Sibyl Post-v0.10 Release Remap Spec

• Status: active RC evidence checklist
• Created: 2026-05-17
• Baseline release: v0.10.0, published 2026-05-17
• Baseline head at planning time: v0.10.0-2-g5dc7a398
• Tracking task: 12b1fee4-7bdd-45c8-8a6a-b13fd6eab308
• Parent roadmap: SIBYL_1_0_ROADMAP.md
• RC plan: SIBYL_1_0_RC_PLAN.md
• Current focus: v1.0 RC Evidence Freeze
• RC audit task: 218ca5c5-1920-4689-9ab7-16ac04a73404

1. Decision

Start the next large Sibyl release from the shipped v0.10.0 baseline, not from the old v0.11/v0.12 planning labels.

v0.10.0 absorbed more of the original 1.0 stack than expected:

• native LLM substrate foundations
• trust-control and auth-session gates
• Reflection OS foundations and the reflection-quality gate
• runtime telemetry and overview performance work
• CLI pending writes and idempotent replay
• single-host deployment scaffolding
• broad RBAC and scope-isolation hardening

The old roadmap labels are still useful as design provenance, but they are no longer a good release schedule. The new schedule should be product-shaped:

1. v0.11: real corpus ingestion and source-grounded synthesis at scale
2. v0.12: Memory Workspace and context-quality product hardening
3. v0.13: Surreal-only runtime closure and Graphiti deletion
4. v1.0 RC: evidence freeze, docs/install rehearsal, and boring release gates

Keep v0.14 in reserve only if v0.13 discovers a release-sized distribution or migration gap.

Current state as of 2026-05-18: the v0.11 Corpus Runtime packets, v0.12 gate additions, and v0.13 runtime-closure packets have landed in the task graph. This document remains active as the RC evidence checklist; future-looking packet text below is retained as release provenance.

2. Why Remap Now

The v0.10 release notes describe the largest release since the SurrealDB cutover. The codebase now already contains gates and surfaces that the roadmap previously assigned to v0.11 and v0.12:

• tools/trust/trust_control_gate.py
• tools/trust/auth_session_gate.py
• tools/trust/reflection_quality_gate.py
• tools/trust/overview_perf_gate.py
• tools/trust/synthesis_gate.py
• tools/trust/adapter_ingest_gate.py
• MemorySpace, audit, inspect, synthesis, import, and reflection surfaces across API, CLI, MCP, and web

The next roadmap should avoid redoing shipped work. It should harden the behavior that v0.10 made possible: ingesting private source corpora, synthesizing reliable artifacts, making automatic memory decisions visible, and deleting the remaining Graphiti support only when native behavior has receipts.

3. Release Principles

3.1 Ship Product Slices, Not Internal Layers

Every release after v0.10 should prove an end-to-end product promise. Internal work still matters, but each release needs a user-visible loop:

• capture or import
• policy check
• retrieval or synthesis
• inspect or correction
• audit receipt
• regression gate

3.2 Treat Gates As Product Claims

Gate names should map to claims the release notes can make. A green gate is not just a test bundle; it is a receipt that a public claim is currently supported.

3.3 Keep Humans On Exceptions

The product direction remains automatic-first memory. Humans should see sensitive material, ambiguous contradictions, policy failures, destructive actions, high-impact sharing, and correction flows. Humans should not approve routine safe memory writes one by one.

3.4 Make Graphiti Deletion Earned

Graphiti deletion is a 1.0 gate, not a vibes cleanup. Remove it only after native source ingest, reflection quality, synthesis, context packs, archive restore, and benchmark receipts cover the behavior that used to depend on Graphiti-shaped code.

3.5 Keep Releases Small Enough To Revert

Use atomic feature commits and release packets with clear verification. If a release starts eating multiple product promises, split it.

4. Baseline: What v0.10 Already Owns

v0.10 is the new floor for planning. Treat these as shipped foundations:

• native model/provider substrate and runtime configuration surfaces
• reflection lifecycle records, dream-cycle maintenance jobs, and automatic promotion receipts
• trust-control gate coverage for memory spaces, API-key restrictions, MCP policy, job policy receipts, inspect/audit parity, and UI visibility
• CLI pending write buffer with idempotent replay
• source-grounded synthesis and source import foundations from v0.9
• Memory Workspace routes for captures, imports, source inspect, and synthesis
• runtime telemetry and overview performance receipts
• single-host Docker/Ansible deployment scaffolding

These foundations may need hardening, but they should not be reintroduced as new milestones.

5. Target Schedule

Release	Name	Product Promise	Primary Workstreams
v0.10.x	Patch Rail	Keep the shipped baseline healthy until v0.11 RC1 is tagged.	CI, release workflow, docs drift, urgent trust fixes
v0.11	Corpus Runtime	A real private corpus can be imported, searched, inspected, corrected, and synthesized without leaking scope.	W5, W2, W7
v0.12	Memory Workspace OS	Humans can see what Sibyl did automatically, correct it, undo it, and debug context inclusion from one product surface.	W2, W4, W7
v0.13	Surreal-Only Runtime Closure	The supported runtime no longer depends on Graphiti or Redis-by-default assumptions.	W8, W9
v1.0 RC	Evidence Freeze	The claim surface is frozen, docs/install agree, and every release claim has a current receipt.	All workstreams

6. v0.11: Corpus Runtime

Release Promise

v0.11 makes source-grounded memory useful on real corpora, not just curated fixtures.

A user should be able to import a large private source set, resume interrupted work, search before expensive extraction finishes, inspect skipped/deduped records, correct visibility, synthesize an artifact with explicit source IDs and gaps, remember that artifact with provenance, and prove no private source leaked into project memory.

In Scope

• source adapter contract hardening
• at least one import target beyond the current mailbox/MBOX path, or a deliberately broader mailbox-family adapter if that is the fastest real-corpus win
• persistent import-run observability: imported, skipped, deduped, errored, attachment, checkpoint, and extraction state
• large-corpus import rehearsal using a private-by-default dogfood fixture
• source search before extraction completion
• stronger section-pack materialization and source selection
• unsupported-claim, hidden-source, redaction, and freshness verification in synthesis
• artifact memory lifecycle: generated artifact -> raw source -> graph memory -> correction impact
• Memory Workspace visibility for import and synthesis receipts

Out Of Scope

• full Graphiti dependency deletion
• full arbitrary public sharing
• broad team/org sharing UX
• complete redesign of synthesis generation
• replacing every import or synthesis metadata shape in one migration

Packets

Packet A: Roadmap And Task Reconciliation

Goal: make planning state trustworthy before implementation starts.

Actions:

• close or retag stale v0.8 tasks that shipped through v0.10
• create the v0.11 Corpus Runtime epic
• create child tasks for the packets in this spec
• update active docs so v0.10 is the baseline and v0.11 is the current implementation focus
• keep historical v0.8-v0.12 docs as receipts, not live release promises

Verify:

• sibyl context shows one active v0.11 epic instead of stale v0.8 blockers
• docs/architecture/SIBYL_1_0_ROADMAP.md points at this remap spec
• git status only shows intended docs/task metadata changes

Packet B: Source Adapter Scale

Goal: make import adapters durable enough for more than one curated path.

Likely files:

• packages/python/sibyl-core/src/sibyl_core/services/source_adapters.py
• packages/python/sibyl-core/src/sibyl_core/models/sources.py
• packages/python/sibyl-core/src/sibyl_core/services/mailbox_adapter.py
• apps/api/src/sibyl/jobs/source_imports.py
• apps/api/src/sibyl/api/routes/crawler.py
• apps/web/src/components/memory/source-import-progress.tsx

Actions:

• clarify stable source identity, adapter version, source version, dedupe key, and transform version
• make skipped records first-class enough to inspect and count
• make attachment and metadata-only records visible without forcing extraction
• add one broader source path or broaden mailbox import to cover the next real corpus
• keep imported records private by default unless policy says otherwise

Verify:

• moon run adapter-ingest-gate
• adapter contract tests cover stable IDs, dedupe, private scope, checkpoint resume, and skipped records
• a fixture can pause/resume and preserve counts without duplicating raw memory

Packet C: Large-Corpus Import Rehearsal

Goal: prove the import loop works under realistic volume and partial failure.

Actions:

• create a dogfood import fixture with private, project, duplicate, skipped, attachment, and error cases
• run import in bounded batches with resumable checkpoints
• expose run receipts in API, CLI, and web
• ensure records become searchable before expensive extraction finishes when metadata/raw content is available
• document the rehearsal command and artifact location
• register moon run large-corpus-rehearsal or make adapter-ingest-gate invoke the same focused rehearsal, with fixtures under packages/python/sibyl-core/tests/fixtures/large_corpus/

Current command and artifact contract:

• command: moon run large-corpus-rehearsal
• fixture: packages/python/sibyl-core/tests/fixtures/large_corpus/dogfood.json
• artifact: .moon/cache/large-corpus-rehearsal/receipt.json

Verify:

• new large-corpus rehearsal task or script exits nonzero on count drift, duplicate drift, policy leak, or unrecoverable resume failure
• the rehearsal command and artifact path are stable enough to cite in release notes
• moon run adapter-ingest-gate includes or invokes the rehearsal in focused form
• Memory Workspace shows imported, skipped, deduped, errored, and pending-extraction states

Packet D: Synthesis Verification Depth

Goal: make generated artifacts trustworthy when source material is messy.

Likely files:

• packages/python/sibyl-core/src/sibyl_core/services/synthesis.py
• packages/python/sibyl-core/src/sibyl_core/tools/synthesis.py
• apps/api/src/sibyl/api/routes/synthesis.py
• apps/cli/src/sibyl_cli/main.py
• apps/web/src/components/memory/synthesis-runner.tsx
• apps/web/src/components/memory/synthesis-verification-panel.tsx

Actions:

• strengthen unsupported-claim detection and gap reporting
• preserve section-level source IDs and hidden-source/redaction reasons
• track freshness and supersession signals from reflection findings
• require artifact provenance when remembering synthesis output
• make correction/redaction affect future synthesis selection

Verify:

• moon run synthesis-gate
• hidden-source and unsupported-claim fixtures fail with stable reasons
• remembered synthesis artifacts carry source IDs, verification status, and correction impact

Packet E: Corpus-To-Artifact Product Loop

Goal: wire import, inspect, correction, synthesis, and remember into one usable product loop.

Actions:

• add workspace affordances for source import receipts and synthesis run receipts
• link synthesis sections back to source inspect views
• show correction impact on recall and synthesis
• make CLI vocabulary match the web surface for import status, synthesis verify, artifact remember, and memory inspect

Verify:

• a browser/workspace trust-flow check covers import -> inspect -> correct -> synthesize -> remember
• CLI commands can produce the same receipt IDs shown in web
• correction tests prove recall and synthesis behavior changes after hiding, redacting, or marking stale

v0.11 Release Gate

Minimum release-candidate commands:

moon run adapter-ingest-gate
moon run large-corpus-rehearsal
moon run synthesis-gate
moon run autonomy-gate
moon run reflection-quality-gate
moon run trust-control-gate
moon run auth-session-gate
moon run memory-trust-gate
moon run overview-perf-gate
moon run docs:lint
moon run docs:build
moon run :check

Release notes must cite:

• large-corpus rehearsal artifact
• adapter ingest gate receipt
• synthesis gate receipt
• hidden-source and unsupported-claim fixture receipts
• workspace trust-flow check receipt

7. v0.12: Memory Workspace OS

Release Promise

v0.12 makes the web workspace and CLI feel like one coherent memory operating system.

A human should be able to answer:

• what did Sibyl remember or import?
• why did it promote, hide, dedupe, mark stale, or route an exception?
• who or what can see it?
• which source proves it?
• how do I correct or undo it?
• why did it appear in this context pack?

In Scope

• exception console for automatic memory decisions
• context-quality gate for wake, recall, and deep_search
• workspace-trust gate for browser-visible trust flows
• product-grade inspect, correction, restore, and promotion preview flows
• context-pack explanation: inclusion, exclusion, redaction, freshness, confidence, source IDs
• CLI simplification around the same nouns as the web UI
• realtime activity that is useful without creating idle query storms

Out Of Scope

• public sharing as a social feature
• arbitrary policy language
• full Graphiti deletion unless v0.11 unexpectedly clears every dependency

Required New Gates

• context-quality-gate
• workspace-trust-gate

The gates should become Moon tasks before v0.12 implementation is considered feature-complete.

Release Gate

moon run context-quality-gate
moon run workspace-trust-gate
moon run memory-trust-gate
moon run autonomy-gate
moon run reflection-quality-gate
moon run trust-control-gate
moon run auth-session-gate
moon run synthesis-gate
moon run overview-perf-gate
moon run :check

8. v0.13: Surreal-Only Runtime Closure

Release Promise

v0.13 removes the remaining supported runtime dependence on Graphiti-shaped code and Redis-required single-machine assumptions.

A fresh install should run with SurrealDB as the only required data service. Legacy Graphiti-shaped archives should remain readable through Sibyl-owned import/projection code that does not import the Graphiti Core module.

In Scope

• delete Graphiti Core from package metadata, default dependency groups, optional extras, and CI paths
• replace remaining Graphiti-shaped entity, relationship, search, embedder, extraction, and restore adapters with Sibyl-native services or explicit migration readers
• replace Graphiti compatibility tests with native archive/import regression tests
• turn no-Graphiti smoke from default-loop proof into supported-runtime proof
• complete Redis-optional coordination boundary for local installs
• backup/restore round-trip for auth, graph, content, raw memory, tasks, settings, and source imports
• Docker, Helm, Homebrew, quickstart, and docs alignment to remove Graphiti from default install paths and prove the Surreal-only runtime

Out Of Scope

• high-throughput distributed swarm runtime
• Gradial Rust runtime decisions
• cloud-only deployment assumptions

Required Packets

Packet A: Graphiti Deletion Inventory

Goal: make the deletion surface explicit before removing compatibility code.

Verify:

• moon run inventory-check
• moon run inventory-typecheck
• moon run inventory-test
• rg "graphiti[_-]core" only returns accepted historical or migration-format labels

Packet B: Backup/Restore Release Gate

Goal: turn the backup/restore round-trip into an executable release gate before runtime deletion work lands.

Actions:

• add moon run backup-restore-gate
• cover auth, graph, content, raw memory, tasks, settings, source import runs, and synthesis artifacts
• write the round-trip receipt to .moon/cache/backup-restore-gate/receipt.json

Verify:

• moon run backup-restore-gate
• restored data preserves policy scope, source IDs, task links, and synthesis provenance

Packet C: Redis-Optional Local Runtime Closure

Goal: finish the local coordination boundary for single-machine installs.

Verify:

• local mode can enqueue crawl, entity creation, task update, backup, and reflection maintenance work without Redis
• Redis mode remains opt-in for distributed deployments

Release Gate

moon run inventory-check
moon run inventory-typecheck
moon run inventory-test
moon run memory-trust-gate
moon run trust-control-gate
moon run reflection-quality-gate
moon run auth-session-gate
moon run autonomy-gate
moon run overview-perf-gate
moon run adapter-ingest-gate
moon run synthesis-gate
moon run backup-restore-gate
moon run :check

Additional deletion receipts:

• rg "graphiti[_-]core" returns only historical docs, archived benchmark notes, or explicit migration-format labels
• no default install, local dev, Docker, Helm, CI, or test task installs Graphiti
• backup/restore round-trip artifact is attached to release notes
• fresh quickstart works with only SurrealDB as required infrastructure

9. v1.0 RC: Evidence Freeze

Release Promise

v1.0 ships only after the boring receipts are current.

No new large product promises land in the RC. The work is claim freeze, docs/install rehearsal, evidence refresh, and fixing the smallest remaining blockers.

In Scope

• active architecture docs agree
• stale release docs are archived or labeled historical
• benchmark ledger current and claim-safe
• install, local dev, Docker, Helm, Homebrew, and docs tell one story
• auth, retrieval, memory trust, source ingest, synthesis, overview performance, product flows, and Graphiti deletion all have current receipts
• the versioning and release-note contract is clear before the explicit release cut

RC Completion Checklist

The RC is complete only when each claim has fresh evidence from the current checkout.

Claim	Evidence
Active docs agree on current focus	docs:lint, docs:build, this spec, the 1.0 roadmap, and the Northstar all point at v1.0 RC focus
Task graph reflects current work	sibyl context, sibyl epic show epic_19e1dea67ebf, and task lists show RC work instead of ghosts
Source ingest and corpus scale are current	moon run adapter-ingest-gate and moon run large-corpus-rehearsal
Synthesis is source-grounded	moon run synthesis-gate
Automatic memory remains policy-safe	moon run autonomy-gate, moon run memory-trust-gate, and moon run trust-control-gate
Auth/session behavior is boring	moon run auth-session-gate
Reflection quality is current	moon run reflection-quality-gate
Context and workspace trust are current	moon run context-quality-gate and moon run workspace-trust-gate
Overview performance has a receipt	moon run overview-perf-gate
Surreal-only runtime closure holds	moon run inventory-check, moon run inventory-typecheck, moon run inventory-test, and grep audit
Backup/restore is release-gated	moon run backup-restore-gate
Benchmark ledger is claim-safe	moon run bench-gate
Install and docs surfaces build	moon run docs:lint docs:build and moon run :check
Release cut has an explicit boundary	VERSION remains the package version source; prerelease cuts validate a pre-committed candidate SHA, same-SHA nightly run, RC gate bundle, and release notes after go-ahead

RC grep audit:

rg -n "graphiti[_-]core" apps packages tools docs \
  --glob '!docs/_archive/**' \
  --glob '!docs/architecture/**' \
  --glob '!contexts/**'

Expected result: no supported runtime, test helper, package metadata, install doc, or active guide contains a Graphiti Core dependency or import-module reference. Historical archive docs may still contain exact legacy names as provenance.

Release Decision Format

Ship v1.0:
  yes/no
Blocking packet:
  smallest remaining blocker, if no
Proof command:
  command or external receipt that will prove the blocker closed
Residual risk:
  accepted risks with owners

10. Task Graph Reconciliation

The project task graph contained stale v0.8-era tasks after v0.10. Packet A reconciled them so recall no longer injects obsolete blockers.

Actions:

• shipped v0.8/v0.10 tasks are done with release receipts
• obsolete v0.8 epics are archived or marked complete
• Graphiti-deletion work is retagged under v0.13
• v0.11 Corpus Runtime packet tasks were created and completed
• v0.12, v0.13, and v1.0 RC epics exist for current planning

Verification:

sibyl context
sibyl epic show epic_19e1dea67ebf
sibyl task list --status doing
sibyl task list --status todo

The output should show v1.0 RC evidence work and any explicitly active v0.13 closure task, not stale v0.8 release ghosts.

11. Gate Matrix

Gate	Current Status	Release Role
memory-trust-gate	exists	Always-on regression gate for memory policy and leak prevention
trust-control-gate	exists	Always-on regression gate for memory spaces, MCP, jobs, CLI, and web trust
auth-session-gate	exists	Required for v0.12+ and any auth/session change
autonomy-gate	exists	Required when automatic promotion or exception routing changes
reflection-quality-gate	exists	Required for v0.11+ because synthesis depends on lifecycle signals
overview-perf-gate	exists	Required for product releases touching overview, activity, or realtime
adapter-ingest-gate	exists	Primary v0.11 gate
synthesis-gate	exists	Primary v0.11 gate
large-corpus-rehearsal	exists	v0.11 dogfood volume, resume, search, dedupe, and policy receipt
context-quality-gate	exists	v0.12 context-pack quality receipt
workspace-trust-gate	exists	v0.12 browser-visible trust-flow receipt
no-Graphiti supported-runtime gate	exists	v0.13 supported-runtime proof via no-Graphiti smoke and inventory gates
backup-restore-gate	exists	Required v0.13 round-trip proof for archive restore scope and provenance

12. Risk Register

Risk	Impact	Mitigation
v0.11 becomes another mega-release	Hard to verify and release	Keep Graphiti deletion and full workspace OS out of scope
Large-corpus rehearsal exposes import model gaps	Release slips	Treat rehearsal as Packet C and fix the smallest contract gap first
Synthesis quality depends on weak reflection signals	Unsupported claims or stale artifacts	Use reflection-quality gate as a v0.11 regression gate
Workspace UI becomes a manual queue	Product regresses to review treadmill	Optimize for exception routing, receipts, and correction flows
Graphiti deletion starts too early	Native behavior loses quality	Do deletion after v0.11/v0.12 receipts, not during source scale
Task graph remains stale	Agents plan from old blockers	Reconcile tasks before implementation and after each release

13. Open Questions

These are decisions for v0.11 kickoff, not blockers for accepting this remap:

1. Which real corpus should drive the large-corpus rehearsal first: Gmail Takeout, Maildir, Apple Mail export, agent transcripts, or a mixed local document archive?
2. Should v0.11 add a second adapter, or deepen mailbox import until it covers the highest-value private corpus?
3. Which synthesis artifact should be the dogfood hero: release notes, architecture docs, onboarding guide, decision log, or audit packet?
4. Should context-quality gate start in v0.11 as a fixture-only spike, then become a release gate in v0.12?
5. Which v0.13 packet should land first: Graphiti deletion inventory, backup/restore gate, or Redis-optional local runtime closure?

14. Recommendation

Make v0.11 Corpus Runtime.

This is the cleanest next release because it builds directly on what v0.10 shipped without reopening settled trust and reflection work. It proves Sibyl can handle the kind of messy, source-heavy memory that makes a second brain valuable, while generating the evidence needed for workspace polish and eventual Graphiti deletion.