Forge
markdowne8ad0934
1# World Modeling Playbook v1
2
3## Purpose
4Meta-playbook for keeping knowledge structure faithful to reality: intuition first, formalization second.
5
6## Scope
7- Domain/world boundary decisions
8- Separation of similar-but-distinct ecosystems
9- Router-layer vs world-layer architecture
10- Ongoing correction when reality disproves structure
11
12## Core Principle
13- Do not force reality into a preselected schema.
14- Let distinctions emerge from lived behavior, failures, and diagnostic practice.
15
16## Intuition-First Loop
17- Sense: identify where context "feels different" in causality, language, and action patterns.
18- Probe: run one concrete scenario to test whether that difference is operationally real.
19- Separate: if behavior diverges, keep worlds separate even when labels look similar.
20- Bridge: connect worlds through router layers, not by merging world layers.
21- Recalibrate: if evidence contradicts the map, update the map immediately.
22
23## What Counts as a Separate World
24- Different causal model of how things work.
25- Different dominant failure modes.
26- Different verification/diagnostic workflow.
27- Different vocabulary that cannot be safely collapsed.
28
29## What Belongs to a Router (Not a World)
30- Tool choice guidance.
31- Shared safety constraints.
32- Cross-world interoperability rules.
33- Navigation/index links.
34
35## Anti-Patterns
36- Premature unification "for convenience".
37- Grouping by surface similarity (same category labels) despite behavioral divergence.
38- Turning uncertain intuition into rigid ontology too early.
39- Repeating known misclassification because "it looks cleaner".
40
41## Recovery Protocol After Misclassification
42- Split immediately when divergence is confirmed.
43- Move shared pieces to router layer.
44- Keep each world internally coherent and testable.
45- Record one concise "why split happened" note for future intuition calibration.
46
47## Metrics
48- Number of re-splits caused by over-merging.
49- Time to correct wrong categorization.
50- Cross-world incident confusion rate.
51- Stability of world boundaries over real project cycles.
52
53## Revisit Triggers
54- Frequent "this feels wrong" feedback during categorization.
55- Repeated incidents where one playbook fails in one of the grouped ecosystems.
56- Rising ambiguity in ownership of diagnostics and decisions.
57
58## Calibration Examples (Fast Intuition Check)
59
60### Example 1: Windows vs Linux
61- Surface similarity: both are "OS environment troubleshooting".
62- Divergence signal: service model, permissions, process lifecycle, and diagnostics toolchain differ materially.
63- Decision: two worlds (`windows-environments`, `linux-environments`) + optional router bridge.
64
65### Example 2: Regex vs Parser Logic
66- Surface similarity: both "extract structure from text".
67- Divergence signal: regex is pattern-matching with dialect/overmatch risks; parser logic is grammar/model-driven with different failure surfaces.
68- Decision: keep regex as its own world; parser architecture belongs to another world (language/data tooling), linked via router.
69
70### Example 3: Docker vs Host Runtime
71- Surface similarity: both "runtime and operations".
72- Divergence signal: container isolation, image lifecycle, volume/network boundary, and security model introduce distinct causal chains.
73- Decision: keep Docker world separate from host-OS worlds; connect through deployment/ops router contracts.
74
75### Example 4: Blazor vs Avalonia/XAML
76- Surface similarity: both are ".NET UI".
77- Divergence signal: runtime model (web vs desktop/native shell), rendering lifecycle, state/navigation contracts, deployment and diagnostics differ.
78- Decision: keep separate UI worlds with shared frontend router patterns only where behavior is truly equivalent.
79
80### Example 5: Roslyn Code Actions vs Manual Edits
81- Surface similarity: both can produce the same textual code change.
82- Divergence signal: Roslyn is symbol/semantic aware, while manual edits are text-level and can miss references/overloads/renames.
83- Decision: keep semantic refactoring workflow as its own tooling world; manual patching remains a fallback world with stricter verification.
84
85
View only · write via MCP/CIDE