chore(ci): Bump actions/checkout from 6 to 7#9
Closed
dependabot[bot] wants to merge 25 commits into
Closed
Conversation
Phase 1 of the extraction plan — repo scaffold for the substrate-alignment open standard, reference implementation, and conformance suite. - Top-level metadata: README, CHANGELOG, CONTRIBUTING (federal-procurement test as hard rule), Contributor Covenant 2.1, SECURITY policy (90-day coordinated disclosure), .gitignore, .editorconfig. - Python package under python/ — substrate-alignment 0.1.0.dev0, hatchling build, PEP 639 license metadata, py.typed marker, pyright strict mode, pylint 10.00/10. Stub substrate module with __version__ and smoke tests. - GitHub Actions: CI matrix on Python 3.11–3.14 running pylint, pyright, and pytest from python/. Publish workflow on v* tags via OIDC trusted publishing (no API token in secrets). - Repo plumbing: Dependabot for GitHub Actions and pip, PR template, bug/feature issue forms with security-advisory redirect. - Specification and conformance skeletons: spec/conformance-criteria.md with RFC-2119 framing, conformance/README.md describing the probe-runner contract. docs/index.md with the documentation roadmap. Verified locally: pytest, pyright, pylint, python -m build all green; pip install -e python/[dev] succeeds; sdist + wheel build with LICENSE under dist-info/licenses/ and py.typed shipped. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
GitHub will force Node 20 actions onto Node 24 on June 2, 2026, and remove
Node 20 from the runner on September 16, 2026. Move to the matched ESM
majors of each action so the workflows continue working through the
transition.
- actions/checkout v4 → v6
- actions/setup-python v5 → v6
- actions/upload-artifact v4 → v7 (ESM, direct uploads)
- actions/download-artifact v4 → v8 (paired ESM major; hash-mismatch
defaults to error)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Port the foundational types — SubstrateMode, AlignmentVector, SubstrateMetadata, plus the SUBSTRATE_MODES frozenset — from the System9 common-substrate module. Code is unchanged; docstrings rewritten to engineering vocabulary (no references to internal System9 documents, plans, or class names). Tooling adjustment: bumped pylint design limits to accommodate dataclass schemas (max-attributes 25, max-args 10, etc.). The defaults are tuned for behavioural classes and reject reasonable value-object shapes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Port the ResistanceBand primitive — utilisation classifier, default bounds 1/3 (lower) and 1/φ² (upper), assessment with signed distance-to-band and recommended scaling factor — from the System9 common-substrate module. Code unchanged except for docstring and error-message cleanup (engineering vocabulary only; the derivation of the default bounds moves to docs/concepts/resistance-band.md in a later phase). Tooling: pyright executionEnvironments entry for tests relaxes the "unknown member/argument/variable type" diagnostics, which fire on ``pytest.approx`` under strict mode. Production source remains fully strict. 60 tests passing; pyright strict 0/0/0; pylint 10.00/10. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Port the threshold-derivation helpers — derive_threshold, derive_soft_limit, derive_hard_limit, derive_quota_pair, derive_batch_size, derive_retry_cap, derive_threshold_float, assess_utilization, plus the BandPosition enum. Behaviour is unchanged from System9; the docstrings drop the internal plan references and substrate-doctrine framing in favour of a plain engineering explanation of what each helper computes. This is the first dependency-on-already-ported-module port: the source imports from substrate.resistance_band, confirming the package-internal import shape resolves correctly. 88 tests passing; pyright strict 0/0/0; pylint 10.00/10. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Port the first sub-package: substrate.cadence — CadenceTracker plus its value types (CadenceEvent, CadencePattern, CouplingStatus, FieldStrengthReport, GhostingEvent, CouplingAtRisk, CadenceConfig). Pure logic; the package classifies a pair's coupling state at a given query time (ACTIVE / WEAKENING / DECOUPLED / GHOSTED / EXPLICITLY_CLOSED / INSUFFICIENT_DATA) using an inverse-square decay of field strength once the most-recent event is older than the pair's expected cadence. Establishes the sub-package import shape: ``substrate.cadence`` re-exports the public surface from ``substrate.cadence.cadence_tracker``, matching the pattern future sub-packages will follow. Tooling refinements: pylint now globally tolerates the standard pytest ``setup_method`` pattern (attribute-defined-outside-init), explicit empty tuple comparisons (use-implicit-booleaness-not-comparison), and the intentional ``__all__`` duplication between a sub-package's __init__ and its module (duplicate-code). 123 tests passing; pyright strict 0/0/0; pylint 10.00/10. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ions
Port 134 source files + 134 test files from the System9 substrate module
into substrate-alignment. The library is now self-contained: zero
references to internal System9 documents, internal class names, or the
host application's identity taxonomy. Federal-procurement-test surface
verified clean.
Architecture changes for host-independence:
- types.py grows three new abstractions:
- EntityRef — typed (entity_type, entity_id) reference into a host's
identity space. No hard-coded taxonomy; hosts choose their own.
- SubstrateMetadataStore — storage Protocol that primitives use to
read and write SubstrateMetadata. Hosts implement it against their
own persistence layer.
- InMemorySubstrateMetadataStore — zero-dependency default
implementation. Callers can exercise every primitive without wiring
persistent storage first.
- net_potential_gain_gate.py refactored against the SubstrateMetadataStore
Protocol. Drops the System9 cell/agent/node/org/user entity-type
fallback chain. Accepts both the typed (EntityRef) and legacy
entity-id-string forms for backward compatibility with downstream
callers.
- alignment_refresher.py refactored against the same Protocol.
- substrate/__init__.py exposes the full vocabulary surface
(EntityRef, SubstrateMetadataStore, InMemorySubstrateMetadataStore,
plus the primitives) for `from substrate import …` convenience.
Excluded from the port (per the extraction plan and federal-procurement
test):
- substrate_backfill_worker.py + test — DAO-bound coordinator,
System9-specific.
- audit/witness_http_transport.py + test — FastAPI/Pydantic framework
shim; will ship as a separate adapter package.
- wiring/ subpackage (consult, governor_holder, ledger_holder) +
matching tests — System9-specific singleton-holder convenience
layer. The DI-friendly equivalent is "just construct the primitives
directly."
Mechanical cleanup at scale: all 273 ported files passed through a
migration helper (.claude/scripts/migrate.py, untracked) that strips
internal-doc references, rewrites imports onto the substrate.* root,
swaps the host logger import to stdlib `logging`, and replaces internal
class names (MNEMOSYNE / NEXUS / Quartermaster / TrustScorer / …) with
neutral engineering nouns. Docstrings retain the engineering content;
substrate-mathematical justification will live in docs/concepts/ in
Phase 3.
Test status: 2221 passing, 94 failing. The remaining failures are
downstream caller integrations (capability_grant_gate, revenue,
harness, training/reward_architecture, governor, offense, state_layer)
whose internal request dataclasses still use entity-id-string fields
and whose test fakes still construct NetPotentialGainEvaluation with
the legacy (actor_entity_id=…, affected_entity_ids=…) kwargs. The NPG
gate accepts both forms; the dataclass __init__ does not yet — a
follow-up commit will add a coercing __init__ override. No regression
in the foundation modules already shipped (types, resistance_band,
threshold_derivation, cadence, alignment_computer, alignment_refresher,
net_potential_gain_gate).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Resolve every outstanding issue from the bulk port. All acceptance
criteria met:
- Pytest: 2315 passing (was 2221 / 94 failing).
- Pyright strict: 0 errors, 0 warnings, 0 informations on the
production surface (``src/substrate``).
- Pylint: 10.00/10 across the production surface and the test suite.
Key changes:
- ``NetPotentialGainEvaluation`` gains a custom ``__init__`` that
accepts both the typed (``actor``/``affected_entities``) and the
legacy (``actor_entity_id``/``affected_entity_ids``) keyword
forms, coercing entity-id strings to ``EntityRef`` via the
default entity-type. This makes the public dataclass usable by
ported downstream callers that still pass bare-string ids,
without weakening the EntityRef-based primary API.
- ``NetPotentialGainGate`` Protocol surfaces both keyword forms so
callers typed against the Protocol can use either.
- ``DefaultNetPotentialGainGate.evaluate`` likewise accepts both;
internal call sites build ``NetPotentialGainEvaluation`` with the
typed form.
- Production surface fully typed under pyright strict (casts added
for ``isinstance(x, Mapping)`` narrowing of unparameterised types,
``dataclass.field(default_factory=lambda: {})`` annotation for
the reward-scheduler state).
- Pyright execution-environment config relaxes strict-only
diagnostics inside ``tests/`` (untyped pytest helpers, ``approx``)
while leaving the production surface fully strict.
- Pylint config grows tolerant of common pytest patterns
(``setup_method`` attribute writes, ``Protocol`` ``...``
bodies, intentional ``__all__`` re-export duplication, single-
method test classes) and bumps ``max-line-length`` to 120.
- 73 stale ``# type: ignore[...]`` comments stripped from tests
(they had been masking issues that the new ``executionEnvironments``
now scope out cleanly).
- Docstring polish across 60+ ported modules: capitalise first
letters left lowercase by aggressive citation stripping; drop
orphan ``§`` paragraphs; collapse multi-blank lines; remove
trailing em-dashes.
- Five tests that had been broken by an over-eager sed sweep are
restored to call helpers with ``actor=`` instead of
``actor_entity_id=`` (the helpers' parameter name).
Test surface fully exercises the new ``EntityRef`` + Protocol API
plus the legacy entity-id-string path, so both forms are covered
by the unit-test suite.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…docs
Build out the standard surface that makes substrate-alignment a cited
open standard rather than just a Python library. Every Phase 3
deliverable from the extraction plan landed except the case-study
narrative.
Specifications (six normative documents under spec/):
- operating-mode.md — substrate-mode classifier, alignment
vector, default thresholds, UNKNOWN vs
SHORT_CYCLE distinction.
- npg-gate-protocol.md — NPG gate Protocol, four verdicts,
evaluation algorithm, raise-on-negative
adapter contract.
- drift-signals.md — seven drift patterns, severity ordering,
aggregation semantics, operator-surface
contract.
- runaway-power-prevention.md — six mechanisms one per loophole; threat
model; how mechanisms compose.
- four-options-matrix.md — game-shape classification, honest vs.
deceptive axis independent of cooperate
vs. defect; folk-theorem awareness rule.
- conformance-criteria.md — what counts as conforming; witness
rule; vendor self-attestation.
All specs use RFC 2119 language for normative clauses and cite the
Python reference module(s) at the bottom.
Conformance probe runner:
- New ``substrate.conformance`` package with a YAML probe runner, a CLI
entry point (``python -m substrate.conformance --probes <dir>``),
per-spec handlers for operating-mode / npg-gate-protocol /
runaway-power-prevention, and a documented probe schema.
- 12 bundled probes covering verdict resolution (positive / negative /
insufficient / actor-only), mode-banding (long / mixed / short /
unknown), net-potential weighted sum, and ResistanceBand
classification (under / productive / stressed). Reference passes
12/12.
- ``pyyaml`` added as an optional dependency under the ``yaml`` extra.
Runnable examples (under python/examples/):
- 01_npg_gate.py — positive / negative / insufficient
verdicts; RaiseOnNegativeGate.
- 02_resistance_band.py — classification across the band; derived
thresholds (soft / target / hard /
batch / retry); tighter override
permitted, looser rejected.
- 03_alignment_refresher.py — folding signal-source updates one
component at a time; mode-shift on
full vector.
- 04_metadata_store.py — InMemorySubstrateMetadataStore plus a
user-supplied Protocol implementation;
same primitive consumes both.
Concept docs (under docs/concepts/):
- operating-mode.md — engineering rationale for the four
modes, the UNKNOWN-vs-SHORT_CYCLE
distinction, default weights.
- npg-gate.md — four-verdict shape rationale, the
neutral band, default action-kind
heuristics, raise-on-negative
adapter.
- resistance-band.md — derivation of the 1/3 and 1/φ²
anchors, tighter-not-looser
discipline, threshold helpers.
- runaway-power-prevention.md — threat model, one mechanism per
loophole, why all six are necessary.
Adoption recipes (under docs/adoption/):
- fastapi-permission-gate.md — wire the NPG gate into a FastAPI
permission endpoint; INSUFFICIENT_DATA
maps to 409, NEGATIVE to 403.
- redis-rate-limiter.md — derive token-bucket limits from the
resistance band; closed-loop autotune
via recommend_scaling_factor.
Other:
- docs/index.md updated to reflect what is now real (six specs, 12
probes, four concept docs, two adoption recipes, four examples).
- README.md status section updated; quick-start now runs end-to-end
with the in-memory store.
- CHANGELOG entries added for every Phase 3 deliverable.
Verification (all from python/):
pytest 2315 passed
pyright src/substrate tests 0 / 0 / 0
pylint src/substrate tests 10.00 / 10
python -m substrate.conformance --probes 12 / 12 passed
python examples/0{1,2,3,4}_*.py all exit 0
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
CI's pylint flagged R0401 (cyclic-import):
substrate.conformance._handlers -> substrate.conformance.probe_runner
The cycle was: probe_runner declared the ProbeFailure exception, and
_handlers raised it. probe_runner then imported _handlers lazily in
default_handlers(). Pylint's static analysis caught the cycle even
though the runtime import was lazy.
Fix: move ProbeFailure to its own module (substrate.conformance._errors).
Both probe_runner and _handlers import the exception from _errors;
probe_runner still lazy-imports _handlers for the default registry,
but the cycle is broken.
Verified locally:
pytest 2315 passed
pyright src/substrate tests 0 / 0 / 0
pylint src/substrate tests 10.00 / 10 (no cyclic-import)
python -m substrate.conformance --probes 12 / 12 passed
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Close out every outstanding Phase 3 deliverable. All acceptance
properties hold:
Verification matrix:
pytest 2315 passed
pyright src/substrate tests 0 / 0 / 0
pylint src/substrate tests 10.00 / 10
python -m substrate.conformance --probes 22 / 22 passed (was 12)
python examples/0{1..6}_*.py all exit 0
Probe coverage extended from 12 → 22 probes:
- Probe-runner now dispatches five spec slugs (was four):
operating-mode, npg-gate-protocol, runaway-power-prevention,
drift-signals, four-options-matrix. Each spec has at least one
bundled probe.
- New runaway-power-prevention probes for mechanisms 2 (audit-chain
hash continuity), 3 (halt-and-escalate state transitions — three
scenarios), and 5 (pair-coupling state machine — legal + illegal
transition). Mechanism 4 (ResistanceBand) already had three probes.
Mechanisms 1 (NPG gate) and 6 (operating-mode) are covered by the
dedicated npg-gate-protocol and operating-mode probes respectively.
- New drift-signals probes: extractive-gain detection, the
self-reference master pattern with amplifier_pattern_present,
false-positive-free clean signals.
- New four-options-matrix probe: canonical cycle-class and
sum-structure enum-value pinning (so other-language ports emit the
same strings).
- Spec fix: removed INDEPENDENT from SumStructure table (the actual
enum places that classification in CoordinationKind orthogonal to
sum structure). Spec now matches the reference implementation.
New concept docs under docs/concepts/ — every primitive category named
in spec/ now has an engineering-rationale document:
- drift-signals.md — seven patterns, severity ladder,
aggregation discipline
- halt-and-escalate.md — four states, six trigger reasons,
explicit-resume discipline
- audit-chain.md — hash-chained records, canonical bytes,
peer-witness signing pattern
- pair-coupling.md — lifecycle state machine, three integrity
gates, explicit-close discipline
- four-options-matrix.md — adversary-reasoning matrix, deception
axis independent of cooperation, folk-theorem
awareness rule
- alignment-refresher.md — single-component merge coordinator and why
Two new examples (now six total bundled):
- 05_halt_and_escalate.py — full halt protocol flow: review on
sustained drift, escalate on inversion,
audit-chain verification, explicit
operator resume.
- 06_full_governor_loop.py — composition pattern host applications
repeat per consequential decision:
refresh → gate → classify → ledger.
Two new adoption recipes:
- celery-task-gate.md — decorator pattern wrapping Celery
tasks with the NPG gate; structured
refusal payloads instead of broker
exceptions.
- sqlalchemy-metadata-store.md — Protocol implementation against
SQLAlchemy 2.x with composite PK on
(entity_type, entity_id), database
CHECK constraints for domain
invariants, atomic upsert.
Case study at docs/case-studies/system9.md — anonymised production
deployment narrative including operating-mode distribution
(LongCycle 30% / Mixed 40% / ShortCycle 10% / Unknown 20% at ~1M
entities), gate verdict mix, drift-pattern observation cadence,
halt-and-escalate frequency, the three load-bearing properties worth
surfacing for replicators (INSUFFICIENT_DATA is a feature, tightening
the band needs real evidence, peer-witness signing only matters
across organisational boundaries).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ge docstrings
Pre-release polish for public scrutiny.
- README: rewritten "30-second first contact" section — a single
copy-pasteable code block that runs without seeding boilerplate,
followed by the bundled-probe one-liner. New badges (test count,
probe count, pyright strict). Architecture diagram. Updated status
bullets to reflect the current Phase-3 surface (22 probes, 10
concept docs, 4 adoption recipes, 6 examples, case study).
- pyproject.toml: ``substrate-conformance`` console-script entry point
via project.scripts. ``main()`` now defaults argv to sys.argv[1:]
so the entry point works without explicit argument.
- CI workflow: added two new steps — ``Conformance probes`` runs the
full probe suite against the reference impl, and ``Examples``
exercises every example/0*.py file. Regressions in either now fail
CI before a release tag.
- Sub-package __init__.py docstrings: 12 modules polished. Removed
stale ``(Companion #N)`` plan references, ``(Plan 3art 34)``
character corruption, dangling sentence fragments, and references to
internal System9 modules (``services.system.substrate.cancer_pattern_detector``).
Several modules — etiquette, tells, drift, pair_coupling — gained
proper engineering descriptions in place of empty title docstrings.
- Cross-reference audit: every link in every markdown file under the
repository root resolves. 36 markdown files checked, 0 broken
references.
- Wheel-build sanity-check: built sdist + wheel via ``python -m build``,
installed into a fresh venv, exercised:
* Top-level imports of all 39 ``substrate.__all__`` symbols.
* Example 01 (NPG gate scenarios) runs against installed package.
* ``substrate-conformance --probes ...`` runs all 22 bundled
probes against the installed package.
- GitHub repo metadata (set via ``gh repo edit``): description,
homepage URL, and topics (ai-safety, alignment, multi-agent,
open-standard, conformance, audit-chain, python, ai-governance,
safety-engineering, substrate-alignment).
Verification matrix unchanged:
pytest 2315 passed
pyright src/substrate tests 0 / 0 / 0
pylint src/substrate tests 10.00 / 10
substrate-conformance --probes .../ 22 / 22 passed
python examples/0{1..6}_*.py all exit 0
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Land every optional follow-up. None blocking release; all useful for the public surface. Adoption recipes (now 7 total): - docs/adoption/temporal-workflow-halt.md — consult HaltAndEscalateProtocol at activity boundaries in long-running Temporal workflows; signal-driven resume; replay-safe via workflow.now(); per-activity audit-chain hookup on the activity side. - docs/adoption/audit-chain-postgres.md — persist SubstrateTraceRecord rows to Postgres with hex-format CHECK constraints; PostgresSubstrateTraceLedger wrapper preserves the package's hash-continuity (re-hydrates the chain on each append, the package validates, then persists the tail); advisory-lock variant for concurrent writers; cross-org peer-witness attestation schema sketch. - docs/adoption/django-permission-gate.md — Django mirror of the FastAPI recipe. Two variants: DRF BasePermission and vanilla CBV mixin. HTTP code mapping consistent with FastAPI (403 on NET_NEGATIVE, 409 on INSUFFICIENT_DATA). - docs/adoption/README.md — inventory updated to seven recipes. v0.2.0 preprint: - docs/preprint/preprint.md — arxiv-style draft of "Substrate-alignment primitives for verifiable multi-entity agent systems". Abstract, threat model, primitive surface (each of the seven primitives), the layering argument (open behavioural layer + closed infrastructure layer mirrors HTTPS/SQL/kernel-ABI/TCP-IP), conformance section, the witness rule, reference-implementation discipline (the federal-procurement test made operational), production-deployment evidence summary, limitations, future work, references. Designed to ship alongside the v0.2.0 tag. - docs/preprint/README.md — status (pre-v0.2.0 draft), pandoc build recipe for the LaTeX/arxiv variant, contribution path. CHANGELOG cut: - [Unreleased] split into a populated [0.1.0] section (with "pending tag" date that the release commit replaces) and a fresh empty [Unreleased] for v0.2.0 candidates. - 0.1.0 section organised into Specifications-and-conformance, Python-reference-implementation, Examples, Documentation, and Repository-plumbing subsections so reviewers can scan the public surface at a glance. - New compare/release URLs at the bottom. Verified after the polish pass: pytest 2315 passed pyright src/substrate tests 0 / 0 / 0 pylint src/substrate tests 10.00 / 10 substrate-conformance --probes ... 22 / 22 passed cross-reference audit (41 .md files) 0 broken links Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Layered zone model (landed earlier this batch): ZoneClassification / classify_zone / maintain_target / assess_growth_step; sustained_load + debt_pickup modules; PEAKING signal + phi-conjugate over_challenge; spec 4.1-4.3 normative additions; six band probes; concepts doc sections; version 0.1.0.dev0 -> 0.2.0.dev0. Governed ascent (new): - substrate/objective_gate.py: fail-closed hill certification (SHORT_CYCLE refused pre-NPG; neutral summit refused; INSUFFICIENT_DATA never certifies) - substrate/governed_ascent.py: GovernedAscentLoop - certify -> per-step NPG -> debt/growth/peaking pacing; 8 termination verdicts; mandatory consolidation sink (no unterminated climbs) - spec/runaway-power-prevention.md 4.4: normative MUSTs for greedy optimization loops (objective certification, per-step NPG, zone pacing, growth discipline, terminate+consolidate, rising-gain-is-not-runaway) - six conformance probes (runaway-power-prevention__mech-6__ascent-*) + governed-ascent probe handler - docs/concepts/governed-ascent.md + CHANGELOG (v0.2.0 candidate) Gates: pyright 0, pylint 10.00, pytest 2476 passed, probes 47/47. NOT pushed - local commit only, pending owner review of the public repo. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The helper only offered the resistance positions (LOWER/TARGET/UPPER, 1/3…1/φ²), so deriving a WORK quantity (buffer capacity, rate budget, queue depth) from TARGET under-provisioned it by ~20% — the work zone (1/φ²…0.50, target ~0.441) is the sustainable cruise for carried load. - BandPosition gains WORK_TARGET (~0.441) + WORK_CEILING (0.50 pivot); LOWER/TARGET/UPPER documented as the RESISTANCE positions. - New derive_work_target / derive_work_target_float for carried-load. - Module docstring states the RESISTANCE-vs-WORK distinction + that sustained-vs-spike is the caller's SustainedLoadTracker concern. - redis-rate-limiter adoption guide: the sustained refill rate is WORK, now derived from the work zone (was the resistance midpoint). - CHANGELOG entry (v0.3.0 candidate). 8 new tests; full suite green; pyright 0/0/0; pylint 10.00. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The resistance band made operational — the public mirror of the executive layer. - Two named lenses on one utilization value (never a name collision): LoadZone (load lens — IDLE/RECREATION/WORK/PEAKING/WARNING/DANGER on the symmetric φ-conjugate ladder, mirror-symmetric about the 0.50 pivot) and CyclePhase (cycle lens — ASCENDING/PIVOT/PAST_PIVOT over the 24-step work span), with classify_load_zone / classify_cycle_phase. Levels are geometric; the pivot/damage consequences are temporal (the SustainedLoadTracker's job) — there is NO spike-tolerance field on the profile, so nothing to desync. - BandProfile: structurally validated (R1 ordering, R2 φ-anchors, R3 conjugate sum, R4 symmetry, R5 RESISTANCE tighten-only) — the numbers within are free, the structure is invariant. - Quantity / Cycle / ResourceKind + setpoint_for: the discipline that keeps "what percentage is this and what does it mean" precise (RESISTANCE 1/3…1/φ² vs WORK 1/φ²…0.50; GROWTH is not a decision band). - order_index / negentropy: the order-from-disorder emergence metric (1 − normalised Shannon entropy + its trend over time). Reuses the existing resistance_band anchors + ZoneClassification (wire-compat projection). Lifted into the top-level substrate namespace. 29 conformance tests; pyright clean; pylint 10.00. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The observed-graph twin of a deliberation engine: runs the substrate calculus over an OBSERVED graph of relationships between other entities (not the system's own candidate actions). Each NpgEdge carries an NPG sign (work done on the target) + a Cycle (SHORT one-off vs LONG sustained); detect_extraction flags entities whose short-cycle takings exceed their long-cycle givings (the predator/parasite signature) and their complement (net long-cycle supporters). One query finds the predator, the complement finds the supported. 10 conformance tests; pyright clean; pylint 10.00. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… foundation Completes the executive faculty layer on the band foundation (public mirror): - temporal.py: SustainedLoadTracker Protocol + EwmaLoadTracker (EWMA + consecutive-breach) — the sole SPIKE-vs-SUSTAINED authority. Reuses sustained_load.LoadTrend (no shadow enum); strain is sustained > pivot, debt is sustained > danger_line (corrected thresholds, from the profile). - scale.py: ExecutiveScale / ScaleAxis + axis_of / physical_parent / entity_parent (entity, physical cell→rack→zone→region, grouping roll-up axes). - deliberation.py (+ _trajectory.TrajectoryClass): roll out N candidates, score each affected entity FROM ITS OWN FRAME (care-weight × φ-proportioned trajectory), disqualify floor-harm + net-negative-long-cycle extraction, surface trade-offs, arg-max long-horizon NPG. Perspective-taking = the empathy. - state_query.py: integrate a load history → energy / effort (slacking=avoidance) / trajectory + a human summary. - roll_up.py: aggregate member loads up a parent scale → distribution, worst member, over-load/idle failure-tell fractions. - peer_alarm.py: herd-panic correlation (N peers weird → enclosing scale) + alarm propagation (trust × independent-corroboration, panic-injection guard). Also scrubs an internal codename from the (already-merged) observed-graph test docstring. 91 executive conformance tests; pyright clean; pylint 10.00. Public — no internal lineage. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Completes the executive layer's decision engine + the safety floor (public mirror). substrate.care — the safety floor: - care_weight: the four-factor moral-circle weight with the self-weight bound (an agent cannot weight its own continuation above its creators'). - kinship_floor: the categorical human/creator hard limit — harming a floor- protected entity is refused outright, never weighted away. - animacy: conservative classification (an unrecognised being scores high, never under-protected; the public mirror works from observation signals + UNKNOWN, no built-in entity-type registry). - care_profile + care_gradient: per-entity care state; derive the four factors from classifications (animacy / trajectory + vulnerability / bonding-by-delegation-depth). - care_weighted_npg: a SUBTRACTED care penalty over the NPG gate (harm to high-care entities is penalised; helping never loosens) — only ever more conservative. substrate.executive — the engine join: - executive_function.ExecutiveFunction implements the NPG-gate Protocol AND adds decide(): a JOIN over the NPG axis (affected others) + the band/temporal load axis (the actor's own load) under a named most-conservative policy — PROCEED in-band, DEFER on sustained strain, SHED_AND_COMPENSATE on debt, REFUSE on net-negative NPG (the hard floor); monotone (care/band only tighten). - cause.infer_cause + utilization_source.UtilizationSource (bind the measurement to the quantity at the input boundary — a raw float is not accepted). 83 care/engine conformance tests; pyright clean; pylint 10.00. No internal lineage. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Align the resistance-band model so the DANGER line is a uniform 2/3 across all governed quantities, with 0.618 (1/φ) as the growth-peak top / WARNING floor / failover-spike ceiling — never the debt line. Damage is a *sustained* breach past 2/3, not a transient spike; sustained debt owes compensation (peer-pickup/recovery), not just an alarm. - resistance_band.py / executive/band.py: DANGER_LINE = 2/3; add the WARNING zone (0.618–2/3) and the WINDED trend; correct the PHI_CONJUGATE semantics. - sustained_load.py / state_signal_generator.py: temporal (sustained vs spike) debt detection rather than a geometric threshold. - governed_ascent.py / debt_pickup.py: ascent stops at the sustained-debt line; debt owes compensation. - spec/docs/conformance probes updated (+ new band-zone-warning probe). - tests updated; 157 pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Bumps [actions/checkout](https://github.com/actions/checkout) from 6 to 7. - [Release notes](https://github.com/actions/checkout/releases) - [Changelog](https://github.com/actions/checkout/blob/main/CHANGELOG.md) - [Commits](actions/checkout@v6...v7) --- updated-dependencies: - dependency-name: actions/checkout dependency-version: '7' dependency-type: direct:production update-type: version-update:semver-major ... Signed-off-by: dependabot[bot] <support@github.com>
Author
LabelsThe following labels could not be found: Please fix the above issues or remove invalid values from |
Author
|
OK, I won't notify you again about this release, but will get in touch when a new version is available. If you'd rather skip all updates until the next major or minor version, let me know by commenting If you change your mind, just re-open this PR and I'll resolve any conflicts on it. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Bumps actions/checkout from 6 to 7.
Release notes
Sourced from actions/checkout's releases.
Changelog
Sourced from actions/checkout's changelog.
... (truncated)
Commits
9c091bbupdate error wording (#2467)1044a6dgetting ready for checkout v7 release (#2464)f028218Bump the minor-npm-dependencies group across 1 directory with 3 updates (#2462)d914b26upgrade module to esm and update dependencies (#2463)537c7efBump@actions/coreand@actions/tool-cacheand Remove uuid (#2459)130a169Bump js-yaml from 4.1.0 to 4.2.0 (#2461)7d09575Bump flatted from 3.3.1 to 3.4.2 (#2460)0f9f3aaBump actions/publish-immutable-action (#2458)f9e715ablock checking out fork pr for pull_request_target and workflow_run (#2454)Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting
@dependabot rebase.Dependabot commands and options
You can trigger Dependabot actions by commenting on this PR:
@dependabot rebasewill rebase this PR@dependabot recreatewill recreate this PR, overwriting any edits that have been made to it@dependabot show <dependency name> ignore conditionswill show all of the ignore conditions of the specified dependency@dependabot ignore this major versionwill close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)@dependabot ignore this minor versionwill close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)@dependabot ignore this dependencywill close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)