trace-first-starter

RES — TFW-23: Templates English Standardization

Date: 2026-04-04 Author: Coordinator Status: 🔬 RES — In progress Parent HL: HL-TFW-23 Mode: Pipeline / Focused


Research Context

5 of 9 TFW templates have mixed Russian/English content. Goal: standardize all to English. This research validates the approach — translation consistency, community practices for multilingual template frameworks, and whether any downstream artifacts/workflows would break.

Briefing

Research Plan

  1. Gather — How do other AI workflow frameworks handle template language? Is English-only the standard? Any multi-language template approaches worth considering?
  2. Extract — Scan existing project artifacts (filled RFs, TSs, HLs) for patterns: do agents already mix section heading language when filling? Do workflows reference specific RU heading text?
  3. Challenge — Is pure English the right call? Could we lose something by removing RU? Should we consider i18n placeholders instead of hardcoded English?

Scope Intent

Guiding Questions

  1. Do you have any preference for specific English terms for ambiguous translations (e.g., “Целевое состояние” → “Target State” vs “Desired State” vs “To-Be”)?
  2. Should the Статус line use the same phrasing as PROJECT_CONFIG status descriptions, or keep the current human-friendly labels?
  3. Is there any plan to support non-English users who might fork this repo and want templates in their language?

Hypotheses (from HL §10)

# Hypothesis HL Status RES Status Evidence
H1 RES.md is already fully English confirmed ✅ confirmed Verified by reading file
H2 KNOWLEDGE.md, RELEASE.md, TOPIC_FILE.md are already English confirmed ✅ confirmed Verified by reading files
H3 No workflows reference template headings by RU name open 🟡 testing  

→ User direction: ___

Decisions

| # | Decision | Rationale | |—|———-|———–| | D1 | English-only templates, content filled in user’s language | Industry standard (G1). Templates = code → code is English | | D2 | Use consistency table from E1 for all translations | D28: “right terminology creates right associations”. 32 terms mapped | | D3 | content_language config deferred to future task | Challenge C1: mixing features violates scope. TFW-20 or new task | | D4 | Translate RU instructional annotations too, not just headings | Challenge C3: instructions are part of template structure | | D5 | Update KNOWLEDGE.md §3 legacy entry to match new EN headings | Challenge C4: avoid confusion when agents read legacy table | | D6 | H3 confirmed: workflows reference §numbers, not heading text | Grep verification (G4). Renaming is safe | | D7 | §3.1 instructional block: Variant A (principle + domain examples). §10 filter: literal translation. Both rewritten to be domain-agnostic — TFW is not code-only | User feedback: HL instructions too tech-focused, TFW serves education, business, research too |

Open Questions

| # | Question | Status | Answer | |—|———-|——–|——–| | 1 | Best EN terms for ambiguous RU headings? | ✅ resolved | Consistency table in E1 — D28 applied to each term | | 2 | Status line phrasing source? | ✅ resolved | Short labels: “Awaiting review”, “Awaiting approval”, “Complete” | | 3 | i18n / fork considerations? | ✅ resolved | Document principle in HL. content_language config = future task |


Stage: Gather — “What do we NOT know?”

G1: External i18n practices for AI prompt templates

External research confirms: English structure + localized content is the industry standard pattern. Key finding: “Language-Agnostic Core — keep the core system logic consistent across all languages. Only localize the tone, style, and cultural adaptation parts.” This is exactly what TFW needs — templates = core logic (English), filled content = user language.

G2: Template i18n architecture options

Option Description Complexity Fit
A. English-only templates Templates in EN, agents fill in user’s language Low ✅ Best — simple, no infra
B. Locale directories templates/en/HL.md, templates/ru/HL.md High ❌ File explosion, sync nightmare
C. YAML key-value i18n {t.vision} placeholders, resolved from locale file Medium ❌ Over-engineered for markdown
D. Config-driven content language tfw.content_language: en in config, agent uses it for filling Low ✅ Good — configurable without file duplication

Conclusion: Option A + D hybrid is best. Templates stay English (Option A). Config has tfw.content_language (Option D) that tells the agent which language to use when filling the template. Default = en. During tfw-init, ask user preferred language.

G3: Naming — D28 principle application

Per D28 (“Naming > Explanation”) and P5 (“right terminology creates right associations”):

G4: H3 verification (workflow heading references)

Grep confirmed: workflows reference sections by § number (§3.1, §10), never by heading text. Zero RU heading references in any workflow file. Renaming headings is safe.

Checkpoint: Gather

| Found | Remaining | |——-|———–| | i18n best practice = EN structure, localized content | — | | 4 architecture options evaluated, A+D hybrid selected | — | | D28 naming principle applicable to heading translation | translation table (→ Extract) | | H3 confirmed: no workflow refs to RU headings | — |

Sufficiency Verdict:

Agent assessment: Gather is sufficient. Architecture decision made (A+D hybrid). H3 confirmed. Depth check: 3 external web searches used (i18n practices, naming conventions, locale config patterns). Recommendation: close stage, move to Extract for naming table Stage complete: YES → User decision: ___

Stage: Extract — “What do we NOT see?”

E1: Translation Consistency Table (applying D28: Naming > Explanation)

Method: For each RU heading, identify the behavior it should trigger in the agent. Then find the shortest English term that the AI already associates with that behavior (D28: “right terminology creates right associations”).

Common fields (all templates)

RU Original EN Translation D28 Rationale
Дата Date Universal
Автор Author Universal
Статус Status Universal
Ожидает ревью Awaiting review Short, clear intent
Ожидает апрува Awaiting approval Standard process term
Ожидает ответов Awaiting answers Direct, no ambiguity
Выполнено Complete Shortest possible

HL template

§ RU Original Candidate A Candidate B Winner D28 Rationale
1 Видение Vision Goal Vision Standard strategic planning term; agents associate it with “why + business value”
2 Текущее состояние (As-Is) Current State (As-Is) As-Is Current State (As-Is) Already has EN annotation; keep both for scan-ability
3 Целевое состояние (To-Be) Target State (To-Be) To-Be Target State (To-Be) Mirror of §2
3.1 Визуализация результата Result Visualization Outcome Preview Result Visualization Agents associate “visualization” with diagrams/ascii — exactly what we want
4 Фазы Phases Roadmap Phases Direct translation; TFW uses “Phase” heavily
5 Definition of Done (DoD) Keep Already EN
6 Definition of Failure (DoF) Keep Already EN
7 Принципы Principles Design Philosophy Principles Short, standard. Section body already says “Design philosophy”
8 Зависимости Dependencies Prerequisites Dependencies Standard engineering term
8 tbl Зависимость \| Статус Dependency \| Status Dependency | Status Direct
9 Риски Risks Risks Universal
9 tbl Риск \| Вероятность \| Влияние Risk \| Probability \| Impact Risk \| Likelihood \| Severity Risk | Probability | Impact Standard risk matrix terms
9 tbl Низкая/Средняя/Высокая Low/Medium/High Low/Medium/High Universal
10 Обоснование RESEARCH RESEARCH Case RESEARCH Justification RESEARCH Case Shorter. “Case” = argument. Agents associate “case” with “make this argument”
10.1 Слепые зоны Blind Spots Unknown Unknowns Blind Spots Already used in workflows in EN. AI associations: strong
10.2 Гипотезы Hypotheses Hypotheses Standard scientific term
10.2 tbl Гипотеза \| Статус Hypothesis \| Status Hypothesis | Status Direct
10.3 Фильтр Filter Filter Direct. But the content is more important — translate the instruction
10.4 Риски незнания Risks of Not Researching Ignorance Risks Risks of Not Researching Explicit, no ambiguity
10.5 Предлагаемый фокус RESEARCH Proposed RESEARCH Focus Proposed RESEARCH Focus Direct, clear
footer При провале On failure On failure Short

TS template

§ RU Original Winner D28 Rationale
1 Цель Objective Stronger action signal than “Goal”; agents associate “objective” with deliverable
3 Затрагиваемые файлы Affected Files Standard term in PR/diff tooling
3 tbl Файл \| Действие \| Описание File | Action | Description Direct
3 note Бюджет Budget Direct; already used in EN context
6 Риски фазы Phase Risks Mirror HL §9

RF template

§ RU Original Winner D28 Rationale
1 Что сделано What Was Done Direct, action-past
1.1 Новые файлы New Files Direct
1.2 Изменённые файлы Modified Files Standard term (git)
1 tbl Файл \| Описание, Файл \| Изменения File | Description, File | Changes Direct
2 Ключевые решения Key Decisions Direct
4 Верификация Verification Direct

ONB template

§ RU Original Winner D28 Rationale
1 Understanding (как понял задачу) Understanding Drop RU annotation
2 Entry Points (откуда начинать) Entry Points Drop RU annotation

REVIEW template

Only Дата, Автор — covered in common fields.

E2: content_language config pattern

Proposed addition to PROJECT_CONFIG.yaml:

tfw:
  content_language: en    # Language for filled artifacts (en, ru, etc.)

Agent instruction (goes in conventions.md or AGENTS.md):

“When filling templates, use the language specified in tfw.content_language. Template structure (headings, labels, field names) stays English always.”

During tfw-init, agent asks: “What language should I use when filling artifacts? Default: English.”

E3: Impact on existing KNOWLEDGE.md entry (D25 in KNOWLEDGE.md §3)

KNOWLEDGE.md line 133 has:

§3.1 Визуализация результата (ascii mandatory), §10 Обоснование RESEARCH (hypotheses + blind spots)

This is a historical record (Source column = TFW-22). Will need updating to match new EN headings.

Checkpoint: Extract

| Found | Remaining | |——-|———–| | Full translation table for 5 templates (32 term pairs) | — | | content_language config pattern designed | — | | KNOWLEDGE.md §3 needs historical record update | HL scope update |

Sufficiency Verdict:

Agent assessment: Extract is sufficient. Complete consistency table built. No gaps remaining. Depth check: Project knowledge (D28, P5) + convention analysis + grep verification of downstream refs. Recommendation: close stage, move to Challenge Stage complete: YES → User decision: ___

Stage: Challenge — “What do we NOT expect?”

C1: Is content_language overengineering?

Argument FOR: User explicitly asked for it. Minimal implementation — one YAML key + one sentence in conventions.md. Enables non-English users without template duplication.

Argument AGAINST: If we add it now, we need to:

  1. Add key to PROJECT_CONFIG.yaml
  2. Document it in conventions.md
  3. Add to tfw-init workflow (ask user)
  4. Add to Config Sync Registry (via /tfw-config)
  5. Update AGENTS.md or conventions.md with the instruction

This is not template standardization — it’s a new feature. Mixing it into TFW-23 violates scope.

Verdict: Split content_language into a separate concern. TFW-23 = English templates only. The content_language feature can be part of TFW-20 (user preferences) or a new task. Today we document the principle: “templates = English, content = user’s language” in the HL, but don’t implement the config mechanism.

C2: Do the chosen heading names actually trigger better behavior?

Testing a few critical names against alternatives:

Term Why it works (D28) Risk
Vision (§1) AI models are trained on business docs — “Vision” triggers strategic framing, not just description Low — universally understood
Objective (TS §1) vs “Goal” “Objective” is more specific: “measurable deliverable”. “Goal” is vaguer. In military/project contexts, “objective” = concrete target Low
RESEARCH Case (§10) vs “RESEARCH Justification” “Case” = shorter, implies “make the argument”. “Justification” = reactive, implies “defend a decision already made”. “Case” is proactive Low
Result Visualization (§3.1) vs “Outcome Preview” “Visualization” literally tells the agent to produce a visual (diagram, ascii). “Preview” is passive — could mean just text description Winner: Result Visualization confirmed

No alternative names are clearly better. The consistency table from Extract holds.

C3: What about the RU instructions inside HL template (not headings)?

Lines 23-28 in HL.md are RU instructional text:

> Покажи результат как будто он уже достигнут. Используй:
> - **ASCII-схемы** (обязательно): ...
> - **Mermaid** (для complex flows): ...
> - **Before→After таблицы** для сравнения state
> Цель: исполнитель и юзер должны увидеть «финальную картинку» до начала работы.

And line 88:

> **Фильтр:** Каждая гипотеза: «Если окажется ложной — изменится ли наш подход?» Если нет — удалить.

These are instructional annotations, not headings. They also need translation. I missed these in the initial HL scope. Must be included.

C4: Should KNOWLEDGE.md historical entries be updated?

Line 133 in KNOWLEDGE.md has §3.1 Визуализация результата. This is a Legacy & Deprecation record. The convention (P2: “Index, don’t duplicate”) says link to sources. The §3.1 heading name is descriptive context in the legacy table. It should be updated to reflect the new name to avoid confusion when agents read it.

Checkpoint: Challenge

| Found | Remaining | |——-|———–| | content_language = separate concern, not TFW-23 scope | Document principle in HL only | | Naming table confirmed — no alternatives are clearly better | — | | RU instructional text inside templates needs translation too | Add to HL scope | | KNOWLEDGE.md historical entry needs update | Add to HL scope |

Sufficiency Verdict:

Agent assessment: Challenge revealed 2 scope additions (instructional text, KNOWLEDGE.md) and 1 scope reduction (content_language → future task). Net change is minor. Depth check: Internal analysis of assumptions, external D28 cross-reference. Recommendation: close stage, proceed to Final Checkpoint Stage complete: YES → User decision: ___


Final Checkpoint

Stage Status Key Findings
Gather ✅ Done EN structure + localized content = industry standard. A+D hybrid architecture. H3 confirmed (no RU heading refs in workflows)
Extract ✅ Done 32-term consistency table built (D28 applied). content_language config pattern designed. KNOWLEDGE.md legacy entry found
Challenge ✅ Done content_language = separate task (scope reduction). RU instructional text also needs translation (scope addition). KNOWLEDGE.md entry update needed

Sufficiency Check

Question: Sufficient for HL finalization? Can we confidently define phases, approach, and dependencies? Self-check:

Verdict: Sufficient for HL finalization

Closure

HL Update Recommendations

| # | What to update | Source | |—|—————|——–| | 1 | Add RU instructional text (HL.md lines 23-28, 88) to scope — not just headings | C3 | | 2 | Add KNOWLEDGE.md §3 legacy entry update to scope | C4 | | 3 | Remove content_language config from TFW-23 scope → note as future (TFW-20 or new task) | C1 | | 4 | Add consistency table reference to HL (or keep in RES as source of truth) | E1 | | 5 | Document principle: “templates = English, content = user’s language” in HL §7 | C1 |

Next Step

Update HL → confirm → TS → User decision: ___

Fact Candidates

# Category Candidate Source Confidence
1 convention Templates = English always; filled content = user’s language. Template structure is code, code is English User direction + external i18n best practice High
2 convention content_language as a config key is the right pattern for i18n in TFW — avoids template duplication, avoids locale directories, minimal infra. Implementation deferred RES G2 analysis Medium
3 process D28 (Naming > Explanation) applies not just to workflow terms but also to template section headings — heading names should trigger correct agent behavior, not just describe content RES E1 + C2 analysis High

Conclusion

Research confirmed that English-only templates with user-language content is the industry standard for AI prompt frameworks. A 32-term consistency table was built applying D28 (Naming > Explanation), choosing terms that trigger correct agent behavior. Challenge stage refined scope: content_language config is a separate feature (deferred), while RU instructional annotations and KNOWLEDGE.md legacy entries were added to scope. All hypotheses confirmed. No unknowns remain.


*RES — TFW-23: Templates English Standardization 2026-04-04*