feat(challenge): governance articles for E0008 challenge refactor

[klappy]

Summary

Eleven governance articles plus an evidence note that set up the oddkit_challenge refactor to mirror the PR #96 encode pattern: governance-driven extraction replacing hardcoded source logic. Governance lands in canon first; the workers/src/orchestrate.ts refactor that extracts against these articles is a follow-up PR. No runtime behavior change here.

The honest framing: the extraction contract is domain-agnostic, but the shipped defaults are not. They are software-engineering-flavored on purpose, labeled as such, with a Domain Adaptation section showing three worked patterns — software engineering (the defaults), thought leadership from books (Tim's pattern), and comparative architectural writing (klappy.dev's pattern). Every other domain is expected to override with its own taxonomy. Custom types are markdown edits, not code changes.

Articles added

Meta governance (odd/challenge-types/)

• how-to-write-challenge-types.md — extraction contract, Domain Adaptation with three worked patterns, six-step procedure for KB stewards to build their own taxonomy

Software-engineering default challenge types

• strong-claim.md — definitive statements; maximum pressure
• proposal.md — future-oriented plans; pressure scales with irreversibility
• assumption.md — implicit premises; make explicit before they compound
• observation.md — reports; lightest pressure; declared fallback type

Architectural-writing overlay challenge types (for this repo's writing work)

• pattern-coinage.md — naming novel patterns; prior-art and precision discipline
• comparative-positioning.md — positioning against a landscape; freshness and fair characterization
• principle-extraction.md — elevating heuristics to principles; sample size and scope

Supporting articles (odd/challenge/, coexist both domains)

• base-prerequisites.md — three universal checks (evidence, source, confidence)
• normative-vocabulary.md — RFC 2119 plus architectural load-bearing terms
• stakes-calibration.md — nine modes spanning software-dev and writing lifecycles; voice-dump mode suppresses all challenge output as invariant

Evidence

• docs/oddkit/evidence/challenge-governance-articles-commit.md — gauntlet record

Gauntlet run

• Preflight — surfaced three constraint docs: ai-voice-cliches, author-identity-language, definition-of-done
• Writing Canon gate per-article — all 11 articles have Summary sections (caught as missing during validate, remediated before commit), blockquotes contain compressed argument, headers pass scan test, metadata references full file paths
• AI voice clichés audit — clean. Zero hits on formulaic transitions, puffing, overclarification, summary clichés, bold-then-explain. Em-dash density 0.13/line vs precedent how-to-write-encoding-types.md at 0.11/line — same neighborhood
• Author identity — no translator claims about Klappy in any article
• Derives-from path audit — every referenced path verified against repo. One broken reference caught and fixed: canon/epistemic-modes.md → canon/definitions/epistemic-modes.md in stakes-calibration.md
• Session capture (OLDC+H) — 7 artifacts encoded (1D, 4L, 1C, 1H); see evidence note for references

Open risks (flagged during challenge, addressable in follow-ups)

• Extraction-contract incompleteness — sections I didn't anticipate needing may emerge during implementation; remediated by amending governance articles
• Detection-pattern overlap noise — operator-mode types and writing-mode types may both fire on ambiguous inputs even with stakes-calibration trimming; tunable via pattern narrowing
• Domain Adaptation discoverability — Tim and other KB stewards may need more worked examples than the three patterns provide; more patterns can be added as domains are stood up
• Voice-dump suppression aggressiveness — suppressing all challenge in voice-dump mode is an invariant here; may need revisiting if a critical check should still fire

Meta observation

oddkit_gate returned NOT_READY during the gauntlet because its hardcoded generic prereqs (problem-statement-defined, constraints-reviewed) could not see session state. This is the same failure mode the challenge refactor addresses — gate has its own version of the governance-disconnect problem. Noted honestly rather than papered over. A future PR can extend the governance-driven pattern to gate too.

Next step

workers/src/orchestrate.ts refactor. Mirror PR #96 encode pattern: discoverChallengeTypes() with per-canonUrl cache, fetch-and-parse for the three supporting articles, runChallengeAction() refactored to use extracted governance, runCleanupStorage() extended to clear four new caches, graceful degradation when articles are missing.

---

Note

Low Risk
Low risk because this PR only adds documentation/canon governance articles and an evidence note, with no runtime or UI changes. Risk is limited to future work that will start extracting behavior from these documents.

Overview
Introduces a new governance set for oddkit_challenge (11 new canon articles) that specifies a markdown extraction contract for challenge types, including multi-match semantics and governance-driven fallback routing.

Adds default challenge type articles (e.g. strong-claim, proposal, assumption, observation), writing-oriented overlay types (e.g. pattern-coinage, comparative-positioning, principle-extraction), and three supporting governance docs for universal prerequisites, normative vocabulary, and mode-based stakes calibration.

Includes a docs/oddkit/evidence/challenge-governance-articles-commit.md gauntlet/evidence record for the governance-only commit.

^{Reviewed by Cursor Bugbot for commit 5d982ae. Bugbot is set up for automated code reviews on this repo. Configure here.}