/wiki-lint-theme-mapping

`/wiki-lint` sub-procedure — Contradiction Theme Mapping (full re-derivation)

Not a slash command — a sub-procedure of [`/wiki-lint`](wiki-lint.md), reached from `contradiction theme --fix`. Invoking it directly runs nothing.

This guide prescribes the procedure for reading all claims in `_contradictions.json` and generating `_contradictions_themes.json` (the theme ↔ claim mapping SoT). Since the output is **structured JSON, not markdown prose**, journalism · consulting form does not apply (unlike overview authoring). Schema conformance · contradiction-axis accuracy · self-validation are the core.

A Claude with no prior knowledge must be able to read this guide alone and reproduce the same quality · structure.

Read before working:

• [`wiki-lint.md` → `## Sub-procedure: Conflict Axis Sync Rule`](wiki-lint.md#sub-procedure-conflict-axis-sync-rule) (SoT hierarchy · cross-assignment principle)
• `wiki/contradictions/_contradictions.json` (main input — all claims; for structure see "Input Data Structure" below)
• For Phase 2 only, `wiki/contradictions/_contradictions_themes.json` (carrying over the Phase 1 output)
• As needed, `wiki/sources/<slug>.md` (source originals — Phase 2 Priority Read targets; for structure see "Source File Structure" below)

---

Input Data Structure

`_contradictions.json` — Main Input

The top level is a JSON array, and each element is a record with the fields below. `python tools/build.py contradictions` auto-generates it by extracting `- contradicts: [[Hub]] — desc` lines from the `## Connections` section of source pages.

{
  "id": "a709b575",
  "source": "sources/agentic-ai-kill-saas-debate.md",
  "claim": "[[osi-open-source-ai-definition]] requires training-data disclosure, whereas this document argues that releasing open weights alone is sufficient to count as open.",
  "status": "open",
  "type": "real",
  "type_score": 0.6,
  "evidence_strength": 0.42
}

| Field | Meaning | Use in theme derivation | |---|---|---| | `id` | First 8 chars of the SHA1 hash of the claim text. The key referenced by `claim_ids` in `_contradictions_themes.json` | **Must** use this value verbatim | | `source` | Path of the Layer 2-1 source file this claim was extracted from (`sources/<slug>.md`) | The actual Read target during Priority Source Read | | `claim` | The desc part of the `## Connections` `- contradicts: [[Hub]] — desc` line (after the prefix is removed). The contradiction is summarized in 1–2 sentences | The primary basis for judging the contradiction axis | | `status` | Currently fixed at `"open"` | Unused | | `type` | Auto-classification result — one of 4 values `real` · `superseded` · `related` · `soft` (explained below) | Priority-Read selection · auxiliary grouping judgment | | `type_score` | Type-category keyword-match score 0.0–1.0 (sum of regex weights capped at 1.0) | Lower = weaker classification signal (re-classification candidate) | | `evidence_strength` | Phase 2 meta-based evidence strength 0.0–1.0 — anchor presence 0.30 + source recency 0.15 + ratio of `## Key Claims` primary sources 0.40 + target hub type 0.15 | Higher = strong contradiction (priority `## Representative Evidence` candidate). Lower = weak signal (consider drop) |

**Orthogonality of the two score dimensions**:

• High `type_score` + low `evidence_strength` → clear keyword match + weak evidence (weasel risk)
• Low `type_score` + high `evidence_strength` → ambiguous classification + strong evidence (re-classification priority)
• Both ≥ 0.6 → strong-contradiction candidate
• Both ≤ 0.3 → soft signal, consider drop

**The 4 values of the `type` field**:

• **`real`**: a clear contradiction · tension — two actors/positions in opposite directions. **The primary target for preliminary theme placement**
• **`superseded`**: a past issue resolved by a timeline update — mostly fragmentary or one-off issues
• **`related`**: a reference/complement relation, not a contradiction — the claim body tends to state "not a contradiction" · "complement" · "similar case." Mostly fragmentary
• **`soft`**: auto-classification failed (the rules could not decide any of {real, superseded, related}). **Emitted with `type_score=0.0` and worth re-reviewing in Phase 2 Priority Read** — a soft claim with high `evidence_strength` is a re-classification priority

Source File Structure — `wiki/sources/<slug>.md`

A Phase 2 Priority Read target. Each source page consists of frontmatter + 4 H2 sections:

| Section | Content | Use in theme determination | |---|---|---| | `## Summary` | A 2–4-sentence summary of the article · document | The fastest entry point for grasping original-text context | | `## Key Claims` | A bullet list of key claims | Confirm which claim a contradiction derives from | | `## Key Quotes` | Quotations from the original text | Confirm speaker · grounds | | `## Connections` | `cites:`/`references:`/`contradicts:`/`defines:` prefix lines. The `contradicts:` line is the extraction source for `_contradictions.json` — grasp other claim context within the same source + infer the related theme · axis |

Priority-Read priority: `## Summary` → `## Key Claims` → `## Connections`. `## Key Quotes` is auxiliary.

---

Purpose & Scope

• **Trigger**: the Claude probabilistic-automation stage when `/wiki-lint contradiction theme --fix` runs
• **Output**: `wiki/contradictions/_contradictions_themes.json` (full overwrite every time)
• **Responsibility scope**: claim → theme derivation and mapping only. Authoring theme.md bodies · Rubric judgment are outside this guide's scope
• **Input scope**: `_contradictions.json` is the main input. Reading `wiki/sources/<slug>.md` is allowed as needed (Phase 2 Priority Read). Phase 2 carries over the Phase 1 output (`_contradictions_themes.json`)

Two-Phase Pipeline

Execution is a sequential pipeline of two Phases. Each Phase runs independently, and the human reviewer reviews the Phase 1 output before instructing Phase 2 to proceed.

| Phase | Pur