StarLedger v1.0.0-beta.5 — post-beta.4 maintenance & template-cleanliness hardening

Post-beta.4 maintenance & template-cleanliness hardening checkpoint. No new runtime features — this release closes the PR1–PR5 cleanup cycle and restores source ↔ template release parity.

Changes since v1.0.0-beta.4

• Public / template-facing documentation hygiene (#330) — scrubbed local/maintainer paths and clarified repo-derived build values (e.g. the Pages base / favicon path are derived from GITHUB_REPOSITORY, not hardcoded).
• Template-builder: explicit skills-classification exclusions (#331) — skills-classification.json, skills-classification-meta.json, skills-classified.md added to the deny set.
• Template-builder: deny-over-allow local docs/reviews exclusion (#332) — docs/reviews/** local review drafts can never ship in the generated template.
• setup-doctor: detect skills-classification artifacts (#333).
• setup-doctor: detect discovery-candidate artifacts (#334).
• Defense-in-depth: builder PREVENTs owner/generated artifacts from being copied; setup-doctor DETECTs them if they ever appear in a supposedly clean tree.

Template parity

F-e-u-e-r/starledger-template was regenerated from this release through the canonical one-way builder and tagged v1.0.0-beta.5 in parity.

Versioning remains git-tag + Release driven (package.json stays 0.1.0, unchanged across the beta series).

StarLedger v1.0.0-beta.4 — WS4-SL Portfolio-family visual alignment

Fourth public beta — the WS4-SL Portfolio-family visual alignment milestone. StarLedger adopts the Portfolio product family's visual identity. No change to the public data contract, annotation/meta byte contracts, or executor behaviour; no new runtime subsystem.

WS4-SL — Portfolio-family visual alignment (since beta.3)

• Light/dark emerald design system — a Portfolio-inherited brand-token palette; light is the default, dark via @media (prefers-color-scheme: dark); green-alpha border language.
• >_ StarLedger identity — a bordered emerald >_ prompt mark beside the StarLedger wordmark; a matching emerald >_ favicon. The mark is decorative; the wordmark is the sole accessible name.
• Optional ← Portfolio return link — opt-in via the build-time VITE_PORTFOLIO_URL variable (fed from the PORTFOLIO_URL Actions repo variable). Absent/empty ⇒ nothing renders. The maintainer's URL is never hardcoded in source, so it does not ship into starledger-template.
• JetBrains Mono technical typography — self-hosted .woff2 (SIL OFL-1.1), same-origin under font-src 'self' (no CDN, no CSP change), metric-matched fallback + font-display:swap for zero layout shift; scoped to technical / system-facing roles.
• Accessibility / contrast hardening — a brand-token contrast regression gate and favicon-wiring tests; AA contrast across light and dark surfaces.
• Durable brand-token contract — docs/brand-tokens.md + docs/adr/ADR-005-portfolio-visual-alignment.md record the token system and the opt-in Portfolio semantics.

Also since beta.3: routine daily stars.json sync + AI-annotation enrichment commits (data-pipeline only; not part of the visual milestone).

Verified: Codex gpt-5.6-sol max + Grok cross-family review; full CI green; GitHub Pages deployed; live light/dark + 320/390/768/1280/1440 responsive verification. Prerelease; the tag points at this repository's own canonical main (001f0628).

StarLedger v1.0.0-beta.3 — #207 5b outcome-contract hardening

Third public beta — the maintenance / freeze checkpoint. Narrowly closes the last deferred item of #207, item 5b (circuit-breaker outcome classification), at the contract level. No change to public data, annotation/meta byte contracts, or executor behaviour; no new runtime subsystem.

#207 item 5b — terminal outcome contract (a classification fix, NOT a new feature)

• plan_summary gains a required deferred_terminal count — a tally of repos skipped as terminally-unavailable (skip-terminal), exactly parallel to the existing deferred_retry. A terminally-unavailable repo's zero-job residual is now machine-readable and can never collapse into the success-like drained-at-base.
• The owner-run P3 completion gate gains a second, independent fail-closed assertion (deferred_terminal === 0) beside the empty_reason precedence, and fails closed on a manifest predating the field.
• No circuit breaker, terminal-state writer, retry engine, or scheduler was added. Both the retry-backoff and terminal failure-bookkeeping paths remain dormant: no runtime code originates attempts > 0 or terminal_unavailable = true (the planner only carries prior state forward). This is forward-compatible scaffolding, so that if such a writer is ever built — or state is seeded externally — its page-the-operator residual is already classified and gated, never mistaken for a drained backlog. The empty_reason enum is unchanged.
• Labeled defensive-contract falsification tests pin the terminal residual at both the planner and the gate; the dormant / no-write-path status is documented consistently across schema, runtime, and the routine spec.

Reviewed across two independent model families (grok-4.7 + gpt-5.6-sol + gpt-5.6-luna); full CI green. Prerelease; tag points at this repository's own canonical main.

StarLedger v1.0.0-beta.2 — #207 classifier hardening

Second public beta. Hardens the P3 AI-classification pipeline per #207 (follow-up to #205); no change to public data, annotation/meta byte contracts, or executor behavior.

Classifier hardening (#207)

• ClassificationManifest gains an optional, additive plan_summary — pre-budget eligible_new/eligible_retry/eligible_refresh counts, an omitted_unfetchable count, a deferred_retry count, and an empty_reason enum — so a zero-job manifest is machine-checkable instead of parsed from stdout prose.
• The owner-run P3 completion gate is now sound: under a 0 refresh/retry budget a changed-fingerprint or retry-backoff repo yields 0 jobs while staying annotated, which the old jobs==0 + node_id-set-equality check falsely passed. The gate (now a pure, unit-tested module with falsification cases) requires empty_reason=drained-at-base with every eligible/omitted/deferred count at zero.
• Corrected two stale "disable the routine" doctrine comments and two executor-runbook / routine-spec doc drifts (base-freshness file list; ai.budget.* key paths).

Reviewed across two independent model families; full CI green. Prerelease.

StarLedger v1.0.0-beta.1 — First public beta

StarLedger v1.0.0-beta.1

This is StarLedger's first coherent public beta.

What is included

• Repository discovery and filtering
• Optional classification-aware browsing
• Identity and repository metadata improvements
• Group-by-category browsing
• Coherent AI annotation readiness
• Deterministic Insights
• Shared @starred/insights deterministic core
• Context Builder:
  • exports the full filtered repository set
  • Markdown or versioned JSON
  • repository description
  • existing category and tags
  • preview before copy
  • deterministic portable context for use with ChatGPT, Claude,
    Codex, and other agents

Product direction

StarLedger currently prioritizes human-directed repository discovery and
portable context handoff.

The previously planned read-only MCP server is intentionally deferred,
not cancelled. It will be reconsidered when real usage shows demand for
agent-directed autonomous exploration.

Release metadata reset

Earlier v1.0.0-rc.*, v1.1.0-alpha.*, v1.2.0-alpha.*, and
v1.3.0-alpha.* tags were experimental development checkpoints and were
not maintained under one consistent public-release contract.

For this beta, the legacy release metadata was intentionally reset.

Going forward:

• public release tags are immutable;
• release tags are created from an explicitly audited commit;
• every public tag must have a matching GitHub Release;
• experimental internal checkpoints should not be published as permanent
  public release tags unless intentionally released.

Release commit

68427a4