Method: Revision Lens Sequence

Draft-zeros compress source material into structured essays. They consistently miss the same concerns — not because the author or agent is unaware of them, but because applying all governance lenses at once exceeds what a single pass can hold in focus. The observed pattern: the author asks the same sequence of improvement questions after every draft-zero, and each question improves the output whether or not the author has read the draft.

This is a signal. When the same corrections are needed every time, they should be automated as sequential passes — not collapsed into a single checklist that produces shallow compliance on each point.

This does not mean multi-lens passes have no value. Quick checks with all lenses active — preflight before execution, inline validation along the way, sanity checks during drafting — are fast, valuable, and appropriate throughout the writing process. They catch obvious issues early and keep the work directionally sound. What they cannot do is substitute for the thorough, single-focus reads that formal verification requires. A quick multi-lens scan might catch that the blockquote is missing. It won't catch that the theological section drifted from Socratic voice into sermonic declaration while maintaining perfect guide posture.

The distinction: multi-lens passes are how you work. Single-lens passes are how you verify. Both have a place. This method governs the verification sequence — the formal gate checks that happen after draft-zero and after each major revision prior to publishing.

The method: after draft-zero — and again after each major revision — the agent executes a defined sequence of single-lens revision passes. Each pass reads the entire document through one specific governance concern. Passes are discrete operations — the output of each becomes the input of the next. The sequence ends with a regression check that re-reads the document looking for damage introduced by earlier passes, followed by an oddkit gauntlet (preflight, challenge, validate) as the formal gate before claiming the document is ready.

Each pass applies one lens. The agent executes the pass, presents what changed, and waits for author direction before proceeding. The author may skip, reorder, combine, or add passes based on what the specific piece requires.

Pass 1 — Guide Posture

Lens: Is the reader the hero? Does the essay open with their pain, not the author's system?

Governing document: canon/constraints/guide-posture.md

What it catches: Opening with system features instead of reader experience. Leading with "here's what I learned" instead of "here's what's frustrating you." The system revealed as product rather than plan.

Three-question test: (1) Who is the hero? (2) Does it open with their pain? (3) Is the system revealed as a plan, not a product?

Pass 2 — Socratic Voice

Lens: Are assertions presented as declarations or as discoveries the reader makes through guided questions?

Governing document: canon/constraints/guide-posture.md + observed author preference across all essay sessions

What it catches: Declarative operator voice where guide voice belongs. Telling the reader what to think instead of asking questions that lead them to the same conclusion. Sermonizing — especially in theological sections.

The test: Read each paragraph. If it asserts a conclusion, ask: could this be reframed as a question that lets the reader arrive at the same place? First-person confession is exempt — testimony is not assertion.

Pass 3 — Confession Before Accusation

Lens: Where conviction is needed, does the essay use "we" not "you"? Is the author included in the indictment?

Governing document: docs/book/governance-additions-2026-02-27.md (Rule 10)

What it catches: Second-person accusation disguised as exhortation. "You need to rest" instead of "we need to rest." Finger-pointing that the reader can dismiss as not applying to them.

The test: Find every sentence that convicts. Is the author in the sentence?

Pass 4 — Relational Sensitivity

Lens: Would every person and organization referenced recognize themselves and feel respected?

Governing document: canon/constraints/relational-sensitivity.md

What it catches: Named individuals who should be anonymous. Organizational friction framed as personal failure. Colleagues cast as villains rather than co-navigators of structural challenges. Quotes attributed without speaker verification.

Five constraints: (1) Structural not personal. (2) Show what worked. (3) Honor both relational and political tensions. (4) Consistent pseudonyms when needed. (5) Shared confession, not finger-pointing.

Pass 5 — Factual Verification

Lens: Are all claims verifiable? Are attributed quotes real? Is scripture accurate?

Governing document: canon/values/axioms.md (Axiom 2: A Claim Is a Debt), docs/book/governance-additions-2026-02-27.md (known hazards)

What it catches: AI-fabricated attributed quotes. Scripture contamination across translations (must verify against BSB). Overclaiming — lived experience stated as demographic fact, directional trends as absolutes. Contradiction with published chapters or essays.

The test: Every attributed quote verified against source material. Every scripture reference mechanically checked. Every "always" and "never" questioned.

Pass 6 — Progressive Disclosure Audit

Lens: Does the document work at every extraction depth — title, blockquote, metadata, summary, full?

Governing document: canon/meta/writing-canon.md

What it catches: Titles that don't signal stance. Blockquotes that tease instead of delivering the compressed argument. Summaries that require the body to make sense. Headers that read as generic form instead of telling the story. Buried claims not present at higher tiers.

Seven-point checklist: Title test, blockquote test, metadata test, summary test, header scan test, no buried claims, axiom space test.

Pass 7 — Author Identity and Frontmatter

Lens: Does the frontmatter comply with the schema? Does author identity language follow constraints?

Governing documents: canon/meta/frontmatter-schema.md, canon/constraints/author-identity-language.md

What it catches: Quoted booleans and integers. Missing required fields for public essays (slug, author, type, public, description, hook). Identity violations ("Bible translator" instead of "software builder in Bible translation"). YAML type mismatches that cause blank pages.

Pass 8 — Regression Check

Lens: Did any earlier pass introduce damage that another pass would have caught?

Governing document: All of the above, re-applied as a single read-through.

What it catches: Socratic voice pass that broke guide posture (e.g., opened with a question about the system instead of the reader's pain). Relational sensitivity pass that introduced declarative voice. Factual verification pass that lost the Socratic framing of a corrected claim. Any pass that solved its own concern while creating a new one.

The test: One complete read-through holding all lenses simultaneously — but only looking for regressions introduced by the current pass sequence, not attempting new improvements. If regressions are found, fix them and note the tension for the author.

Pass 9 — oddkit Gauntlet

Lens: Does the document pass the epistemic tools?

Actions: oddkit_preflight → oddkit_challenge (with explicit counter-arguments) → oddkit_validate

What it catches: Constraints the agent missed. Claims that don't survive pressure-testing. Completion gaps. NEEDS_ARTIFACTS is expected for text commits — not a failure.

This sequence applies to all public essays (writings/) and book chapters — after draft-zero and again after each major revision prior to publishing. A "major revision" is any change substantial enough that earlier passes may no longer hold: new sections added, source material incorporated, structural reordering, or feedback that reshapes the argument. Minor edits (typos, single-line fixes, frontmatter updates) do not require a full sequence. Internal canon documents (canon/, docs/, odd/) follow the Writing Canon gate but do not require the full lens sequence — the Socratic voice, confession-before-accusation, and relational sensitivity passes are specific to public-facing content.

Not every piece requires every pass at equal depth. After each milestone (draft-zero, major revision, feedback incorporation), the agent proposes the sequence and notes which passes are high-value for this specific piece. Examples:

• An essay with theology → Pass 3 (confession) is high-value
• An essay with named colleagues → Pass 4 (sensitivity) is high-value
• An essay with no scripture → Pass 5 (factual) is lower-value
• A governance doc → Passes 2, 3, 4 are skipped; Passes 6, 7 are primary

1. After each milestone: The agent proposes the lens sequence with a brief note on which passes are high-value for this piece.
2. One pass at a time: The agent executes one pass, presents what changed (specific edits, not just descriptions), and waits for author direction.
3. Author controls the pace: The author may say "do the next three" or "skip sensitivity, we'll do that after the reviewer reads it" or "combine Socratic and confession."
4. Regression check is strongly recommended: After all content passes, the agent runs the regression check unless the author explicitly skips it. It exists because the observed failure mode is real: each lens fixes its own concern while occasionally breaking another.
5. oddkit gauntlet closes: Preflight, challenge, validate. This is the formal gate before claiming the document is ready.

This method was encoded because the author observed the same pattern across multiple essay sessions: after every draft-zero, the same questions were asked in the same order, and each question consistently improved the output. The questions themselves became a signal — if they're always needed, they should be proactive, not reactive.

The specific trigger was the "Your Context Window Needs a Sabbath" session (April 7, 2026), where the author asked for a Socratic voice pass after the draft-zero and then asked: "How do we address this with governance of multiple passes to address these missed concerns, as it's too much in a single pass but governance of new lenses each pass?"

The answer: sequential single-lens passes, each a complete read-through with one concern, ending with a regression check to catch what earlier passes broke, followed by an oddkit gauntlet to formally gate completion. Subsequent observation confirmed that even sequential passes from the same agent have structural blind spots — an independent reviewer with fresh context and the same governance consistently catches what the authoring agent cannot, because the same lenses used to create are the same lenses used to evaluate.

• Canon-Integration Audit — the three checks this method is missing: concept audit, adjacent-canon audit, validator-completeness audit; extends the lens sequence with integration-layer gates
• Writing Canon — the progressive disclosure requirements each pass must preserve
• Guide Posture — the constraint governing Passes 1 and 2
• Relational Sensitivity — the constraint governing Pass 4
• Book Governance Additions — the source of Passes 3 and 5
• Self-Audit Checklist — the reflection layer this method extends