# ODD Manifesto v1.2 (Extended) > A governance framework for human-AI collaboration that optimizes learning, not execution. ODD operationalizes epistemic discipline: defining outcomes, enforcing constraints, and verifying reality. AI accelerates execution; governance preserves trust. Confidence scores are grounded in observed system behavior across 246+ canonical documents, working MCP tooling, and documented incidents — not theory. ## Description Outcomes-Driven Development (ODD) operationalizes governance for human-AI collaboration. The core thesis: development is about defining outcomes, enforcing constraints, and verifying reality—not writing code. AI accelerates execution; governance preserves trust. The pillars include Prompt Over Code, KISS, DRY with Isolation, Consistency, Maintainability, Antifragile design, and Scalability. ODD treats restartability as a feature, applies progressive governance based on maturity (PoC → Pilot → Production), requires evidence over assertion, treats AI as accelerator not authority, demands falsifiable outcomes, prefers reversibility, and requires stop conditions. Memory is the bottleneck, not computation. ODD is portable epistemic infrastructure — it works in monorepos, multi-repos, or any structure where canon is addressable. ## Content > ODD v1.2 — Extended (Internal / Agent-Governance) → for canon, MCP, agents (this file) --- ## Purpose This document operationalizes Outcomes-Driven Development as a governance framework for human–AI collaboration. It is designed to: - guide autonomous agents - enforce verification and evidence - scale judgment without repeating it - adapt rigor as projects mature This version is not optimized for persuasion. It is optimized for execution and enforcement. --- ## Core Thesis The primary job of development is not writing code. It is defining outcomes, enforcing constraints, and verifying reality. AI accelerates execution. Governance preserves trust. --- ## Values-First Foundation ODD's epistemic discipline is grounded in moral commitments, not just procedural constraints. Four foundational axioms — Reality Is Sovereign, A Claim Is a Debt, Integrity Is Non-Negotiable Efficiency, You Cannot Verify What You Did Not Observe — define an unconditional commitment to truth from which all constraints, validators, and definitions of done are derived. These values are explicit, intentional, and forkable. See `canon/values/axioms.md` for the full axioms and `canon/values/orientation.md` for the creed agents carry into every task. --- ## Pillars (Operational Interpretation) ### Prompt Over Code - Intent is expressed declaratively. - Code is treated as ephemeral. - Regeneration is cheaper than preservation. ### KISS - Complexity is a liability. - Escalation requires evidence of failure. - Expand only when it hurts — canon explores many concepts so the user surface stays simple. ### DRY (With Isolation) - Duplication is tolerated across bounded contexts. - Shared abstractions require proven reuse. ### Consistency - Behavioral predictability matters more than visual uniformity. - Consistency is scoped, not global. ### Maintainability - Systems must survive creator turnover. - Documentation and explicit tradeoffs are part of the product. ### Antifragile - Failure is assumed. - Recovery paths are preferred over prevention. - Learning velocity is a design constraint. ### Scalable - Growth must be bounded in: - cost - complexity - human attention - Scalability includes cognitive and operational load. --- ## Restartability Over Salvage ODD assumes that restarting from refined intent is often more effective than steering a system that has already drifted. As systems grow, prompts accrete, assumptions harden, and local fixes compound. At a certain point, continued steering optimizes for preserving effort rather than improving outcomes. Restarting is not failure. Restarting is a recognition that: - intent has become clearer - constraints are better understood - evidence from prior attempts now exists In an AI-accelerated environment, restarting is cheap. Misalignment is expensive. ODD therefore treats restartability as a design feature: - prompts are disposable - implementations are ephemeral - canon and intent persist The goal is not to preserve artifacts, but to preserve learning. A clean restart with better constraints is progress. --- ## Progressive Governance (Maturity-Aware ODD) ODD enforcement depends on project maturity. Level 0 — PoC / Exploration - Goal: learn quickly - Artifacts are non-authoritative - Verification demonstrates possibility - Over-governance is prohibited Level 1 — Pilot / Product - Goal: deliver value safely - Evidence and visual proof required - Tradeoffs must be explicit - Silent failure is unacceptable Level 2 — Production / Long-Term - Goal: sustain trust - Outcomes must be measurable - Observability, reversibility, and security are mandatory - Autonomous actions require stop conditions and human gates Maturity must be stated explicitly. See `odd/maturity.md` for the full maturity model. --- ## Evidence as the Gate Completion requires: - observed behavior - produced evidence - self-audit against constraints - explicit declaration of confidence and gaps Assertions do not count as completion. See `canon/constraints/definition-of-done.md` for the full evidence policy. --- ## Trust, Authority, and AI AI is an accelerator, not an authority. - AI may propose and generate - AI may self-audit and verify - AI may not silently assume trust Authority boundaries and escalation points must be explicit. --- ## Outcomes Must Be Falsifiable Outcomes are only valid if they can be: - observed - tested - disproven Non-falsifiable outcomes are treated as goals, not success criteria. --- ## Reversibility and Cost Awareness Prefer decisions that are: - cheap to undo - bounded in cost - limited in blast radius Irreversible decisions require human approval. --- ## Stop Conditions Every autonomous loop must define: - success criteria - failure criteria - exit conditions Endless optimization is a failure mode. --- ## Memory Is the Bottleneck AI didn't just make coding faster. It changed what's scarce. In ODD, generated artifacts are abundant, but **durable intent** is not. So the work shifts toward: - preserving what was learned, - verifying reality, - discarding what cannot be trusted, - and elevating only what repeatedly reduces future drag. ODD stays legible by using **Progressive Elevation & Decay**: most artifacts die at the working layer; only proven patterns elevate into Canon and Decision Trace. See `odd/appendices/progressive-elevation.md` for the elevation model. --- ## Relationship to Canon - ODD → why - Constraints → assumptions - Decision Rules → how - Maturity Model → when - Evidence Policies → proof - Values & Axioms → grounding Together, these form a complete governance layer. --- ## Closing (Internal) ODD is not a philosophy of optimism. It is a discipline of restraint, verification, and curation— designed for a world where generation is infinite, but trust is not. --- ## Confidence, Risks, and Known Failure Modes (ODD v1.2 — Evidence-Based Self-Assessment) This section captures an evidence-based assessment of how well ODD, as currently applied, aligns with its stated principles and where it remains vulnerable. This is not a guarantee of correctness. It is an explicit acknowledgment of uncertainty — now grounded in observed system behavior rather than theoretical projection. --- ### Confidence Model Confidence scores express current belief that ODD will behave as intended when applied thoughtfully. Scale: 0.0–1.0 - 0.9+ — robust under most conditions - 0.7–0.85 — strong, with known boundaries - 0.5–0.7 — plausible, fragile under misuse - <0.5 — likely misaligned without correction Scores are updated based on evidence from real-world application. --- ### What Changed Since v1.1 v1.1 confidence scores were theoretical self-assessments written before ODD had tooling, incident history, or a canon exceeding a few dozen documents. v1.2 scores reflect: - **246+ baseline documents** actively queried by agents - **OddKit MCP server** providing search, retrieval, orientation, challenge, and gating - **Documented incidents** that triggered constraint creation (progressive disclosure failure, anti-cache lying) - **Real agent usage** across exploration, planning, and execution modes - **Progressive Elevation** demonstrated through the promotion pipeline --- ### Principle-Level Confidence Snapshot **Prompt Over Code / Convention Over Configuration** Confidence: 0.85 (up from 0.80) Evidence of strength - 246+ canonical documents function as the source of intent, actively queried by agents through OddKit. - Conventions are enforced by MCP tooling — agents orient against canon, challenge claims against it, and gate transitions using it. - Intent is genuinely first-class; regeneration from canon is practiced, not just theorized. Remaining risks - Conventions silently becoming configuration sprawl as the corpus grows. - Multiple URI schemes (`https://klappy.dev/page/`, `oddkit://`, `odd://`) serving different content sources — functional but not yet documented as a deliberate convention. Failure mode - "Prompt over code" degenerates into "prompt + hidden config everywhere." --- **KISS (Keep It Simple, Stupid)** Confidence: 0.75 (held from 0.75) Evidence of strength - The user-facing surface is simple: orient, search, get, challenge, gate, encode. - The principle "expand only when it hurts" is working — the corpus explores many concepts for depth, but tiers and OddKit routing mean users absorb only what's relevant. - Progressive disclosure ensures documents are useful at every extraction depth, preventing mandatory deep reads. Remaining risks - Meta-layers (tiers, modes, maturity, epochs, elevation levels) are each justified but collectively represent significant conceptual surface area. - Onboarding new humans or agents who lack OddKit may find the corpus overwhelming. Failure mode - Governance becomes heavier than the systems it governs — mitigated by tooling routing, but the risk persists if tooling fails. --- **DRY (With Isolation)** Confidence: 0.70 (held from 0.70) Evidence of strength - Canon centralizes worldview and defaults. - OddKit's search and catalog provide a single discovery surface. Remaining risks - Three URI schemes (`https://klappy.dev/page/`, `oddkit://`, `odd://`) differentiate content source but are not formally documented — a decision record would move this toward 0.75. - Reuse pressure creating brittle shared abstractions too early. Failure mode - "One source of truth" becomes "many partial truths." --- **Consistency** Confidence: 0.72 (up from 0.65) Why this improved - OddKit enforces URI resolution, metadata schemas are standardized, and the Writing Canon (`canon/meta/writing-canon.md`) provides structural conventions. - Consistency violations are now detectable — the progressive disclosure incident demonstrated that violations get caught and documented. - Tooling has moved consistency from pure discipline to enforceable convention. Remaining risks - Small inconsistencies compounding across documents as the corpus grows. - URI scheme convention needs formalization. Failure mode - The system remains logically sound but ergonomically frustrating. --- **Maintainability** Confidence: 0.70 (held from 0.70) Evidence of strength - Separation of stable principles from evolving operations is holding. - Epoch model, migrations, and maturity framework support managed evolution. - Writing Canon checklist prevents structural degradation in new documents. Remaining risks - Curation burden grows with corpus size. - Version semantics are implied but not enforced — manifesto has been v1.1 for an extended period with no v1.2 criteria documented. Failure mode - Canon becomes respected but stale. --- **Antifragile** Confidence: 0.72 (up from 0.60) Why this improved significantly - The original 0.60 was "intentionally cautious" because "antifragility depends on real-world stress, not theory." - Since v1.1: the progressive disclosure failure was documented, analyzed, and produced a constraint change integrated into the Definition of Done. The anti-cache lying incident produced a new canon constraint. These are stress → learn → strengthen cycles — the definition of antifragile. - The system didn't just survive failures — it got better from them. Remaining risks - MCP or tooling layers becoming hidden single points of failure. - Incident documentation becoming rote rather than genuinely analytical. Failure mode - System recovers technically but loses trust socially. --- **Scalable** Confidence: 0.72 (up from 0.70) Evidence of strength - 246+ documents with working search, catalog, and tiered obligation. - OddKit routing means agents don't need to understand the full corpus to work effectively. - Progressive disclosure means documents scale in quantity without proportionally increasing cognitive load. - ODD is structure-agnostic — compatible with monorepos, multi-repos, or any layout where canon is addressable. Remaining risks - Human cognitive load becoming the true bottleneck as conceptual surface area grows. - Discovery degrading if search quality doesn't keep pace with corpus growth. Failure mode - System scales in size but not in usability. --- ### Cross-Cutting Risks **Premature Formalization** ODD is vulnerable to being "locked in" too early, reducing exploration. Mitigated by maturity model and "expand only when it hurts" — but the risk remains that documented patterns calcify before they're proven. **False Authority** Well-written governance can be mistaken for correctness without evidence. Mitigated by the evidence-as-gate requirement and axiom "A Claim Is a Debt" — but the risk persists, especially for agents that absorb canon without questioning it. **Silent Drift** Small deviations, left unnamed, can erode trust over time. Mitigated by incident documentation and drift checks — but drift in less-visible areas (naming conventions, metadata schemas, cross-references) requires ongoing vigilance. **Tooling Dependency** OddKit now mediates most agent interactions with canon. If OddKit degrades or serves stale content, the epistemic surface degrades with it. The anti-cache lying constraint was created specifically to address this risk, but tooling as a single point of failure remains a cross-cutting concern. --- ### Intended Use of This Section This section exists to: - prevent ideological hardening - make risks discussable - encourage re-evaluation - model intellectual humility It is expected to change. --- ### Re-evaluation Philosophy ODD should be reassessed when: - it is applied to real production systems - autonomous agents operate for extended periods - failure modes surface that are not addressed here - confidence scores remain unchanged for more than two epochs without new evidence Confidence should be updated based on evidence, not optimism. --- ## Status - ODD v1.2 Extended updated - Confidence scores updated from evidence-based assessment - Cross-references verified against current repo structure - Tooling (OddKit MCP) reflected as realized infrastructure - Fully aligned with Writing Canon progressive disclosure requirements --- For common failure modes and practical misapplications of ODD, see `odd/misuse-patterns.md` and `odd/prompt-architecture.md`.