From ae6b1c7564b00c854170ceececfd1fe89d045a5a Mon Sep 17 00:00:00 2001 From: Edwin Amirian Date: Sun, 9 Aug 2026 18:24:59 -0700 Subject: [PATCH] docs close governed cognition milestone --- ROADMAP.md | 95 ++++++++------ docs/evidence/README.md | 16 ++- .../gc1-public-external-consumer-v1.md | 122 ++++++++++++++++++ tests/test_evidence_index_integrity.py | 22 +++- tests/test_public_roadmap_positioning.py | 38 +++--- 5 files changed, 219 insertions(+), 74 deletions(-) create mode 100644 docs/evidence/gc1-public-external-consumer-v1.md diff --git a/ROADMAP.md b/ROADMAP.md index 658b976..0f12772 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -71,19 +71,24 @@ Roadmap outcome states are used strictly: ## Current release checkpoint -`ace-core` 0.4.2 is published on PyPI and GitHub. It preserves the governed-Intelligence +`ace-core` 0.4.4 is published on PyPI and GitHub. It preserves the governed-Intelligence foundation and public P1/P2 evidence carried by 0.4.1—Core + Intelligence in one install, governed LIVE source ingress, the LIVE bridge, Brief synthesis, the inert JSON Domain Pack -compiler, exact replay, packaged external-consumer conformance seams, schema head v175, supported -CLI identities, and exactly eleven thin MCP tools—while adding the public `ace cognition` builder -workflow over the existing authenticated cognition boundary. - -The [0.4.2 GitHub Release](https://github.com/augmented-cognition-engine/core/releases/tag/v0.4.2) -and public `ace-core==0.4.2` package expose proposal teaching, semantic-diff inspection, authorized -human review, materially attributed fresh use, exact receipt inspection, and lifecycle governance. -The [GC1 public builder-surface evidence](docs/evidence/gc1-public-builder-surface-v1.md) binds the -tag, trusted publication, artifact hashes, and fresh public-index smoke test. GI1 remains the -packaging, contract, and publication outcome first passed by 0.4.0. +compiler, exact replay, packaged external-consumer conformance seams, supported CLI identities, +and exactly eleven thin MCP tools—while exposing the public `ace cognition` builder workflow over +the existing authenticated cognition boundary. Patch releases 0.4.3 and 0.4.4 hardened canonical +cognition persistence and task-source timestamp compatibility; schema head is v176. + +The [0.4.4 GitHub Release](https://github.com/augmented-cognition-engine/core/releases/tag/v0.4.4) +and public `ace-core==0.4.4` package expose proposal teaching, semantic-diff inspection, authorized +human review, materially attributed fresh use, exact receipt inspection, lifecycle governance, and +the canonical persistence fixes required by a real external consumer. The +[GC1 public external-consumer evidence](docs/evidence/gc1-public-external-consumer-v1.md) binds the +tag, trusted publication, artifact hashes, clean public-index runtime, independent Market journey, +restart durability, failure controls, and released conformance coverage. The earlier +[builder-surface evidence](docs/evidence/gc1-public-builder-surface-v1.md) remains the exact +point-in-time receipt for 0.4.2. GI1 remains the packaging, contract, and publication outcome first +passed by 0.4.0. The [0.4.1 GitHub Release](https://github.com/augmented-cognition-engine/core/releases/tag/v0.4.1) and public `ace-core==0.4.1` package remain the exact Core identity used by the completed GI2 @@ -98,10 +103,12 @@ compiles and activates both inert packs through unchanged platform APIs, retires the World binding remains active and byte-identical. The exact receipts and limitations are in the [GI2 public cross-domain evidence](docs/evidence/gi2-public-cross-domain-falsification-v1.md). -This release does not by itself close the broader governed-cognition builder-experience gate in -issue #3. The supported public surface is now proven, but the external-consumer journey must still -pass proposal, human disposition, exact revision use, restart persistence, retirement failure, -and attribution against a running deployment. **GC1** therefore stays active in 0.4.x. +The broader governed-cognition builder-experience gate is now passed. An independent Market +Intelligence consumer reproduced proposal, human disposition, exact revision use, restart +persistence, retirement failure, and attribution against a running public-package deployment. +Released Core conformance tests cover rejection, revision, rollback, expiry, unavailable +dependencies, effectiveness classification, and non-selectable revision/retirement proposals. +**GC1** is therefore passed in 0.4.x. The release remains bounded to the documented single-node topology and trusted in-process Python extensions. It does not establish hostile-code isolation, distributed operation, general @@ -134,8 +141,8 @@ Release targets describe sequence, not calendar commitments. ACE follows three r |---|---|---|---|---| | 0.2.x | State Engine stabilization | **Maintenance** | Inherits the passed R0, R1, R2, R3, R4, R5, R6, R7, G1, IA-R1, I1, I2, I3, F1, L1, K1, K2, and K3 foundation; patch work is limited to compatibility, migration, recovery, observability, reliability, security, and documentation hardening | [#1](https://github.com/augmented-cognition-engine/core/issues/1) | | 0.3.x | Productized State | **Maintenance** | PS1 passed in 0.3.1 through the K1–K3 spine, R1/R4 onboarding pattern, G1 and IA-R1 inspection, I1–I3 receipts, and E1 extension/governance boundary; later patches are compatible hardening only | [#2](https://github.com/augmented-cognition-engine/core/issues/2) | -| 0.4.x | Governed Cognition | **Active** | GI1 passed in 0.4.0 with governed LIVE ingress, the Intelligence bridge, Brief synthesis, Domain Pack compiler, exact replay, and external conformance seams; GC1 remains the supported teach, inspect, approve, use, measure, revise, and retire journey | [#3](https://github.com/augmented-cognition-engine/core/issues/3) | -| 0.5.0 | Reasoning into Action | **Next** | Advances T1 and B1 from approved decision to bounded attributable action after GC1 passes; uses I1 authority receipts and begins the execution-adapter slice of E2 | [#37](https://github.com/augmented-cognition-engine/core/issues/37) | +| 0.4.x | Governed Cognition | **Passed** | GI1 passed in 0.4.0; GC1 passed against public 0.4.4 through the supported teach, inspect, approve, use, measure, revise, rollback, and retire lifecycle, an independent Market consumer, restart durability, attribution, and failure controls | [#3](https://github.com/augmented-cognition-engine/core/issues/3) | +| 0.5.0 | Reasoning into Action | **Next** | Advances T1 and B1 from approved decision to bounded attributable action; uses I1 authority receipts and begins the execution-adapter slice of E2 | [#37](https://github.com/augmented-cognition-engine/core/issues/37) | | 0.6.0 | Measured Intelligence | **Later** | Productizes the bounded L1 evidence loop, connects I3 material use to product-owned outcomes, advances the SI4 orientation/attention evaluation slice, and advances F2 only where evidence or demonstrated user need justifies a broader consequence contract | [#38](https://github.com/augmented-cognition-engine/core/issues/38) | | 0.7.0 | Domain and Extension Platform | **Later** | Completes the stable third-party platform promise across GI1, E1, and E2 and advances SI3: pack schemas, SDKs, conformance, compatibility, permissions, isolation, heterogeneous evidence sources, telemetry, adapters, and lifecycle policy | [#39](https://github.com/augmented-cognition-engine/core/issues/39) | | 0.8.0 | Reasoning Workspace | **Later** | Builds on G1 and IA-R1 to expose the I1–I3, E1, B1, L1, and SI1–SI4 lifecycle through a coherent permission-aware human experience | [#40](https://github.com/augmented-cognition-engine/core/issues/40) | @@ -155,9 +162,11 @@ decision cadences. A failure found by either domain returns to Core only as a do contract, compiler, conformance, governance, or runtime requirement. A domain-specific workaround does not become platform code. -This work proceeds in parallel with GC1 and preparation for 0.5.0. A domain demonstration does not -promote an ACE capability by itself; promotion still requires reproducible public evidence, -declared limitations, and reconciliation in this roadmap. +This work proceeds in parallel with preparation for 0.5.0. A domain demonstration does not promote +an ACE capability by itself; promotion still requires reproducible public evidence, declared +limitations, and reconciliation in this roadmap. The Market GC1 journey contributes to the passed +Core outcome only because it is composed with the public builder surface and released Core +conformance evidence. ### 0.2.x — State Engine stabilization @@ -210,7 +219,7 @@ or retire. Signal, attention disposition, and governed Brief resources with provenance and exact replay. - Preserve the eleven-tool MCP surface, naked-kernel startup, schema v175, and single-node limits. -**Active in 0.4.x — GC1:** +**Passed in 0.4.x — GC1:** - Make the canonical E1 teaching lifecycle obvious through supported public interfaces. - Create inspectable proposals from authorized tasks, corrections, conversations, and documents. @@ -219,17 +228,17 @@ or retire. - Support rejection, rollback, supersession, conflict, expiry, disablement, and retirement without erasing history or enabling silent self-modification. -The public `ace-core==0.4.2` release adds a thin `ace cognition` builder workflow over the existing -authenticated HTTP boundary. It makes proposal, semantic-diff inspection, authorized review, -materially attributed fresh use, receipt inspection, and lifecycle governance explicit without -adding authority or a new cognition model. Its public builder surface is now verified from a fresh -PyPI install. This does not close GC1: the external-consumer, restart, failure-control, and -running-deployment acceptance journey remains the outcome gate. +The public `ace-core==0.4.2` release added a thin `ace cognition` builder workflow over the existing +authenticated HTTP boundary. Releases 0.4.3 and 0.4.4 hardened canonical persistence and +task-source timestamp compatibility discovered by the external-consumer journey. Public 0.4.4 now +passes the complete supported gate: an independent consumer taught a reusable capability through +public interfaces, a human inspected and governed the change, fresh invocations materially used +the exact approved revision with complete attribution before and after restart, and retirement +failed a distinct later required use closed. Released conformance tests cover the remaining +rejection, revision, rollback, expiry, unavailable-dependency, and effectiveness branches. -Release gate: GI1 is passed. GC1 passes when an external consumer can teach ACE a reusable -capability through public interfaces, a human can inspect and govern the change, and a fresh -invocation materially uses the exact approved revision with complete attribution. Until then, the -0.4.x milestone remains active and 0.5.0 remains next. +Release gate: **passed.** GI1 and GC1 are passed, the 0.4.x milestone is complete, and 0.5.0 remains +next under its own T1 and B1 gates. ### 0.5.0 — Reasoning into Action @@ -474,8 +483,8 @@ flowchart LR |---|---|---|---|---| | 1. Reasoning foundation | Can ACE reliably reason, remember, and explain what happened? | **Passed** | R0–R7, G1, IA-R1, I1–I3, and F1 passed | through 0.2.x | | 2. Product state | Can a product maintain grounded state and reason about change and consequences? | **Passed** | K1–K3, E1, and PS1 passed; Productized State is published in 0.3.1 | 0.3.x | -| 3. Governed cognition and intelligence | Can a builder run domain-neutral Intelligence and teach the product how to reason without silent self-modification? | **Active** | GI1 passed in 0.4.0; GC1 remains the supported teaching journey | 0.4.x | -| 4. Reasoning into action | Can an approved decision safely produce attributable work? | **Next; gated by GC1** | T1 and B1 not ready; I1 authority foundation passed | 0.5.0 | +| 3. Governed cognition and intelligence | Can a builder run domain-neutral Intelligence and teach the product how to reason without silent self-modification? | **Passed** | GI1 passed in 0.4.0; GC1 passed against public 0.4.4 and an independent Market consumer | 0.4.x | +| 4. Reasoning into action | Can an approved decision safely produce attributable work? | **Next** | GC1 and I1 are passed; T1 and B1 remain not ready | 0.5.0 | | 5. Measured intelligence | Can ACE prove when retained intelligence or a capability helped, hurt, or remains unproven? | **Bounded evidence gate passed; productization next** | L1 and I3 passed; F2 not ready | 0.6.0 | | 6. Domain and extension ecosystem | Can any product adopt, specialize, operate, and retain ownership of ACE intelligence? | **Pack foundation passed; ecosystem gated** | GI1 and E1 passed; E2 and H1 not ready | 0.7.0 and 0.9.0 | | 7. Human experience | Can people inspect and govern the full loop without learning ACE internals? | **Read-only and governance foundations passed** | IA-R1, G1, E1, and L1 passed; the writable workspace remains bounded by B1 and H1 | 0.8.0 | @@ -614,20 +623,20 @@ Both predate the P1/P2 sequence above and are superseded by it. These point-in-time records did not move an outcome by themselves. Their former publication gate has now passed: Core 0.4.1, World 0.8.0, and Market 0.6.0 are public, the tagged World conformance suite passes, and the clean public-index two-domain activation and retirement-isolation journey -passes. GI2 is therefore reconciled to `passed`; GI1 remains passed, GC1 remains active, and -SI1–SI4 remain bounded future outcomes. +passes. GI2 is therefore reconciled to `passed`; GI1 and GC1 are passed, while SI1–SI4 remain +bounded future outcomes. -### 1. Complete the governed-cognition builder journey +### 1. Completed governed-cognition builder journey E1 passed for ace-core 0.3.0 and establishes one canonical cognition model. GI1 passed in 0.4.0 and -establishes the governed Intelligence substrate. GC1 now makes the cognition lifecycle an obvious +establishes the governed Intelligence substrate. GC1 makes the cognition lifecycle an obvious supported public experience: ```text teach → propose → inspect → approve → use → measure → revise or retire ``` -The dependency sequence is: +The accepted dependency sequence was: 1. preserve the converged recipe and legacy `Skill`/`Job`/`Phase` migration behavior; 2. create inspectable learning proposals from tasks, corrections, conversations, and documents; @@ -637,9 +646,11 @@ The dependency sequence is: 5. discover and load only relevant approved cognition within explicit context and cost budgets; 6. record the exact cognition revisions considered, selected, used, omitted, or unavailable. -Exit condition: an extension can teach ACE a reusable reasoning capability, a human can govern the -change, and a fresh invocation can materially use the approved revision without widening the thin -MCP contract. +Exit condition: **passed.** The public 0.4.4 builder and independent Market consumer journeys prove +that an extension can teach ACE a reusable reasoning capability, a human can govern the change, +and fresh invocations can materially use the exact approved revision across restart without +widening the thin MCP contract. The exact receipts and composed negative-path coverage are in the +[GC1 public external-consumer evidence](docs/evidence/gc1-public-external-consumer-v1.md). ### 2. Strengthen the runtime before writable action @@ -731,7 +742,7 @@ general-intelligence claim. | PS1 | passed | Make the passed State Engine an obvious supported extension-first product journey | [Productized State evidence](docs/evidence/productized-state-journey-v1.md) passes authenticated adapter discovery/ingestion, `ace state` orchestration, integrated G1/IA-R1 and I1–I3 receipt inspection, v171 schema-zero, v168→v171 upgrade, restarts, correction, and later material use; the [0.3.1 release evidence](docs/evidence/productized-state-v0.3.1-release-readiness.md) binds the exact tag, official CI, trusted PyPI publication, matching registry hashes, and clean public-index installation | | GI1 | passed | Ship the governed Intelligence foundation as one installable distribution with inert Domain Pack compilation and governed LIVE seams | [ace-core 0.4.0](https://github.com/augmented-cognition-engine/core/releases/tag/v0.4.0) packages Core + Intelligence, the inert Domain Pack compiler, LIVE source ingress, the LIVE bridge, Brief synthesis, exact replay, external conformance seams, schema v175, unchanged eleven-tool MCP, trusted PyPI publication, and clean public-index installation. This is a packaging, contract, and publication outcome; the domain-neutrality claim it previously carried moved to GI2 | | GI2 | passed | Prove the Intelligence foundation is domain-neutral through an independent, publicly reproducible second-domain falsification | [GI2 public cross-domain evidence](docs/evidence/gi2-public-cross-domain-falsification-v1.md) binds public Core 0.4.1, World 0.8.0, and Market 0.6.0 artifacts. The tagged World suite passes 81 tests and reproduces `case:412426eee708d56f6bda931ccf9e5d8b` and `brief:25d8232c9bfa27050bdcb160fb75f06c`; a clean public-index install compiles and activates both packs through unchanged APIs, then retires Market without changing the active World binding. This is domain-neutral substrate evidence, not SI1–SI4 completion or a claim of general real-world accuracy | -| GC1 | active | Make governed cognition an obvious supported teach, inspect, approve, use, measure, revise, and retire journey | E1 and GI1 are passed; issue [#3](https://github.com/augmented-cognition-engine/core/issues/3) remains open until one public external-consumer journey passes proposal, human disposition, exact revision use, failure controls, restart, and attribution through supported interfaces | +| GC1 | passed | Make governed cognition an obvious supported teach, inspect, approve, use, measure, revise, and retire journey | [GC1 public external-consumer evidence](docs/evidence/gc1-public-external-consumer-v1.md) composes the public 0.4.4 artifact identity, independent Market prepare/restart/resume/retire journey, exact material-use attribution, fail-closed post-retirement request, and released Core coverage for rejection, revision, rollback, expiry, unavailable dependencies, and effectiveness | | T1 | not ready | Strengthen task recovery, replay, portability, cancellation, and resource reporting | [TP1 evidence](docs/evidence/state-engine-tp1-reliable-memory-lifecycle-v1.md) closes reliable observation claiming/recovery; cancellation, portability, reporting, and topology guarantees remain | | E1 | passed | Stabilize the extension and governed-cognition boundary | [E1 release evidence](docs/evidence/e1-governed-cognition-release-v1.md) binds the canonical E1-A–G lifecycle to ace-core 0.3.0, exact current/N-1 and mixed-package evidence, public artifacts and fresh install, deployment inventory, independent AI security acceptance, and release-owner countersignature; the pass is limited to trusted in-process extensions and is not a human penetration test or certification | | L1 | passed | Use resolved conditional forecasts to improve later reasoning and decision quality | [L1 evidence](docs/evidence/l1-foresight-impact-evidence.md) preserves the negative retrospective probe and v5/v6 failures, then records the preregistered v7 correction replicate: all 192 cases and 48 clusters were eligible, and ACE passed persistence, naïve/base-rate, and matched model-only under the frozen all-controls interval rule; the claim is limited to the executable benchmark | diff --git a/docs/evidence/README.md b/docs/evidence/README.md index 2a0c44c..4a7fb9f 100644 --- a/docs/evidence/README.md +++ b/docs/evidence/README.md @@ -48,14 +48,18 @@ superseded by later work and is kept for audit trail only. - [State Engine K1-K3 extension-first product closeout](state-engine-k1-k3-product-journey-v1.md) - [Productized State integrated public journey](productized-state-journey-v1.md) -## Governed cognition builder surface +## Governed cognition -The [GC1 public builder-surface record](gc1-public-builder-surface-v1.md) binds the public -`ace-core==0.4.2` tag, trusted publication, artifact hashes, and fresh public-index CLI smoke test. -It proves the supported builder surface, not the remaining external-consumer, restart, and -failure-control journey; GC1 therefore remains active. +The [GC1 public external-consumer closeout](gc1-public-external-consumer-v1.md) composes the public +builder surface, the independent Market Intelligence prepare/restart/resume/retire journey, and +the released Core conformance coverage for negative lifecycle and effectiveness branches. GC1 is +therefore **passed**. -- [GC1 public governed-cognition builder surface](gc1-public-builder-surface-v1.md) — public surface passed; GC1 active +The earlier builder-surface record remains an immutable point-in-time receipt. Its statement that +GC1 was still active was true before the external-consumer gate passed and is now historical. + +- [GC1 public external-consumer closeout](gc1-public-external-consumer-v1.md) — public, passed +- [GC1 public governed-cognition builder surface](gc1-public-builder-surface-v1.md) — public surface passed; historical pre-closeout state ## Governed Intelligence platform substrate diff --git a/docs/evidence/gc1-public-external-consumer-v1.md b/docs/evidence/gc1-public-external-consumer-v1.md new file mode 100644 index 0000000..069cff1 --- /dev/null +++ b/docs/evidence/gc1-public-external-consumer-v1.md @@ -0,0 +1,122 @@ +# GC1 — public governed-cognition external-consumer closeout (v1) + +**Status:** public, passed + +**Date:** 2026-08-09 + +**Outcome:** GC1, supported governed-cognition builder journey + +## Claim + +The public governed-cognition lifecycle is reproducible from published artifacts by an independent +domain consumer. Market Intelligence used only the supported `ace` CLI from a clean +`ace-core==0.4.4` installation to teach a sourced capability, inspect it, approve it through human +authority, materially use the exact approved revision, stop and restart the API over the same +durable database, use the same revision again, retire it, and prove that a distinct later required +use failed closed. + +This record composes that external journey with the public Core builder-surface receipt and the +released Core conformance tests for the lifecycle branches that the Market journey did not repeat. +It does not treat one successful domain demonstration as proof of reasoning quality or general +beneficial impact. + +## Public identities + +| Component | Public identity | Exact source | +|---|---|---| +| Core + Intelligence | `ace-core==0.4.4` | [release `v0.4.4`](https://github.com/augmented-cognition-engine/core/releases/tag/v0.4.4), commit `ca7ee1f1e04c02e43a2db05c3bb6355feb011180` | +| Market Domain Pack | `ace-domain-market-intelligence==0.6.0` | unchanged inert distribution | +| External consumer packet | Market repository main | [pull request #3](https://github.com/augmented-cognition-engine/domain-market-intelligence/pull/3), squash commit `eb596664a2e2736a9b4d97e296c3e59ebcd9b88f` | + +The [trusted publication run](https://github.com/augmented-cognition-engine/core/actions/runs/31345474622) +published the exact Core tag to PyPI. Public artifacts were: + +- `ace_core-0.4.4-py3-none-any.whl`, SHA-256 + `25949e984d68e1f917fc0aef9a123e550618d401ec4be73d2f49f21c77a9245e`; +- `ace_core-0.4.4.tar.gz`, SHA-256 + `85af82368f71abfa4da94a11b37049e760e42da7c30d4c0633a8f9a160b6eff9`. + +A fresh Python 3.12 environment installed 0.4.4 from the public PyPI index. The distribution, +`ace.__version__`, `ace_mcp_client.__version__`, and Core runtime version all reported `0.4.4`. +The database migrated from schema v0 to v176. No Core source checkout or locally built Core wheel +was on the consumer runtime import path. + +## External journey result + +The independent consumer packet is archived in the +[Market repository](https://github.com/augmented-cognition-engine/domain-market-intelligence/blob/main/docs/evidence/gc1-market-external-consumer-v1.md). +Its machine-readable receipt has SHA-256 +`b239b314b34c3dfedddb72907b63abe5bf3197013b8d1460466ff4942f714682`. + +| Boundary | Exact result | +|---|---| +| Source task | `task:9295nllkmh074tv3pr58` | +| Proposal | `cognition_proposal:a1306d345dfd5bc0e9fb7182905ab383`, initially non-selectable | +| Semantic diff | SHA-256 `5211778bfd65a2177894180029898fa80a27374246fb8819959261d46db342b5` | +| Human review | `cognition_review:2dead1ebed5de7e6fea92a553b6c3e5c`, disposition `approve` | +| Approved revision | `cognition_revision:02db82ccd8f2f5776bc49ef88f71ec62` | +| Active head | `cognition_head:dcc2299f646ad9a0a811f8bd4898a1c9`, generation 1 | +| Pre-restart use | task `task:3ohtjhggviqo25arnoax`, material-use hash `919ac46758ee244fbf9e029b79155313d568aa08295e0c0e8097a7ca516872f5` | +| Post-restart use | task `task:ino203d0bq12znb1dlmz`, same required revision and material-use hash | +| Retirement | `cognition_lifecycle:5d673126b4c5e84b46014c038b29fd74`, resulting lifecycle `retired` | +| Distinct later request | failed with `cognition_use_attribution_incomplete` | + +The revision hash +`e04977c31966cef53b85f64bbefa20b871f3223245463eab18f6700642d29730` and active-head hash +`7bb755abaabf53da8933de9981b4c377c79aab34f01087e7ad94a81cabf11f05` were byte-identical before +and after the API restart. Both successful tasks had matching non-empty selection and use revision +sets that included the approved Market revision. + +The failure probe intentionally used a different post-retirement request. ACE replays an identical +task inside its idempotency window; repeating the earlier request would retrieve its valid durable +receipt rather than make a new eligibility decision. + +## Composed lifecycle coverage + +The external journey proves the public prepare → restart → resume → retire path. The released Core +suite supplies deterministic branch coverage for the rest of the supported lifecycle: + +| Promise | Released verification | +|---|---| +| model cannot approve or mutate governed state | `tests/test_governed_cognition_governance.py` and `tests/test_governed_cognition_lifecycle.py` | +| rejection creates a durable receipt without a revision or head change | `test_rejection_creates_receipt_without_revision_or_head_change` | +| approval creates an immutable revision and generation-checked active head | `test_human_approval_creates_immutable_revision_and_cas_head_atomically` | +| revision and rollback preserve history and restore selection | `test_governed_cognition_chain_survives_fresh_database_connection` | +| expired cognition is filtered before scoring | `test_expired_active_head_is_filtered_before_scoring` | +| unavailable required dependencies fail the candidate closed | `test_unavailable_required_dependency_fails_candidate_closed` | +| selected-but-unused cognition cannot be reported as successful use | `test_cognition_use_fails_closed_without_material_use_attribution` | +| matched outcomes distinguish helped, hurt, unproven, unused, and stale | `tests/test_governed_cognition_effectiveness.py` | +| harmful or stale evidence proposes revision or retirement but cannot activate it | `test_matched_high_confidence_help_and_harm_are_distinct` and `test_unused_and_stale_are_explicit_and_only_stale_can_propose_retirement` | + +These tests are part of the exact public 0.4.4 source identity. The Core 0.4.4 release gates passed +7,378 normal and 7,376 extension-disabled tests; all six official pull-request checks and trusted +publication passed. The Market consumer repository passed 133 tests with one expected skip in its +CI-equivalent locked environment; its official pull-request CI passed. + +## Boundary result + +- Core owns immutable proposals and revisions, human review receipts, active heads, discovery, + selection/use attribution, lifecycle transitions, effectiveness receipts, and persistence. +- Intelligence and the Market Domain Pack supply the external consumer context without adding + Market nouns to the cognition model or receiving approval authority. +- The consumer imports no `ace` or `core.engine` Python modules. It drives the public CLI over the + authenticated application boundary and is excluded from the inert Market wheel. +- No twelfth MCP tool, second cognition model, model-write authority, or domain-specific Core path + was added. + +## Limitations + +This evidence is bounded to ACE's documented single-node topology and trusted in-process Python +extensions. It does not prove cross-database portability, distributed approval, hostile-code +isolation, general causal accuracy, autonomous learning, or beneficial real-world outcomes. The +Market revision was selected alongside Core-default cognition: the receipts prove exact inclusion +and material use, not exclusive use. Effectiveness classification and revision/retirement +recommendations are proven by released Core conformance tests, not by a real Market outcome cohort. + +## Reconciliation decision + +GC1 advances from `active` to `passed`. The public builder surface, external-consumer lifecycle, +restart durability, exact attribution, failure controls, measurement classifications, revision, +rollback, and retirement promises now have reproducible evidence with explicit boundaries. This +closes the 0.4.x Governed Cognition milestone; 0.5.0 Reasoning into Action remains `next`, with its +own T1 and B1 acceptance work still not ready. diff --git a/tests/test_evidence_index_integrity.py b/tests/test_evidence_index_integrity.py index 4eff43c..82ea3d7 100644 --- a/tests/test_evidence_index_integrity.py +++ b/tests/test_evidence_index_integrity.py @@ -7,9 +7,10 @@ EVIDENCE_DIR = REPO_ROOT / "docs" / "evidence" EVIDENCE_README = EVIDENCE_DIR / "README.md" -# Historical candidate labels remain intentional after publication: the point-in-time P1/P2 -# records are immutable. The later public GI2 receipt is separately indexed as the authoritative -# passed closeout. Keep these checks aligned with tests/test_public_roadmap_positioning.py. +# Historical candidate labels remain intentional after publication: the point-in-time P1/P2 and +# GC1 builder-surface records are immutable. The later public GI2 and GC1 receipts are separately +# indexed as authoritative passed closeouts. Keep these checks aligned with +# tests/test_public_roadmap_positioning.py. LOCAL_MARKDOWN_LINK = re.compile(r"\[[^\]]+\]\(([^)]+\.md)(?:#[^)]*)?\)") @@ -50,6 +51,7 @@ def test_evidence_readme_indexes_manifesto_and_new_platform_records() -> None: required_targets = [ "../../MANIFESTO.md", "gc1-public-builder-surface-v1.md", + "gc1-public-external-consumer-v1.md", "gi2-public-cross-domain-falsification-v1.md", "platform-p1c1-declarative-source-mapping-v1.md", "platform-p1c2-governed-live-source-ingress-v1.md", @@ -74,12 +76,17 @@ def test_evidence_readme_indexes_manifesto_and_new_platform_records() -> None: def test_evidence_readme_labels_candidate_and_historical_records_honestly() -> None: text = EVIDENCE_README.read_text(encoding="utf-8") - gc1_line = next( + gc1_builder_line = next( line for line in text.splitlines() if line.startswith("- [GC1 public governed-cognition builder surface](") ) - assert "public" in gc1_line - assert "passed" in gc1_line - assert "GC1 active" in " ".join(text.split()) + assert "public surface passed" in gc1_builder_line + assert "historical pre-closeout state" in gc1_builder_line + + gc1_closeout_line = next( + line for line in text.splitlines() if line.startswith("- [GC1 public external-consumer closeout](") + ) + assert "public, passed" in gc1_closeout_line + assert "GC1 is therefore **passed**" in " ".join(text.split()) gi2_line = next(line for line in text.splitlines() if line.startswith("- [GI2 public cross-domain falsification](")) assert "public, passed" in gi2_line @@ -113,6 +120,7 @@ def test_every_evidence_markdown_file_referenced_from_this_pr_is_indexed() -> No "productized-state-golden-journey-v1.md", "gi2-public-cross-domain-falsification-v1.md", "gc1-public-builder-surface-v1.md", + "gc1-public-external-consumer-v1.md", } for name in newly_added: assert (EVIDENCE_DIR / name).is_file(), f"expected fixture file missing: {name}" diff --git a/tests/test_public_roadmap_positioning.py b/tests/test_public_roadmap_positioning.py index d68241f..c3f671e 100644 --- a/tests/test_public_roadmap_positioning.py +++ b/tests/test_public_roadmap_positioning.py @@ -5,31 +5,31 @@ ROADMAP = (Path(__file__).resolve().parents[1] / "ROADMAP.md").read_text(encoding="utf-8") ROADMAP_ONE_LINE = " ".join(ROADMAP.split()) -# The 0.4.2 checkpoint assertions are the coordinated post-release gate. Historical 0.4.1 GI2 and -# P1/P2 identities remain exact point-in-time evidence, while the later GC1 public-surface receipt -# owns only the builder-interface claim. Keep these aligned with test_evidence_index_integrity.py. +# The 0.4.4 checkpoint assertions are the coordinated GC1 closeout gate. Historical 0.4.1 GI2, +# 0.4.2 builder-surface, and P1/P2 identities remain exact point-in-time evidence. Keep these +# aligned with test_evidence_index_integrity.py. -def test_current_release_and_active_milestone_are_not_conflated() -> None: - assert "`ace-core` 0.4.2 is published on PyPI and GitHub" in ROADMAP - assert ROADMAP.count("| 0.4.x | Governed Cognition | **Active** |") == 1 +def test_current_release_and_passed_milestone_are_not_conflated() -> None: + assert "`ace-core` 0.4.4 is published on PyPI and GitHub" in ROADMAP + assert ROADMAP.count("| 0.4.x | Governed Cognition | **Passed** |") == 1 assert "| 0.4.0 | Governed Cognition | **Delivered** |" not in ROADMAP assert "| 0.4.0 | Governed Cognition | **Now** |" not in ROADMAP assert "| 0.5.0 | Reasoning into Action | **Next** |" in ROADMAP -def test_042_public_builder_surface_is_recorded_without_closing_gc1() -> None: - assert "[0.4.2 GitHub Release]" in ROADMAP - assert "public `ace-core==0.4.2` package" in ROADMAP_ONE_LINE - assert "[GC1 public builder-surface evidence]" in ROADMAP - assert "Its public builder surface is now verified from a fresh PyPI install" in ROADMAP_ONE_LINE - assert "This does not close GC1" in ROADMAP - assert "**GC1** therefore stays active in 0.4.x" in ROADMAP_ONE_LINE +def test_044_public_external_consumer_closes_gc1_without_rewriting_042() -> None: + assert "[0.4.4 GitHub Release]" in ROADMAP + assert "public `ace-core==0.4.4` package" in ROADMAP_ONE_LINE + assert "[GC1 public external-consumer evidence]" in ROADMAP + assert "point-in-time receipt for 0.4.2" in ROADMAP_ONE_LINE + assert "independent Market Intelligence consumer" in ROADMAP_ONE_LINE + assert "**GC1** is therefore passed in 0.4.x" in ROADMAP_ONE_LINE -def test_governed_intelligence_pass_does_not_overstate_governed_cognition() -> None: +def test_governed_intelligence_substrate_is_not_retroactively_made_gc1_evidence() -> None: assert "| GI1 | passed | Ship the governed Intelligence foundation as one installable" in ROADMAP - assert "| GC1 | active | Make governed cognition an obvious supported" in ROADMAP + assert "| GC1 | passed | Make governed cognition an obvious supported" in ROADMAP assert "This does not close GC1, SI1–SI4, or any domain product" in ROADMAP @@ -73,11 +73,11 @@ def test_041_and_both_domain_falsification_packages_are_public() -> None: assert "Core 0.4.1, World Intelligence 0.8.0, and Market Intelligence 0.6.0 artifacts" in ROADMAP_ONE_LINE -def test_platform_substrate_section_reconciles_only_gi2() -> None: +def test_platform_substrate_section_preserves_gi2_and_acknowledges_later_gc1_closeout() -> None: assert "### Governed Intelligence platform substrate" in ROADMAP assert "GI2 is therefore reconciled to `passed`" in ROADMAP - assert "GI1 remains passed, GC1 remains active" in ROADMAP - assert "SI1–SI4 remain bounded future outcomes" in ROADMAP + assert "GI1 and GC1 are passed" in ROADMAP + assert "SI1–SI4 remain bounded future outcomes" in ROADMAP_ONE_LINE def test_041_public_p2a_domain_pack_conformance_is_recorded() -> None: @@ -110,6 +110,6 @@ def test_041_publication_gate_is_complete_and_bounded() -> None: assert "clean public-index two-domain activation and retirement-isolation journey passes" in ROADMAP_ONE_LINE assert "does not establish hostile-code isolation" in ROADMAP_ONE_LINE - assert "| GC1 | active |" in ROADMAP + assert "| GC1 | passed |" in ROADMAP for outcome in ("SI1", "SI2", "SI3", "SI4"): assert f"| {outcome} | not ready |" in ROADMAP