HISTORY

Changelog

Current version: v0.3.2, the first public release of the alpha. All versions are Milber Ferreira's, developed with OpenAI's and Anthropic's models.

Versions

Lineage: initial research → v0.1 → research review (section 0 of v0.2) → v0.2 → external review and assessment → v0.3 → alpha readiness review, SliceOps verification and naming assessment → v0.3.1 → name decision and pre-release copy review → v0.3.2.

Newest first. Each entry lists what was added, changed, removed and fixed.

v0.3.2 (Public Alpha) — 17 October 2026

Patch release: no rule changes. First version published in the public repository.

Added

• What AI changes, and what it doesn't: a worked authentication example (section 1.2).
• Quick estimate: a one-minute F1 estimate from the four parts of the forecast formula, with starting suggestions per risk tier (section 8.8, informative), and a quick-estimate view in the calculator.
• Definitions of the hierarchy, horizons, sizes, AI leverage and risk, previously only in v0.1 (sections 5.0 and 8.0).
• How to join the pilot, and a plain data-handling notice in the pilot protocol.

Changed

• The public name is SLICE Framework, SLICE for short; "working name" is removed everywhere (decision D30).
• The homepage, README and spec opening say directly that SLICE is for planning and estimating software when AI writes the code (D31).
• The workflow in section 10 is written out in text.
• The pilot size in section 14 now matches the pilot protocol: 5–10 teams.
• Version history moved out of the spec's opening and into this changelog.

Fixed

• Leftovers from the original drafting tool in the archived specs: an "@" before the author's name, references to tabs that don't exist on this site, and placeholders where diagrams were.

v0.3.1 (Public Alpha) — 16 October 2026

Patch release: no rule changes.

Added

• Public-alpha status note and the alpha principle (section 14.3).
• A SliceOps comparison in the landscape (section 3).
• The public alpha kit: start-here page, Quick Start, templates and pilot protocol.

Changed

• Pilot thresholds labelled exploratory; F3 forecasting marked unavailable during the alpha.
• SLICE confirmed as a provisional working name; the public name is still to be decided.

Fixed

• v0.3 wrongly said SliceOps could not be found.

v0.3 — 16 October 2026

Added

• A fixed vocabulary: generation speed, implementation effort, delivery lead time, verification effort, total delivery cost, cost of learning, value delivered, Accepted, Resolved.
• Proportionate Planning: Coordination Budget, Decision Gate, Learning Investment.
• Mode E, Experiment to decide, with a table matching each test to the uncertainty it can resolve.
• Item types Bet, Obligation and Improvement, with a guard on Improvement.
• Separate delivery (Ready → Accepted) and outcome (Proposed → Resolved) lifecycles.
• Three minimum records: Roadmap item, Slice card, Evidence record.
• A high-risk rule for solo founders; delivery-economics metrics; pilot hypothesis H6 on overhead.
• Faros Speed Trap, OpenSpec, Kiro and Evidence-Based Management in the evidence and landscape.

Changed

• Thesis: "Building is cheap" became "When implementation stops being the slowest step, delivery is limited by deciding and proving."
• C: Capacity-bound became Constraint-aware, with verification as the default constraint.
• Forecasting: sample minimums became floors; F3 needs back-tested calibration; small teams report P50 and P85 only; every range shows its sample count.
• Spec storage: repository-only became one authoritative, accessible, versioned source.
• Mandatory metrics: review wait became wait at the constraint; validated-outcome rate became outcome resolution.
• Kill criterion became decision criterion; the checkout example now uses count-based criteria.

Removed

• Decision Budget and Mode B (Build to decide).
• Removal rate as a metric.

Fixed

• Spec Kit had been described as feature-only; it now includes idea assessment.
• The 4% → 6% conversion example could not be detected at startup volumes.
• The independent-review rule was impossible for a solo founder.
• Delivery "done" had been tied to outcome review.

v0.2 — 16 October 2026

Added

• Planning split into coordination, decision and proof.
• Decision Budget with Mode B (Build to decide) and a fifth stop-rule question.
• Bets with outcome hypothesis, kill criterion and review date.
• Specification as agent context, in three layers.
• Verification capacity as the WIP limit, with risk-tiered verification.
• Decision wait in the forecast formula; codebase type in reference classes; minimum evidence per F-level.
• Metrics: decision latency, idea-to-evidence time, review wait, unreviewed merges, validated-outcome rate, removal rate.
• Evidence base (DORA, METR, Faros, Stanford, Andrew Ng) and landscape (AI-DLC, BMAD, Spec Kit, Shape Up, Kanban, XB Software).
• Pilot hypotheses H1–H5 and a red team.

Changed

• C: Confidence-led became Capacity-bound.
• Core line became "Decide deliberately. Build small. Count only what is proven."
• Scope narrowed from "solo founder to enterprise" to teams of 1–25.
• Conformance rules grew from seven to ten.
• Planning Overhead Ratio replaced by coordination minutes and decision latency.

Fixed

• Normative voice; interface leftovers removed.
• Two corrections to the initial research: METR's 2026 update, and keeping the multiplier explorer out of the spec.

v0.1 — 16 October 2026

• Founding specification, built on the initial research: SLICE Roadmapping and SLICE Forecasting; principles Strategic, Lightweight, Iterative, Confidence-led, Evidence-driven; the five-level hierarchy; Now/Next/Later; four item types; the Planning Budget with modes 0–3 and the stop-planning rule; XS–XL sizing, AI leverage, risk and F0–F3; Direction/Target/Commitment; seven conformance rules.