# Tml Infer > Surfaces inferences made during a TML interview — the gap between what was explicitly stated and what the system concluded. Produces a structured Inferences document that separates confirmed facts from model-generated conclusions, flags low-confidence inferences for human review, and makes the reasoning chain visible. Part of the TML capture pipeline — runs after tml-interview, before tml-map. - Skill: `cobrachickenai/tml-infer` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add cobrachickenai/tml-infer` - Raw SKILL.md: https://api.skillmd.com/api/skills/cobrachickenai/tml-infer/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: DevOps & Infra - License: MIT - Author: CobraChickenAI (https://skillmd.com/u/cobrachickenai) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/cobrachickenai/tml-infer --- # TML Infer Takes the confirmed primitives from a `tml-interview` and makes the model's reasoning visible. Every interview produces two kinds of output: things the human explicitly said, and things the system concluded from what the human said. The map conflates these by design — it reads as authoritative. This skill separates them. It answers: *what did I decide on your behalf, and how confident am I that I got it right?* This is the debugging surface. If the inference layer is wrong, the map is wrong. Catching it here is cheaper than catching it six months from now. ## When to Use - After `tml-interview`, before presenting the map for confirmation - As part of a `tml-capture` orchestration (runs automatically between interview and map) - When a human wants to audit what the system concluded vs. what they actually said - As a standalone review of any existing TML map — "show me what you inferred" --- ## What Gets Inferred Inferences happen in three ways during a TML interview: ### 1. Boundary inferences The human described something. The system had to decide where the edge of the scope is. That decision is an inference. *Example: Human said "my team." System inferred the scope as "the Customer Success team, excluding shared services." That boundary is a conclusion, not a quote.* ### 2. Classification inferences The human said something. The system had to decide which primitive it belongs to. That classification is an inference. *Example: Human said "we send weekly status emails." System classified that as a Binding (write path to external stakeholders). It could also have been a Capability. The choice was a conclusion.* ### 3. Gap-filling inferences Something wasn't addressed in the interview. The system either flagged it as a gap or filled it from context. Both are inferences. *Example: The human never mentioned who can change the approval threshold. System inferred it defaults to the manager archetype based on general patterns. That's a conclusion — and it might be wrong.* --- ## The Analysis For each primitive in the confirmed primitive set, examine: 1. **What was explicitly stated?** Direct quotes or clear paraphrases from the interview. 2. **What was concluded?** The system's interpretation, classification, or extension of what was stated. 3. **How confident is that conclusion?** High (obvious from context), Medium (reasonable but could be wrong), Low (speculation or pattern-matching). 4. **What would falsify it?** The single most likely way this inference is wrong. --- ## Output Format ```markdown # Inferences: [Scope Name] **Interview date:** [date] **Participant:** [name] **Generated:** [date] --- ## How to Read This This document shows what the system concluded vs. what you actually said. - **Stated** = you said this, or confirmed it when played back - **Inferred** = the system concluded this from context - **Gap** = this wasn't addressed; the system flagged it rather than filling it Review the inferences. If any are wrong, correct them before the map is finalized. A wrong inference in the map is harder to find later. --- ## Scope **[Scope name]** | Element | Status | Source / Reasoning | |---------|--------|--------------------| | Name | Stated / Inferred | [quote or reasoning] | | Boundary | Stated / Inferred | [what was said, what was concluded about the edge] | | Description | Stated / Inferred | [quote or reasoning] | **Confidence:** High / Medium / Low **Most likely wrong if:** [what would falsify this] --- ## Domains ### [Domain 1 name] | Element | Status | Source / Reasoning | |---------|--------|--------------------| | Name | Stated / Inferred | [quote or how the name was derived] | | Ownership | Stated / Inferred / Gap | [who owns this and how we know] | | Accountability boundary | Stated / Inferred | [what's in vs. out of this domain] | **Confidence:** High / Medium / Low **Most likely wrong if:** [what would falsify this] [repeat for each domain] --- ## Capabilities ### [Capability 1 name] | Element | Status | Source / Reasoning | |---------|--------|--------------------| | Name (verb-noun) | Stated / Inferred | [quote or how the name was derived from what was said] | | Outcome | Stated / Inferred | [what was said about what this produces] | | Domain assignment | Stated / Inferred | [why it was placed in this domain] | **Confidence:** High / Medium / Low **Most likely wrong if:** [what would falsify this] [repeat for each capability] --- ## Archetypes ### [Archetype name] | Element | Status | Source / Reasoning | |---------|--------|--------------------| | Role name | Stated / Inferred | [quote or derivation] | | Can-do list | Stated / Inferred | [what was explicit vs. concluded from context] | | Cannot-do list | Stated / Inferred / Gap | [what was explicit vs. defaulted from policies] | **Confidence:** High / Medium / Low **Most likely wrong if:** [what would falsify this] [repeat for each archetype] --- ## Policies ### [Policy name] | Element | Status | Source / Reasoning | |---------|--------|--------------------| | Rule statement | Stated / Inferred | [quote or paraphrase] | | What it constrains | Stated / Inferred | [explicit vs. concluded from context] | | Enforcement mechanism | Stated / Inferred / Gap | [how this is actually upheld] | **Confidence:** High / Medium / Low **Most likely wrong if:** [what would falsify this] [repeat for each policy] --- ## Connectors & Bindings ### [Connector/Binding name] | Element | Status | Source / Reasoning | |---------|--------|--------------------| | Direction (read/write) | Stated / Inferred | [how the read/write distinction was determined] | | What flows | Stated / Inferred | [what data or effect, and how we know] | | External endpoint | Stated / Inferred / Gap | [named system or vague reference] | **Confidence:** High / Medium / Low **Most likely wrong if:** [what would falsify this] [repeat for each connector and binding] --- ## Views [If no views were identified, state: "No views were identified in this interview. This may be a gap — consider whether different roles see different filtered versions of the same information."] --- ## Provenance | Element | Status | Source / Reasoning | |---------|--------|--------------------| | Existing tracking | Stated / Inferred / Gap | [what was mentioned vs. assumed] | | Gaps | Stated / Inferred | [what's not tracked and how we know] | --- ## Summary: Inferences to Review The following inferences are Medium or Low confidence and should be explicitly confirmed before the map is finalized: | Primitive | Inference | Confidence | What to ask | |-----------|-----------|------------|-------------| | [type] | [what was concluded] | Medium/Low | [exact question to confirm or refute] | --- ## Gaps Not Filled The following were not addressed in the interview. They are flagged rather than inferred: | Primitive | What's missing | Suggested question | |-----------|---------------|-------------------| | [type] | [what's absent] | [how to get the answer] | --- *This document is not the map. It is the reasoning behind the map. Correct the inferences here, and the map corrects with them.* ``` --- ## Rules 1. **Never present an inference as a fact.** If it was concluded rather than stated, say so. 2. **Low confidence is not a failure.** It is the system being honest. Flag it and move on. 3. **The "most likely wrong if" field is mandatory.** If you can't articulate how an inference could be wrong, it means you haven't thought about it carefully enough. 4. **Gaps are better than guesses.** An honest gap is more valuable than a wrong inference that makes it into the map. 5. **Use direct quotes where available.** The provenance chain from statement → inference → primitive is the point. --- ## Handoff The Inferences document is reviewed by the human before `tml-map` is run. Corrections to inferences propagate to the map automatically. In a `tml-capture` orchestration, the Inferences document is presented alongside the map as the fourth artifact, not before it — the map is the deliverable, the inferences are the audit surface. --- *v1.0 — The reasoning layer. Makes model conclusions visible so humans can correct them before they compound.*