# Building a Portfolio That Does Not Outrun Its Evidence **State:** VERIFIED PRE-PUBLICATION METHOD / FULL RELEASE CASE NOT YET EARNED **Method version:** `portfolio-evidence-method-v1` **Reviewed:** August 25, 2026 The portfolio is being built as an interface to evidence, not as a substitute for it. That changes the order of work. Source recovery comes before résumé language. Claim boundaries come before case-study polish. Public/private separation comes before downloads. Reproduction comes before a `verified` label. Publication and application submission remain separate approval gates. ## The method ### 1. Recover provenance Start with existing artifacts, prior decisions, source manifests, observed failures, and the surrounding system. Preserve found character and distinguish authored texture from defects, incomplete work, and speculative opportunities. ### 2. Inventory before selecting Record the broader candidate pool before choosing homepage work. This prevents the homepage from becoming the entire portfolio and makes omission a deliberate curation decision rather than accidental loss. Phase 1 inventoried 33 public-facing documents/specifications, 21 stable slugs, and 12 publication briefs. Duplicate slugs and missing brief-boundary fields were zero after correction. ### 3. Separate evidence from claims Every substantive résumé or website statement must map to an inspected source and retain an honest boundary. A test result may support a scoped implementation claim; it does not automatically support production readiness, adoption, savings, client outcome, or universal behavior. Unsupported résumé claims—including chronology, education, certifications, quantified client outcomes, production readiness, and broad consulting scale—remain omitted. ### 4. Split public and protected material Internal source manifests retain provenance paths and protected mappings. Public artifacts use sanitized structural derivatives or explicitly fictional fixtures when real material is unnecessary or unsafe. Examples: - Harbor Room is wholly fictional and synthetic. - The public Studio incomplete packet preserves missing-data categories without claiming to be the real room packet. - The acoustic source board remains private pending permission and outcome closure. - Internal router traces were not converted into a public explorer when component participation remained unproven. ### 5. Reproduce the bounded package A public package earns verification only after its stated contract is rerun, outputs are inspected, privacy boundaries are checked, and integrity hashes are recorded. Current completed examples: - deterministic tag lane: 22/22 tests and 12/12 replay; - Harbor Room fixture: 7/7 twice with byte-identical archive/result output; - Studio Readiness Gate: 6/6, with incomplete and complete states remaining distinct; and - Public Test and Evaluation Index: nine records with zero validator errors. ### 6. Preserve failure after correction Passing reruns do not erase the original failure. Structural, behavioral, evaluator, evidence-retention, environment, and integration failures require different corrections and different promotion decisions. The Router Trace Explorer demonstrates the rule: direct contract evidence passed, but real interface participation and a frozen ablation did not. The explorer was withheld and a closure record was published locally instead of reconstructing the missing proof editorially. ### 7. Keep status visible Public metadata distinguishes `verified`, `partial`, `blocked`, and `open`. These states are not collapsed into one success percentage. Synthetic status and known limitations remain visible at index, detail, and package levels. ### 8. Gate external action separately Local assembly is not publication. A website preview is not deployment. A finished résumé is not application submission. A cross-domain link is not endorsement. Each external action requires its own explicit approval and final verification. ## Current control state `controls.v1.json` records twelve method controls. Eight are complete, two are active, and two remain gated. The active/gated controls are not defects to hide; they represent work that cannot honestly be marked complete before final website verification, publication approval, and bounded post-launch observation. ## What this method has earned It supports a public Evidence Library note explaining how this portfolio controls provenance, claims, privacy, synthetic fixtures, reproduction, failure preservation, and publication authority. It does not yet support a full website-design-and-publication case. That later case requires implemented deeper pages, final accessibility/link/security/mobile/desktop verification, approved `.ai`/`.co` relationship language, a deployment identity and rollback record, explicit publication approval, and post-launch observation. ## Claim boundary This is a pre-publication process record. It does not establish production security, public availability, audience impact, usability, uptime, search performance, conversion, adoption, or the effectiveness of a deployment that has not yet occurred.