Update log

September 26, 2026 / Protocol 0.3-preview.1

Decompress the engineering reasoning before accepting generated implementation.

H.A.R.D. 0.3 makes decision compression the explicit motivating model. The review now follows Decompress -> Model -> Challenge -> Prove -> Decide. For consequential software choices, reviewers expose only the relevant source of truth, ownership, state, boundary, contract, failure and time/ordering model instead of treating generated code as its own explanation.

Consequential assumptions are first-class records with supported, conflicted or unassessed status, consequence if false, evidence needed, retained evidence and a revisit trigger. Bounded challenges must identify their evidence level, so a walkthrough pass cannot be presented as implementation or runtime proof. G3 and G4 deepen these checks while the six gates and fifteen criterion IDs remain.

G6 now favors the smallest coherent software slice that can expose the important assumption end to end before a broad build. CSV transport 2.1 adds assumptions.csv, decision-surfaces.csv and challenge-scenarios.csv. The MCP, Skill, schemas, examples, readable pages, PDFs and packages use the same contract.

Decision compression and implementation outrunning understanding remain conceptual motivations, not measured causal findings. The 241-test regression suite verifies implementation behavior only. A finalized release still requires genuine version-matched bounded external use.

September 26, 2026 / Protocol 0.2-preview.3

Review decisions, with an official path for small teams.

The review now connects important choices to purpose, alternatives, rationale provenance, tradeoffs, verification and accountable human decisions. Artifact Review supports plans, prototypes and code at an explicit stage. Independent evaluator measures are optional and report N/A reasons when roles or order make them ineligible.

CSV templates and worked examples share a canonical schema and retain all four evidence levels. Machine-readable artifact populations prevent mixed-population summaries. Severity and deferred-scope rules are explicit. The optional model companion has been withdrawn from the current distribution.

This change responds to author intent and reported practical friction. Software regression results and practitioner feedback do not demonstrate method effectiveness. Earlier protocol releases retain their original versions.

September 25, 2026 / Method section layout

Read the purpose, then follow the review.

The method section no longer places its research explanation beside an equally prominent procedure. The explanation uses a bounded reading column, followed by a separate practice section with a smaller heading. Three numbered steps sit across wider screens and stack vertically on smaller screens. No extra buttons or expandable panels were added.

Academic Humanize V2 informed the shorter step headings and direct instructions. The structure view remains optional, and the proposed study has not established an effect. This is a presentation change, not evidence of improved usability. No gate, criterion, risk rule, formula, tool version or published download changed.

September 25, 2026 / Research framing and alignment

The research question, with an app-focused example.

The overview now states the research question directly: how does visual fidelity affect readiness judgments and attention to structural issues? Personal recollection and architectural imagery have been removed from the current page. A newly generated app concept shows an interface beside requirement, recovery and evidence prompts. It is an illustration, not a working H.A.R.D. application or a study result.

The introduction and method use equal columns with shared top alignment. Review-path cards share title, note and action rows, and the four review perspectives use a two-by-two layout on wider screens. Shared Plot components and theme tokens remain in use.

Academic Humanize V2 was applied to the revised explanation. The proposed study remains untested; no improved user outcome is claimed. No protocol gate, risk threshold, formula, tool version or published download changed. Earlier images and update entries remain preserved.

September 25, 2026 / Naming revision

Decision describes the next step.

The name is displayed as H.A.R.D. to read as initials. The overview now explains the missing review layer under a polished artifact, using one constructed booking example. Its two first-screen paths are a workflow review and the Developer Library.

The library owns detailed rules, installation and integration reference. Repeated overview buttons and the persistent setup rail are removed; setup unfolds when requested. The update-log table now loads the shared style it needs.

The protocol, MCP and Skill explain a separate structural view of information, workflow, implementation evidence and human operation. Preserve the original artifact and mark unknowns. This is an unvalidated practice aid, not a new gate or an unplanned intervention in the fidelity study. The architecture anecdote is research context, not a measured result.

The full name is now Human-centered AI Readiness and Decision Protocol. H.A.R.D. Protocol 0.2 and Public Preview stay the same. Deployment sounded like technical release approval; Decision describes the actual purpose: decide whether to invest in building a bounded workflow, revise it, or gather missing evidence.

The six gates, fifteen criteria, risk depth, formulas and evidence requirements have not changed. This review still supports an engineering commitment, not permission to put a system into use.

On smaller screens, scroll the version table sideways.

New downloads have separate URLs. Past packages and records keep their original names, versions and hashes. Existing package names, commands and ai-ready invocation remain supported. No new pilot, empirical validation or operational performance is claimed.

September 24, 2026 / Website layout update

Clearer starting paths and a shared visual system.

The overview now pairs its download card with the decision, evidence and next action a review should leave behind. Starting roles use Plot's existing persona-grid components. They are role-based guidance, not research-validated personas.

Detailed requirements live in the searchable developer library. The overview links to all fifteen criterion IDs without repeating the full reference. MCP setup is maintained in one shared install component.

The design-system contents now link directly to twenty molecules, twenty organisms and five templates. Eleven native SVG drawings replace text-based anatomy and layout sketches. Their labels and shapes follow the selected theme.

Theme corrections include status and example panels, muted text, placeholders, input boundaries, accent-button text and keyboard focus. The component search uses a native dialog. Selected controls and layouts were checked in light and dark themes; these checks are not a whole-site accessibility certification or a user study.

Reference and adaptation

The developer library takes navigation and reference-reading cues from Microsoft Learn. The overview takes task-oriented section and resource-grouping cues from Microsoft Graph Adoption. Both use Plot's own components, typography and theme tokens. No affiliation is implied.

What did not change

Protocol 0.2-preview.1, MCP 0.2.0rc7 and Skill/contract 0.2.0-rc.7 remain unchanged. No frozen download, decision gate, criterion, risk rule, formula or past run has been relabeled. Public Preview remains the release status.

No new pilot, live-provider evaluation or performance evidence is reported by this website update.