Skip to content

HL — TFW-25: Values & Principles Consolidation

Date: 2026-04-04 Author: Coordinator (AI) Status: 📝 HL_DRAFT — Updated after RESEARCH RES: RES-TFW-25

1. Vision

TFW accumulated 14 principles and 33 decisions across 24 tasks. They live in 3 separate places (README Values, KNOWLEDGE.md §0, knowledge/ topic files) with heavy duplication and no clear hierarchy. The README Values section — the philosophical face of TFW — is stuck at v1 level with 5 generic entries that don't reflect the framework's real values.

Impact: New agents and users read README Values and get a shallow picture. Real principles are buried in KNOWLEDGE.md tables. Knowledge files duplicate what's already self-evident from conventions and workflows.

"The values section should tell you what TFW believes. Right now it tells you what any framework believes."

2. Current State (As-Is)

Three-layer duplication

Layer File Lines Role
Framework values .tfw/README.md §Values 26 lines, 5 items Public philosophy
Project principles KNOWLEDGE.md §0 17 lines, 14 items (P1-P14) Internal principle registry
Verified facts knowledge/*.md 57 lines, 29 facts Knowledge consolidation output

Duplication map (6 concepts appear 2-3 times)

Concept README Values KNOWLEDGE P# knowledge/ fact conventions/workflows
Anti-sycophancy ✅ Candor Over Flattery P9 (coordinator mindset) philosophy/F3 AGENTS.md conduct
Ref-inside-step — P11, P12 convention/F5 plan.md, base.md (inline)
Naming > Explanation — D28 process/F5 —
Filesystem = state — P14 convention/F9, philosophy/F4 conventions §4, base.md
Token density — P10 constraint/F2 base.md limits table
4 roles — D30 process/F9 conventions §15, glossary

P1-P14 classification (not all are "principles")

Tier Items What they really are
Values (why TFW exists) P1, P3, P7, P9 Traces > code, narrative > DRY, role separation, quality > speed
Design principles (how TFW is built) P2, P5, P8, P14 Index don't duplicate, meta-awareness, research ≠ checklist, filesystem = gate
Implementation rules (how to follow TFW) P4, P6, P10, P11, P12, P13 Glossary/conventions split, lightweight docs, token density, inline enforcement, DNA/library, progressive disclosure

Problem: Implementation rules (P10-P13) are recorded as "principles" but they are engineering patterns. They belong in conventions.md, not in a philosophy section.

KNOWLEDGE.md bloat analysis (179 lines)

Section Lines Redundancy
§0 Principles (P1-P14) 17 Many are implementation details, not principles
§1 Architecture Map 12 Duplicates file listing in conventions.md §2
§2 Key Artifacts 20 Links to task files — useful index
§3 Legacy & Deprecation 38 Growing linearly with every task. 50% resolved items
§4 Tech Stack 8 4 lines of content
§5 Project Facts 10 Index to knowledge/ topic files

§3 Legacy = 38 lines, 17 resolved. Resolved items = historical but take space. §4 Tech Stack = trivially obvious from the repo.

knowledge/ topic files (29 facts)

Some facts are now self-evident from the code they describe: - convention/F9 (filesystem state machine) = obvious from conventions.md §4 + base.md Step 0 - convention/F10 (RES = synthesis) = obvious from templates/RES.md structure - process/F9 (4 roles) = obvious from conventions.md §15 - process/F10 (Resume Protocol) = obvious from base.md Step 0

3. Target State (To-Be)

3.1 Result Visualization

Principle hierarchy (after consolidation):

.tfw/README.md §Values and Principles  ← PUBLIC: what TFW believes (5-8 items, narrative)
  │
  ├── Traces Over Code                   (was: P1, exists)
  ├── Candor Over Flattery               (exists, enrich with P9)
  ├── Completeness Over Speed            (exists)
  ├── Structural Enforcement             (NEW: from P14, P8, philosophy/F4)
  ├── Naming Creates Behavior            (NEW: from [D28](../../knowledge-index.md#architecture-decisions), process/F5)
  ├── Single Source of Truth             (exists)
  ├── Portability                        (exists)
  └── Determinism and Safety             (exists)

KNOWLEDGE.md §0 Principles              ← INTERNAL: battle-tested design rules (compact)
  │
  ├── P1–P3 (philosophy tier) → STAY (compact)
  ├── P4–P6 (implementation) → REMOVE (obvious from conventions.md)
  ├── P7–P9 (roles/quality) → MERGE into README Values / STAY compact
  ├── P10–P13 (engineering patterns) → MOVE to conventions.md or REMOVE
  └── P14 (filesystem state) → MOVE to README Values as "Structural Enforcement"

KNOWLEDGE.md §3 Legacy                  ← PRUNE: remove resolved items
KNOWLEDGE.md §4 Tech Stack              ← REMOVE: obvious from repo

knowledge/ topic files                   ← PRUNE: remove facts now obvious from code

Before → After sizes:

                        BEFORE    AFTER
KNOWLEDGE.md            179 lines  ~115 lines  (-36%)
  §0 Principles         17 lines   ~8 lines    (14 → 7 items)
  §3 Legacy             38 lines   ~20 lines   (35 → 17 items)
  §4 Tech Stack         8 lines    0 lines     (removed)
README.md §Values       26 lines   ~40 lines   (5 → 8 items)
knowledge/ facts        29 facts   18 facts    (-38%)

4. Phases

Phase A: Audit & Consolidate 🔴

  1. .tfw/README.md §Values — rewrite to 8 items: add "Traces Over Code", "Structural Enforcement", "Naming Creates Behavior". Rename "Determinism and Safety" → "Honesty Over Convincingness" (rewrite as genuine value, not rule list). Enrich "Candor" with P9 coordinator mindset
  2. KNOWLEDGE.md §0 — prune from 14 → 7 items. Remove P4, P6, P10-P13. Promote P14 to README. Compress remaining to 1-liners
  3. KNOWLEDGE.md §3 — remove 18 resolved Legacy items (keep 17 recent: TFW-22+)
  4. KNOWLEDGE.md §4 — remove entire section
  5. knowledge/*.md — prune 11 self-evident facts (6 convention, 5 process). Keep convention/F5 (named pattern value per D28)

Phase B: Cross-reference & Apply 🟡

  1. .tfw/conventions.md — absorb P10-P13 content into existing sections or add §X "Design Rules"
  2. AGENTS.md — verify alignment with updated values
  3. Adapters — sync

5. Definition of Done (DoD)

  • ✅ 1. README §Values has exactly 8 values including "Traces Over Code", "Structural Enforcement", "Naming Creates Behavior", "Honesty Over Convincingness"
  • ✅ 2. KNOWLEDGE.md §0 has exactly 7 principles (no implementation details)
  • ✅ 3. KNOWLEDGE.md §3 Legacy: 18 resolved items removed (17 remain)
  • ✅ 4. KNOWLEDGE.md §4 Tech Stack: section removed
  • ✅ 5. KNOWLEDGE.md total ≤120 lines
  • ✅ 6. knowledge/ topic files: 18 facts (11 self-evident ones removed)
  • ✅ 7. No principle/value appears in >2 places with full text
  • ✅ 8. conventions.md or README absorbs P10-P13 content
  • ✅ 9. Cross-references work (no dead refs)

6. Definition of Failure (DoF)

  • ❌ 1. Values become a list of 15+ items (values ≠ encyclopedia)
  • ❌ 2. Pruning removes something that's NOT obvious from other artifacts
  • ❌ 3. README Values section becomes longer than Anti-patterns section below it

On failure: Restore from git. Values = 8 max. If in doubt, keep in KNOWLEDGE.md rather than promote to README.

7. Principles

  1. Values = beliefs, not rules — README Values answers "what does TFW believe?" not "how does TFW work?"
  2. Promote up, don't duplicate — if a fact is now obvious from code/conventions, remove from knowledge/ rather than keeping "for completeness"
  3. Compact > comprehensive — 8 strong values > 15 complete ones

8. Dependencies

Dependency Status
TFW-22 (source of P12, P13, D28) ✅ Done
TFW-24 (source of P14, D30-D33) ✅ Done
TFW-18 (knowledge consolidation infrastructure) ✅ Done

9. Risks

Risk Probability Impact Mitigation
Pruning removes something valuable Low High Git history. If in doubt, compress rather than delete
README Values grows too large Medium Medium Hard cap: 8 values max. "Determinism and Safety" = merge candidate
P10-P13 migration creates dead P# refs in task HLs Low Low Historical refs stay valid (source column unchanged)

10. RESEARCH Case

Blind Spots

  • How do other frameworks (LangChain, CrewAI docs, Cursor rules) structure their values/principles sections?
  • Is the 3-tier split (values/design/implementation) the right taxonomy, or is 2-tier better?

Hypotheses

# Hypothesis Status
H1 P10-P13 are implementation rules, not principles — they belong in conventions.md ✅ confirmed (RES R3)
H2 ≥5 knowledge/ facts are now self-evident from code and can be safely removed ✅ confirmed — 11 facts (RES R4)
H3 README Values section should stay under 8 items to maintain narrative impact ✅ confirmed — exactly 8 (RES R2)

Filter: Each hypothesis: "If proven false, would our approach change?" H1: Yes — if they are principles, they stay in §0. H2: Yes — if not self-evident, we keep them. H3: Yes — more items might be needed.

Risks of Not Researching

Task is mostly internal audit — facts are in the codebase. Low risk of unknown unknowns. External research could validate the tier taxonomy.

Proposed RESEARCH Focus

  1. Gather: external framework docs — how do they structure principles?
  2. Extract: exact list of removable facts from knowledge/
  3. Challenge: is 3-tier taxonomy correct?

Why Not Just...?

  • Why not just delete everything and rewrite? — Lose provenance. Source links in P# and D# are valuable for tracing decisions
  • Why not keep everything as-is? — KNOWLEDGE.md grows ~15 lines/task. At TFW-30 it'll be 200+ lines and unreadable

HL — TFW-25: Values & Principles Consolidation | 2026-04-04