BriefLoop v0.15.3 — Solar Stock Market Data and Review Surfaces

BriefLoop v0.15.3 — Solar Stock Market Data and Review Surfaces

BriefLoop v0.15.3 extends the fresh SQLite/Codex path with the structured
market-data surface needed by Solar Stock Periodic. It also carries forward the
v0.15.1 post-final AI Second Opinion and the v0.15.2 deletion of the legacy JSON
control-plane runtime.

The release remains deliberately narrow: BriefLoop records how sources, claims,
market data, checks, and Human decisions move through a briefing workflow. It
does not prove investment conclusions, provider reliability, semantic truth, or
report quality.

What is included

• Fresh-only SQLite schema 19 and strict briefloop.market_data_snapshot.v2.
• Profile-bound ingestion of the TOYO weekly XLSX layout, plus strict JSON/CSV
  market-data inputs.
• A bounded Yahoo adapter that may fill missing adjusted-close history, FX,
  securities, or fields; verified workbook cells remain authoritative.
• Product-owned recalculation of latest price, period returns, USD conversion,
  and USD market capitalization from frozen inputs. Excel formula caches are
  comparison-only.
• Store-bound primary and overseas comparison tables, an event timeline, seven
  deterministic PNG charts, a JSON read model, a hash-bound chart manifest,
  and a Market Data tab in the local HTML report.
• Solar market-data Gates for missing required series, report-window mismatch,
  blocking workbook gaps, and input conflicts.
• The schema-18 Solar discovery contract carried forward: 20 independent
  Tavily tasks, up to 20 advanced Search results per task, one deterministic
  targeted backfill for an under-covered task, Batch Extract in groups of 20,
  and an 800-unique-URL safety envelope.
• Store-qualified post-final AI Second Opinion, Human observations, separately
  approved guidance, and explicit same-workspace successor opt-in.

Start a fresh Solar workspace

Python 3.12 is required. Existing development-schema workspaces are not
upgraded in place.

briefloop new solar-stock-periodic ./solar-weekly \
  --search-backend tavily \
  --report-window-start 2026-08-03 \
  --report-window-end 2026-08-12

briefloop run --workspace ./solar-weekly --runtime codex
briefloop runtime continue --workspace ./solar-weekly

For a profile-bound workbook:

briefloop market-data ingest \
  --workspace ./solar-weekly \
  --file ./weekly.xlsx \
  --profile toyo-weekly-v1

briefloop market-data project --workspace ./solar-weekly

Authority and evidence boundaries

• SQLite ControlStore Receipts remain the sole runtime authority.
• Search snippets are discovery-only. Only successful, non-empty extracted
  content may enter the evidence path.
• Structured market data does not become Claim Ledger evidence and does not
  prove that an event caused a price move.
• Embedded workbook charts are display-only; they are not evidence.
• AI Second Opinion and approved guidance are advisory. They do not change
  Gates, finalization, delivery, repair, or Core next-action authority.
• Provider calls never auto-retry, and exact replay never redials.

Honest limits

Solar Stock Periodic, the Codex runtime, Tavily acquisition, and AI Second
Opinion remain Experimental. Live provider usefulness, report coverage,
acquisition-to-finalized_local success rate, and advisory utility are
NOT MEASURED by this release.

The source tag v0.15.3 was created on 2026-08-14. Its first CI run exposed
stale Homebrew formula metadata; the formula was corrected on main, and the
post-fix main workflow passed. The GitHub Release record was published later
from the existing tag without moving or rewriting that tag.

For the complete change history, see the
CHANGELOG.md.

v0.15.2: Merge pull request #564 from Stahl-G/worktree-legacy-json-removal

fix!: remove the legacy JSON control-plane runtime (v0.15.2)

BriefLoop v0.14.0 — Codex SQLite Runtime and Read-Only Review Surfaces

BriefLoop v0.14.0 — Codex SQLite Runtime and Read-Only Review Surfaces

BriefLoop v0.14.0 makes the current product boundary easier to see and operate:
new runs use the Experimental Codex SQLite path, initialization can use a
one-shot local web wizard, and a completed report can be exported as a
self-contained three-page read-only HTML view.

The engineering changes in this release were implemented, tested, and
iteratively repaired with Codex. Human maintainers authorized every merge, the
release tag, and the GitHub Release. Codex did not approve its own work.

What is included

• SQLite ControlStore (briefloop.db), accepted strict requests, Receipts, and
  ledger relations are the sole runtime authority for new runs.
• The packaged Codex Skill follows the Store-derived CoreRunNextAction
  protocol: delegate, deterministic, human_decision, blocked, or
  complete.
• briefloop init <workspace> --web starts a one-shot loopback wizard on the
  same strict bootstrap path as terminal initialization.
• briefloop quality html --workspace <workspace> [--open] writes a
  self-contained three-page read-only view for quality status, optional LAJ
  advisory findings, and improvement status.
• Legacy JSON control-plane modules are deleted. JSON-only workspaces have no
  importer, migration, dual-read, dual-write, or fallback path.

Legacy control files and report/status/handoff/finalize/Quality Panel exports
are non-authoritative projections. Strict action, envelope, and human-request
JSON payloads remain input carriers: the runtime revalidates them against the
Store, and the payloads are never authority by themselves.

Honest product limits

• LAJ remains Experimental and its utility is NOT MEASURED. It is an
  advisory second opinion, not a correctness, completeness, or quality proof.
• The three HTML pages and init wizard are interaction surfaces, not workflow,
  Gate, approval, delivery, or next-action authority.
• The Improvement Ledger lifecycle is unavailable in v0.14.0. The third page
  reports that state; there is no automatic learning or accepted-suggestion
  effect on a later run.
• package_ready is not delivered. Human approval and delivery remain
  explicit recorded actions.

Breaking changes

• fix!: retire the eval-cases command.
• fix!: retire the experiments 080 command family.
• fix!: retire the D1 fold-in keys guidance_manifestation and
  support_wording.
• fix!: retire the Improvement Ledger / Memory file lifecycle; the event
  vocabulary remains available for compatibility and audit classification.
• fix!: make the Quality Panel semantic_support section consistently
  not_available until a Store-native producer exists.
• fix!: retire the v1-RC readiness gate whose runner depended on the
  deleted legacy stack.

Start the Codex path

briefloop init ./my-brief --demo --force
briefloop runtime install --workspace ./my-brief --runtime codex
briefloop run --workspace ./my-brief --runtime codex

Open and trust the workspace in Codex, then follow the installed BriefLoop
Skill. The Skill only executes the exact current Store-derived action and never
falls back to a legacy JSON workspace, operator, or another runtime.

To try the new read-only surfaces:

briefloop init ./my-brief --web
briefloop quality html --workspace ./my-brief --open

The web wizard binds only to loopback. The HTML export contains no action or
write endpoint.

BriefLoop v0.13.0 — SQLite ControlStore Sole Authority And LEGACY-DELETE Tier-1

BriefLoop v0.13.0 — SQLite ControlStore Sole Authority And LEGACY-DELETE Tier-1

v0.13.0 lands the CX cutover: the SQLite ControlStore becomes the sole runtime
authority for BriefLoop workspaces, and the first LEGACY-DELETE batch removes
the retired legacy CLI handler layer. It also carries forward the Python 3.12
floor and the explicit runtime-identity contract.

Highlights

• Makes the SQLite ControlStore the sole runtime authority. JSON-only
  workspaces are classified unsupported — there is no importer, migration,
  dual read/write, or fallback. The Codex runtime host
  (briefloop run --workspace <path> --runtime codex, followed by
  briefloop runtime next, invocation-start, invocation-accept|fail, and
  apply) is the active execution path, verified green on macOS and Windows
  without fail-fast.
• Retires the legacy JSON/operator public commands: they fail closed with
  typed rejections (runtime_command_unsupported /
  legacy_workspace_unsupported) and zero writes. Deterministic source
  acquisition executes only the initialization-frozen provider plan; mutable
  sources.yaml is never re-read after initialization.
• LEGACY-DELETE tier-1 removes the retired handler layer, six
  import-graph-unreachable modules, and three unreferenced scripts. Parser
  registrations stay so the typed rejections keep working. The legacy JSON
  runtime-state stack ships as declared internal debt tracked for
  LEGACY-DELETE-2.
• sources decide is retired by design; source discovery runs through the
  runtime-host route. Finalize, approval, and delivery run as typed Store
  actions through runtime apply.
• Moves requires-python from >=3.9 to >=3.12 and makes runtime identity
  explicit for every new run (hermes, claude, opencode, codex,
  codebuddy, or operator); historical auto / manual manifests remain
  read-only diagnostics.

Boundaries

This release adds no migration path, legacy fallback, dual authority, or
compatibility CLI for retired surfaces. The Codex runtime host is
Experimental; LAJ advisory tooling remains advisory-only with utility and
efficacy NOT MEASURED. The v1.0 RC readiness gate
(docs/v1-pilot-evidence.md) still reports not_satisfied; v1.0-ready
wording remains governed by that gate.

For the complete change list, see CHANGELOG.md.

BriefLoop v0.12.1

BriefLoop v0.12.1 — Delegation Boundaries And Product Governance Reference

v0.12.1 packages two already-merged documentation and adapter-boundary
increments on top of v0.12.0. It clarifies how experimental WorkBuddy /
CodeBuddy hosts divide drafting from deterministic control-plane work, and it
publishes the repo-local Northstar governance Skill plus the bilingual
Architecture Reference v0.4.0 reading edition.

Highlights

• Defines one two-phase WorkBuddy / CodeBuddy path: the exact checked-in role
  agent drafts only its handoff-assigned artifact, then a command-capable main
  session re-reads the handoff and runs the permitted BriefLoop CLI
  transaction.
• Makes missing main-session command capability a hard stop. A generic helper
  or narrative claim is not evidence that the host invoked a BriefLoop role;
  the host-visible invocation and return of the exact checked-in role is the
  delegation evidence boundary.
• Adds the repo-local Northstar product-governance Skill and bounded evals.
  Northstar assesses options and recommends decisions; human product owners
  retain outcome acceptance and effective product decisions.
• Publishes bilingual Markdown and HTML Architecture Reference v0.4.0
  editions, diagrams, reviewer notes, and a deterministic source/render guard.
  The report is explicitly bound to the historical v0.11.12 code snapshot.

Boundaries

This patch release does not add a new persisted authority, schema, writer,
gate, delivery permission, release authority, semantic proof, or completed
first-user Pilot claim.

The WorkBuddy / CodeBuddy surfaces remain experimental and source-clone-only.
Role agents do not receive BriefLoop transaction authority. Current product
architecture and support truth remain in docs/architecture-status.md and
docs/support-matrix.md; the v0.4 report is a historical architecture
reference, not a replacement for those live sources.

For the complete change list, see CHANGELOG.md.

BriefLoop v0.12.0

BriefLoop v0.12.0 — Trusted Control Reads And Cross-Platform Merge Gate

v0.12.0 packages the control-plane hardening line that landed after v0.11.12.
It strengthens how BriefLoop acquires runtime control inputs, interprets
recovery and Registry state, and verifies release candidates across supported
platforms.

Highlights

• Adds descriptor- and handle-bound workspace control reads on POSIX and
  Windows, including no-follow handling for unsafe links and reparse points.
• Binds Recovery's five control inputs — Manifest, Workflow, Registry, Event
  Log, and Finalize Report — to one opened workspace session; safe optional
  Registry and Finalize Report absence remains typed and legal where the
  runtime contract permits it.
• Adds strict runtime-contract loading and a read-only trusted Artifact Registry
  interpreter. Status now exposes Registry-derived values only from a canonical
  interpreter result; unsafe or stale Registry state stays value-free.
• Adds a protected full-platform Merge gate for ready candidates and keeps a
  post-merge supported-platform shadow matrix for main.

Boundaries

This is a control-plane hardening release. It does not make a generated brief
semantic proof, infer delivery from file existence, approve delivery or public
release automatically, or claim a completed first-user Pilot.

The WorkBuddy / CodeBuddy surfaces remain experimental and source-clone-only.
They do not prove that a host delegated role agents unless that host records
the delegation itself. Unmerged consumer work and parser-hardening evidence are
not included in this release claim.

For the complete change list, see CHANGELOG.md.

BriefLoop v0.11.12

BriefLoop v0.11.12 — WorkBuddy Adapter And Semantic Support Auditor

v0.11.12 packages the post-v0.11.9 launch-hardening line for WorkBuddy,
operator-runtime handoff, deterministic first-run polish, and proposal-only
semantic support audit records.

Highlights

• Adds experimental source-clone WorkBuddy Skill packaging, install docs,
  first-use routing, and smoke checklist coverage.
• Promotes operator as the host-agnostic compact runtime name while keeping
  manual as a legacy alias.
• Adds proposal-only Semantic Support Auditor surfaces plus
  semantic-support adjudicate records for human accept/reject decisions.
• Hardens delivery, post-merge control projections, screened-candidate discard
  metadata, public-safety checksum scanning, demo determinism, and Quality Panel
  missing-state display.

Boundaries

This release does not publish WorkBuddy Marketplace support, add a WorkBuddy
runtime adapter, assume WorkBuddy subagents ran, create semantic support truth,
write Claim-Support Matrix rows from semantic proposals, approve delivery,
authorize release, prove output quality, or make pipx install briefloop a
current install path.

For the complete change list, see CHANGELOG.md.

BriefLoop v0.11.9

BriefLoop v0.11.9 — Quality Panel Polish, Launch Docs, And Runtime-State Cleanup

v0.11.9 packages the accumulated post-v0.11.6 cleanup and launch-productization
work into a release line.

Highlights

• Adds the bilingual, CSS-only Quality Panel HTML toggle while keeping
  quality_panel.json as the untranslated machine facts source.
• Keeps the launch install path source-clone based while documenting future
  pipx / PyPI packaging prep.
• Adds the MinerU-derived Markdown bridge for evidence_extract sources when
  adjacent derived Markdown already exists.
• Cleans up stale architecture references, orphaned modules, runtime handoff
  helper ownership, shared workspace test helpers, and runtime-state transaction
  module boundaries.

Boundaries

This release does not publish a PyPI package, make pipx install briefloop a
current install path, run MinerU automatically, parse PDFs by itself, judge
semantic support, prove output quality, approve delivery, or authorize release.

For the complete change list, see CHANGELOG.md.

BriefLoop v0.11.6

BriefLoop v0.11.6 — Product Baseline, Reader Quality, Topology, And Launch Readiness

v0.11.6 ships the accumulated v0.11 product-baseline line on top of the
BriefLoop control spine. It promotes the modest CLI product entries for
industry weekly, management monthly, and document-review workspaces while
keeping wider Product OS, quality, release, and evidence-extraction surfaces
bounded by their support status.

This release does not provide semantic truth proof, output-quality proof,
automatic publication, public-release authorization, legal/compliance advice,
investment advice, IR/disclosure readiness, or a completed Evidence Extraction
Mode.

What Changed

• Promoted the product-facing briefloop new industry-weekly,
  briefloop new management-monthly, and briefloop new document-review
  workspace entries as the supported v0.11 baseline, while keeping canonical
  internal ReportPack ids stable.
• Added product-baseline, public-claim, skill-freshness, launch-smoke, and
  release-checklist guards so release prep fails closed when public docs,
  support status, or source-checkout setup drift.
• Added reader-quality projection surfaces for materiality selection, reader
  template conformance, support-calibrated wording, citation profile split,
  coverage/omission, Quality Panel closeout, and scoped final-abstract warnings.
• Added public-safe reference and evaluation packets, including same-evidence
  reader-quality regression coverage, a Product OS reader-quality reference
  note, and a minimal comparative evaluation packet.
• Hardened role topology guidance so default topology can satisfy Screener
  through Scout without losing distinct candidate_claims.json and
  screened_candidates.json artifacts, while strict topology remains available.
• Added deterministic trajectory-regulation decision narrowing for repeated
  retry, repair-cycle, or blocker loops, limiting exhausted current-stage
  decisions to request_human_review and block_run.
• Added Evidence Extract source locks, logical page inventory seeds, and UTF-8
  text-span seed registries for registered document-review sources.
• Added release/evidence synthetic blocker cases and release-branding readiness
  metadata checks for internal readiness workflows.
• Added launch-facing wording and evidence-link polish, plus
  docs/release-checklist.md for operator release preparation.

Compatibility

This release keeps these compatibility surfaces:

• multi-agent-brief remains the engine CLI.
• briefloop remains the shell CLI alias.
• /briefloop remains the Claude writer command.
• /mabw remains a compatibility Claude command.
• multi_agent_brief remains the Python package/module path.
• multi-agent-brief-workflow remains the Python distribution package name.
• Existing workspace artifacts, MABW experiment IDs, and control files are not
  renamed.

Support Status

Supported for the v0.11 baseline:

• briefloop new industry-weekly
• briefloop new management-monthly
• briefloop new document-review
• baseline ReportSpec / ReportPack contracts for the supported packs

Still experimental unless explicitly stated in the support matrix:

• solar-periodic
• Quality Panel / Summary / static HTML projections
• SourceHub Lite
• internal release approval records
• Evidence Extract source-lock/page/span seed artifacts beyond bounded
  registration
• reader-quality warning/projection surfaces
• synthetic evaluation fixtures and minimal comparative evaluation packets

Boundaries

BriefLoop still does not prove that a source semantically supports every claim,
eliminate hallucinations, replace human review, provide legal/compliance or
investment judgment, or make reports automatically ready to send.

v0.11.6 does not complete issue #156 as a full page-cited Evidence Extraction
Mode. It adds bounded source-lock, logical-page, and text-span seed contracts
that future extraction work can build on.

The public reference and evaluation packets are measurement infrastructure and
traceability evidence. They are not benchmark-win claims, output-quality proof,
or publication approval.

BriefLoop v0.10.7 — Product OS Hardening And Quality Panel

BriefLoop v0.10.7 ships the accumulated experimental Product OS hardening line on top of the existing control spine.

Highlights:

• Product OS contracts and defaults: ReportSpec, ReportPack, ReportTemplate, PolicyProfile, workspace skeletons, deterministic policy resolution, and solar/evidence-extract product packs.
• Source evidence hardening: durable source evidence pack materialization, Evidence Extract source/scope registration, SourceHub Lite setup, source taxonomy normalization, and Source Appendix/audit-trace upgrades.
• Bundle and output hygiene: clean delivery/audit bundle archives, source-link rendering, UTF-8 artifact fallback names, and Quality Panel artifacts included only in audit bundles when valid.
• Internal control surfaces: release-mode approval records with event linkage, Quality Panel JSON projection, hash-bound Markdown/HTML summaries, and briefloop quality summarize.

Boundaries:

• No quality score.
• No semantic truth proof.
• No support-sufficiency gate.
• No delivery approval or public release authority.
• No industry compliance, investment advice, IR/disclosure readiness, hidden crawling, or automatic publication.

Validation:

• Full pytest: 2251 passed.
• Version/release/capability/generated-config/runtime-asset/skill-contract checks passed.
• git diff --check passed.