
- Single points of failure count
- Security vulnerability backlog and patch cadence
- Disaster recovery coverage and last tested date
- Compliance audit findings and remediation age
Model supply chain vulnerabilities , prompt injection vectors , data leakage through AI tooling . New threat models are emerging fast .
Automation
How much of the build , test , deploy , and operate cycle runs without human intervention ? Automation measures the maturity of your pipelines , self-healing capabilities , and increasingly , the integration of AI-assisted workflows into your engineering practice .
- CI/CD pipeline coverage and pass rate
- Infrastructure as code percentage
- Auto-scaling and self-healing coverage
- Test automation coverage across unit , integration , and E2E
- AI-assisted code review and generation adoption
- Automated incident triage and root cause analysis
- AI-driven test generation and coverage gap detection
- Intelligent alerting , noise reduction , and anomaly detection
Supportability
How easy is it for someone new to understand , debug , and maintain this layer ? Supportability measures the knowledge transfer readiness and long-term maintainability .
- Bus factor , how many people can operate this layer independently
- Documentation freshness and runbook coverage
- Onboarding time for a new team member
- Observability maturity , logs , metrics , traces , dashboards
AI-generated documentation , conversational codebase exploration , context-aware onboarding . Still early days .
The rating scale
Every facet receives a rating on a four-point scale . The scale is intentionally coarse . Fine-grained scores (7 out of 10 vs 6 out of 10) create a false sense of precision . Four levels are enough to drive the right conversations and the right actions .
Every intersection of a layer and a dimension is a facet . Each facet gets a current rating and a target rating . The gap between them is where the work lives .
Example landscape matrix . Each row is a layer , each column a dimension . The badges show current and target ratings .
Why four ?
Four is a cognitive sweet spot . Research on working memory suggests people can comfortably hold about four items in focus at once . The method exploits this :
- Four layers means you can see the full stack without scrolling .
- Four dimensions means you assess from enough angles to catch blind spots , but not so many that the assessment becomes a chore .
- Sixteen facets (4×4) is a landscape you can absorb in a single view .
Four also maps naturally to quarters . A year has four , and each quarter becomes a natural checkpoint for re-assessment . You can align gap activities to quarterly planning cycles , focus one quarter per layer , or rotate dimensions across the year . The number four isn't just a cognitive constraint , it's a planning rhythm .
The constraint also enforces MECE thinking , mutually exclusive , collectively exhaustive . When you only have four slots , you're forced to group things meaningfully . You can't have a catch-all "other" bucket when the limit is four .
Recursive Decomposition
The 4×4 structure doesn't just apply at the top level . Any layer can be decomposed into exactly 4 sub-layers , and each of those sub-layers can be decomposed further . The structure is fractal .
Edge & Networking
├── CDN & WAF
├── DNS & Domains
├── Load Balancing
└── API Gateway
├── Rate Limiting
├── Auth & Routing
├── Transformation
└── Observability At every level of depth , the same 4×4 matrix applies . The same four dimensions . The same rating system . This means :
- Zoom out to see the health of the entire landscape in 16 facets .
- Zoom in to see the health of a specific sub-system in 16 facets .
- Compare across depths using the same vocabulary and scale .
The recursion is opt-in . You only decompose a layer when you need more detail . A layer rated "Strong" across all dimensions probably doesn't need sub-layers . A layer rated "Critical" on two dimensions probably does , you need to know where in that layer the problems live .
The rule of four
Every decomposition must produce exactly four children . Not three . Not five . This constraint forces clarity :
- If you can only think of three , you're missing something , look harder .
- If you have five or more , two of them probably belong together , merge them .
- The exercise of fitting things into four slots is itself valuable . It surfaces assumptions and reveals boundaries you hadn't articulated .
The Assessment Loop
The method follows a repeating cycle :
Map the layers
Define your four layers (or sub-layers) . Name them , describe them , give them icons . This is the structure of your landscape .
Rate each facet
For every layer-dimension intersection , assign a current rating and a target rating . The current rating is an honest assessment of where things stand . The target is the pragmatic level required for sustainability , not aspirational perfection .
Identify gap activities
Where current and target ratings differ , there's a gap . Each gap produces concrete actions prioritised as P0 (immediate) , P1 (next cycle) , or P2 (when capacity allows) .
Act and re-assess
Execute the gap activities . Then re-assess . The landscape is a living document , ratings change as work gets done , and new gaps emerge as the architecture evolves .
Reading the Landscape
The 4×4 matrix is designed to be read at a glance . Here's what to look for :
Patterns across layers
If a single layer is critical or weak across multiple dimensions , that layer needs dedicated investment . It's not just an automation problem or a risk problem , the layer itself is underdeveloped .
Patterns across dimensions
If a single dimension is weak or critical across multiple layers , you have a systemic issue . For example , if "Automation" is weak everywhere , the problem isn't in any one layer , it's a gap in engineering culture or tooling .
The gap count
The total number of gap activities , and their priority distribution , tells you about overall landscape health :
- Many P0s , the architecture is in crisis , focus on stabilisation .
- Mostly P1s , healthy backlog , functional but needs planned investment .
- Mostly P2s , good shape , improvements are incremental .
Build , Iterate , KTLO , Retire
The Fourward Method assumes a lifecycle for every layer :
Build
A new layer is being constructed . Ratings are expected to be weak or critical . Gap activities focus on getting to "Adequate" , functional , safe , and operable .
Iterate
The layer is operational but not yet mature . Ratings are moving from adequate to strong . Gap activities are about hardening , automating , and reducing operational load .
KTLO
The layer is mature . Ratings are strong or adequate with minimal gaps . A small team can own and operate it . The Supportability dimension directly measures KTLO-readiness .
Retire
The layer has reached end of life . It's being replaced , consolidated , or decommissioned . Gap activities shift from improvement to migration planning , data preservation , and safe shutdown . The goal is to exit cleanly without disrupting the layers that depend on it .
The lifecycle is a loop , not a line . Layers don't stay in KTLO forever . Technology evolves , requirements change , and what was "Strong" two years ago may be "Adequate" or "Weak" today . The assessment loop catches this drift . Sometimes that means cycling back to Iterate . Sometimes it means moving to Retire and replacing the layer entirely . Either way , the method surfaces the signal before the decay becomes a crisis .
Principles
- Constraint enables clarity . Four is not arbitrary , it's a forcing function . The method works because of the limit , not despite it .
- Assess before you act . The matrix exists to make the current state visible . Resist the urge to jump to solutions before you've mapped the landscape .
- Gaps are normal . A matrix full of "Strong" ratings is either dishonest or means you're over-investing . Gaps are healthy , they tell you where to focus .
- Depth is earned . Don't decompose a layer into sub-layers until you need to . Start coarse , go deeper where the problems are .
- The matrix is a conversation tool . The most valuable output isn't the matrix itself , it's the conversations it provokes . When engineers , managers , and leadership look at the same 4×4 grid , they can align on priorities without a 50-page document .
- Forward , four at a time . Don't try to fix everything . Pick the highest-priority gaps , execute , re-assess . Progress is iterative .