Skip to content

title: "Extract — Iteration 2: "What do we NOT see?"" source: "tasks/TFW-32__methodology_and_positioning/research2/extract.md"


Extract — Iteration 2: "What do we NOT see?"

Parent: HL-TFW-32 Predecessor: RES1, Gather2 Goal: Design decisions for naming, visualization, business processes, multi-iteration.

Findings

E1: Naming — honest alternatives analysis (no status-quo bias)

The question: Is "Strategic Insights" genuinely the best name, or am I validating user preference?

Test: Apply D28 strictly. The name must tell the agent: 1. WHAT to look for (target) 2. WHERE it comes from (source) 3. WHY it matters (value)

6 candidates tested:

# Name What agent looks for Source signal Risk D28 score
A Strategic Insights (current) Business context, motives, priorities, domain knowledge Human-only. Synthesis-mode "Strategic" might over-narrow: agent skips non-strategic human knowledge (e.g., "we tried X and it failed" — is that strategic?) 7/10
B Knowledge Candidates (RES1 D2) Anything that could be knowledge Human-sourced Too broad — triggers retrieval mode. Agent writes everything 3/10
C Domain Signals Domain-specific facts, market rules, business logic Human + external research Misses stakeholder emotions, priorities, corrections 5/10
D Stakeholder Intelligence Priorities, constraints, decisions, politics, emotions Human-only Too corporate. Misses pure domain knowledge (e.g., "revenue is seasonal") 6/10
E Strategic Signals Strategic decisions, priorities, corrections, domain patterns Human-only, synthesis "Signal" implies filtering — agent looks for high-value items only. Tighter than "Insight" 8/10
F Context Signals Background, constraints, domain, motives, emotions — anything that shapes decisions Human + environment "Context" is broad but "Signal" narrows it: only what changes decisions. Inclusive but filtered 7/10

Honest assessment of "Strategic Insights" (option A): - PRO: "Strategic" frames synthesis mode (confirmed by external research). Agent doesn't list facts, it identifies implications - PRO: Current name, already works, no migration cost - CON: "Strategic" might make agent skip NON-strategic human input: emotional signals, personal experience ("I tried this, it broke"), cultural context. These are valuable but an agent might not classify them as "strategic" - CON: "Insight" implies the agent must generate the insight, not just record it. Sometimes the human IS the insight

Alternative "Strategic Signals" (option E): - PRO: "Signal" implies detection + filtering (Agent: "Is this a signal I should capture?"). Better alignment with Human-Only Test - PRO: "Strategic" keeps the D28 behavioral frame - CON: "Signal" is less common in methodology language — might confuse human users who aren't signal-processing-minded

Counter-test: When does "Strategic Insights" fail? User says: "Мне понравилась возможно запускать несколько ресерчей отдельно, друг за другом." Is this a "Strategic Insight"? It's a PROCESS preference, not strategy. Agent might hesitate to capture it under "Strategic." But it IS exactly the kind of human-only knowledge that matters.

Under "Strategic Signals" — agent sees "signal" and asks: "Does this change how we work? Yes → capture." Under "Context Signals" — same behavior, broader net.

Decision recommendation: Keep "Strategic" framing (D28 confirmed — synthesis mode). But acknowledge the gap: rename from "Strategic Session Insights" to one of: - Option A (conservative): "Strategic Insights" (drop "Session" — already proposed in RES1 E2) - Option E (sharper): "Strategic Signals" — adds filtering behavior

Both are improvements over RES1's "Knowledge Candidates." The real choice is between A and E. This is a coordinator decision, not a researcher one.

E2: Fixing "Fact Candidates" (§6) — the actually vague name

Current problem: "Fact Candidates" in RF §6 and RES §FC is a catch-all. Agent doesn't know what KIND of thing to capture.

But wait — is this actually a problem? Let me check what RF §6 actually captures in practice.

Looking at RES1 FC1-FC9: - FC1: "Documentation = facts about the project..." — user-sourced definition → should be Strategic Insight - FC2: "Product-oriented people learn TFW faster..." — user-sourced stakeholder insight → should be Strategic Insight - FC3: "TFW = team tool..." — user-sourced philosophy → should be Strategic Insight - FC6: "SECI model maps to TFW..." — research finding → documentation/methodology knowledge - FC7: "Competitive positioning..." — research finding → documentation - FC9: "Orchestration of docs+knowledge..." — research design → documentation

Pattern: RES1's Fact Candidates are a MIX of: 1. Human-sourced strategic knowledge (FC1-FC5, FC8) — should be in Strategic Insights 2. Research-derived findings (FC6-FC7, FC9) — operational/methodology observations

The problem: researcher put human-only insights into "Fact Candidates" because the RES template has no "Strategic Insights" section (only HL has §11).

Fix options:

Option Change Impact
X1: Add "Strategic Insights" section to RES and RF templates Each artifact captures BOTH types explicitly More sections, but each is precise
X2: Keep "Fact Candidates" but add subsection headers §6a Operational, §6b Strategic Minimal template change, agent still writes to one section
X3: Rename "Fact Candidates" → "Project Observations" and add "Strategic Insights" Two sections with precise names Clean separation, but "Observations" already exists in RF §5

Recommendation: X1 — add "Strategic Insights" section to RES and RF templates. - HL §11 already has it → extend to RES and RF - RF §5 Observations = agent technical findings (stays) - RF §6 Fact Candidates = becomes operationally scoped: agent-observable patterns about the project itself - RF §7 (new) or integrated: Strategic Insights = human-sourced domain knowledge captured during execution - RES: add Strategic Insights section after Fact Candidates

This resolves TD-76 DIFFERENTLY from RES1: instead of renaming everything, EXTEND the precise naming to all artifacts.

E3: Visualization section — naming options with framing analysis

The problem: One name across all artifacts, but different content per context. The name must FRAME what the agent draws.

Testing 5 name candidates against arc42 and D28:

# Section Name What agent draws in HL What agent draws in RF What agent draws in RES Framing quality
V1 Diagrams Architecture vision Implementation detail Research discoveries Neutral — doesn't specify PURPOSE. Agent might skip if "no diagram needed"
V2 Visual Map Value flow map System/data flow maps Concept maps, findings "Map" implies spatial relationships. Misses sequences, timelines
V3 Process & Flow Value delivery flow API/DB/sequence flow Research flow/findings "Process" is specific to flows — excludes architecture diagrams, mockups
V4 Blueprints Vision blueprint Technical blueprint Research blueprint "Blueprint" = planning orientation. Doesn't fit RF (result, not plan)
V5 Visual Evidence "Here's what it should look like" "Here's what was built" "Here's what we found" "Evidence" implies PROOF. Agent creates visuals that DEMONSTRATE something. Works across contexts
V6 Schemas (RU: Схемы) Value delivery schema Technical schemas Analytical schemas "Schema" = structured representation. Works in both RU and EN. But might be confused with DB schemas

Additional consideration from user input:

"Везде должно быть одна общая глава с одним четким названием, которая на стадии знаний будет куда-то собираться"

The name must also work as a COLLECTION target during 📚 KNW. During knowledge consolidation, all "Visual Evidence" sections from all artifacts → collected into knowledge as reference diagrams.

I notice: HL §3.1 "Result Visualization" already exists — it's close to V5 but narrower. Could be evolved.

My concern with all options: The section name alone won't ensure agent draws the RIGHT thing. The INSTRUCTIONS inside the section (like HL §3.1 currently has) are what guide behavior. The name is a trigger, the instructions are the behavior.

Design: One name + per-artifact instruction block.

## {Section Name}
> {Per-artifact instruction — changes by template}

With this pattern, V1 "Diagrams" or V5 "Visual Evidence" both work — the instruction block does the real framing. V1 is simpler, V5 is more behaviorally loaded.

This needs more testing. Recommend parking for iteration 3 with user input on preferred direction.

E4: Multi-iteration research — control file and enforcement design

From lived experience (this session = iteration 2):

What I needed at session start: 1. Which hypotheses were tested (and results) 2. Which hypotheses were NOT tested (and why) 3. What NEW questions emerged during RES1 4. User's satisfaction with RES1 findings

What I had to manually reconstruct: - Re-read entire HL (291 lines) - Re-read entire RES1 (101 lines) - Re-read all 4 stage files (~500 lines total) - Parse checkpoint results from each stage

What iterations.yaml SHOULD contain (minimal viable state):

task: [TFW-32](../HL-TFW-32__methodology_and_positioning.md)
min_iterations: 2  # set by coordinator
current: 2

log:
  - iteration: 1
    folder: research/
    res_file: RES__TFW-32__methodology_and_positioning.md
    hypotheses_tested: [H1, H2, H3, H5, H7, H8]
    hypotheses_deferred: [H6]
    decisions: [D1, [D2](../../../knowledge-index.md#architecture-decisions), [D3](../../../knowledge-index.md#architecture-decisions), [D4](../../../knowledge-index.md#architecture-decisions), [D5](../../../knowledge-index.md#architecture-decisions), [D6](../../../knowledge-index.md#architecture-decisions), [D7](../../../knowledge-index.md#architecture-decisions), [D8](../../../knowledge-index.md#architecture-decisions), D9]
    gaps:
      - "H6 (visualization)  not investigated"
      - "[D2](../../../knowledge-index.md#architecture-decisions) naming  'Knowledge Candidates' may be too vague"
      - "[D4](../../../knowledge-index.md#architecture-decisions) (iterations.yaml)  designed but untested"
      - "Business process representation  not covered"
    new_hypotheses: []
    researcher_note: "H6 deferred without investigation. H2 deferred pragmatically. [D4](../../../knowledge-index.md#architecture-decisions) is least validated."

  - iteration: 2
    folder: research2/
    res_file: null  # pending
    hypotheses_tested: [H5b, H6, H6b, H7b]
    hypotheses_deferred: []
    decisions: []
    gaps: []
    new_hypotheses: [H5b, H6b, H7b]
    researcher_note: null

Researcher exit protocol (end of each iteration):

## Iteration Status
- **Iteration:** {N} of {min_iterations}
- **Tested:** {list}
- **Deferred/Remaining:** {list}  
- **Gaps discovered:** {list}
- **New hypotheses:** {list or "none"}

### Recommendation
{CONTINUE / SUFFICIENT}
- If CONTINUE: "Gaps remain. Launch `/tfw-research` iteration {N+1} in a new agent. Focus: {gaps + new hypotheses}"
- If SUFFICIENT: "All hypotheses tested. Recommend proceeding to `/tfw-plan` for TS."

> ⚠️ Coordinator decides whether to continue or proceed. Researcher recommends but does NOT decide.

Coordinator gate (plan.md extension): - After receiving RES from researcher, coordinator checks iterations.yaml - If current < min_iterations → MUST launch next iteration (no bypass) - If current >= min_iterations AND researcher says SUFFICIENT → may proceed to TS - If current >= min_iterations AND researcher says CONTINUE → coordinator decides - Coordinator can INCREASE min_iterations at any point

Key enforcement: min_iterations is a HARD FLOOR. User's point: "без ямла все закончится на фаст ран" — the YAML IS the structural enforcement. Without it, the researcher writes "done" and the coordinator has no gate to check.

E5: Two-level business process representation

Design based on user input + BPMN + VSM patterns:

Level 1 — HL (Vision): Value Delivery Flow

User/Trigger → [Step 1] → [Step 2] → ... → Value/Result
- ASCII or mermaid - Shows: who triggers, what flows, where value is delivered - Purpose: "Does the executor understand WHERE the value ends?" - Abstraction: business-level (no technical details)

Level 2 — RF (Result): Technical Process Detail

API Call → DB Query → Transform → Response → UI Update
- Sequence diagrams, data flow, API contracts - Shows: exact implementation of the value flow - Purpose: "Can the next agent reproduce this?" - Abstraction: implementation-level

Both levels live in the visualization section (whatever it's named). The section instructions differ per template: - HL template: "Show the value delivery flow — from trigger to outcome" - RF template: "Show the technical process — databases, APIs, sequences" - RES template: "Show research findings visually — concept maps, relationship diagrams, comparison matrices"

Checkpoint

# Finding Status Open question
E1 "Strategic Insights" is good but not perfect. "Strategic Signals" might be sharper. Both beat "Knowledge Candidates" Options A vs E for coordinator Need user input on "Insight" vs "Signal"
E2 Fix §6 "Fact Candidates" by adding Strategic Insights section to RES/RF templates (X1) Design ready Does this create too many sections?
E3 Visualization section name: V1 "Diagrams" vs V5 "Visual Evidence" — instructions matter more than name Needs more testing Park for iteration 3
E4 iterations.yaml format + researcher exit protocol + coordinator hard gate Design ready Insertion point in plan.md
E5 Two-level business process: HL=value flow, RF=technical detail Design ready Where does RES fit?

Sufficiency: - [x] External source used? (arc42, C4, BPMN, VSM, naming research — all from Gather) - [x] Briefing gap closed? (all 4 hypotheses have design proposals)

Deep mode criteria: - [x] Hypothesis tested? H5b: BOTH "Strategic Insights" and "Knowledge Candidates" tested honestly. Neither is perfect — "Strategic Signals" is a third option - [x] Counter-evidence sought? Against "Strategic Insights": tested edge case where process preference ≠ strategy. Against keeping names: tested if Fact Candidates is actually problematic (it is — FC1-FC5 from RES1 prove misclassification)

Metacognitive check: Found something uncomfortable — "Strategic Insights" has a real gap (non-strategic human knowledge gets missed). I could have just confirmed user preference but the edge case test revealed a genuine weakness. Also found that the REAL fix for TD-76 is extending Strategic Insights to RES/RF, not renaming it.

Stage complete: YES → User decision: ___