Migration: H.A.R.D. 0.2-preview.3 to 0.3-preview.1

H.A.R.D. Protocol 0.3 · Public Preview

Exact current versions: protocol 0.3-preview.1; MCP 0.3.0rc1; Skill/contract 0.3.0-rc.1.

Why this is a 0.3 preview

Preview.3 already inspected purpose, alternatives, rationale provenance, tradeoffs, four evidence stages and human disposition. 0.3 changes the machine-readable meaning of a consequential choice by requiring explicit engineering-deepening triage and by adding structured system-model and challenge records. That is a contract change, not an editorial revision.

The six engineering gates and fifteen criterion IDs are retained. Published 0.2 files and runs remain frozen and must not be relabeled.

ChoiceRecord changes

Every current ChoiceRecord now requires:

• assumptions,
• engineering_deepening_required,
• deepening_rationale.

When engineering deepening is true, it also requires:

• at least one deepening_trigger,
• explicit required_surface_kinds,
• matching decision_surfaces,
• at least one challenge_scenario,
• next_coherent_slice.

Decision surfaces may be truth, ownership, state, boundary, contract, failure_recovery or time_ordering. Their status is supported, conflicted, unassessed or not_applicable with the evidence rules in the schema. Consequential assumptions are separate first-class records with supported, conflicted or unassessed status plus consequence-if-false, evidence-needed, retained evidence and revisit-trigger fields.

A historical rationale is still different from a new current justification. The new fields do not permit reconstruction of private model chain-of-thought or unrecorded creator intent.

Gate mapping

• G2 now requires the choice's assumptions and explicit deepening rationale.
• G3 evaluates required decision surfaces when deepening applies. Missing/unassessed surfaces are missing evidence; conflicted surfaces are failures.
• G4 evaluates bounded challenge scenarios and exact decision-context fingerprints. A changed decision surface invalidates affected prior validation.
• G6 retains human disposition and the bounded engineering commitment. For software work, the next step should be the smallest coherent slice that exposes the important assumption where applicable.

CSV transport 2.1.0

choice-review.csv adds:

• engineering_deepening_required_yes_no
• deepening_rationale
• deepening_triggers
• required_surface_kinds
• next_coherent_slice

Two canonical files are added:

• assumptions.csv, one row per consequential assumption lifecycle
• decision-surfaces.csv
• challenge-scenarios.csv, with explicit evidence_level for every assessed challenge

Templates and completed examples must use the exact same headers. The verifier does not infer missing rows from prose.

Migration rule

Do not mutate an old 0.2 review into a 0.3 pass. Preserve the old record and create a linked new review. Carry forward only evidence that still refers to the same artifact and decision context; reassess applicability and rerun affected checks when the new decision model changes the context fingerprint.

Compatibility

Existing command names, MCP tool names and the ai-ready invocation remain supported. Exact version matching is still mandatory. Old 0.2 records fail current schema validation by design and remain valid only as records of the version that created them.

Claim boundary

This migration supplies software-tested structure for reasoning-decompression records. It does not establish that decision compression occurs at a measured rate, that H.A.R.D. reproduces expert cognition, or that the new lens improves software outcomes.

Execution versions: protocol 0.3-preview.1 · MCP 0.3.0rc1 · Skill/contract 0.3.0-rc.1.