CR-7: synthetic-clock test for ActivityMonitor.run() (skipped pending
      richer harness) + PerceptionEvent round-trip tests pin schema
IM-2: _SessionReader delegates to get_active_session; removes
      duplicate validity logic + IM-1 last_read_failed flag (spurious
      flush on Redis recovery is bounded by drop-log)
IM-4: advisor prompt builders raise ValueError on missing required
      context fields (no more ?-filled prompts → Ollama)

IM-8 (integration test polling) lands in a follow-up.

Co-Authored-By: Rooty

IM-8 from PR #6 review. T12-T16 now poll for Redis keys / NATS
message arrival instead of sleeping a fixed 2-3s after publish.
Test logic unchanged; flakiness drops and runtime tightens.

Co-Authored-By: Rooty

S2:  shorten severity-keyed-window comment in correlator
S3:  expose Domain = Literal[...] alias in contracts.py
S4:  promote (domain, entity) → TrackingKey NamedTuple in
     feedback_collector
S5:  PendingAdvice.explicit_rating is Literal["y","n","no_response"]
S6:  strip multi-paragraph docstrings that narrated WHAT
S12: thread-safety test now verifies no lost increments (was
     just asserting join completed)
S13: parallel DOMAIN_DESCRIBERS registry next to DOMAIN_HANDLERS;
     test pins their keys identical

Co-Authored-By: Rooty

N1: drop dead redis import (only referenced in a type-hint string
    since IM-2 removed _SessionReader's specific error handling)
N2: skipped run() test enumerates the wiring it would cover
    (publish ordering, intensity reset timing, session-change reset)
N3: comment on feedback_collector.get_session_id explaining the
    intentional lax semantics vs get_active_session

Co-Authored-By: Rooty

feat(perception): activity_focus + activity_intensity domains

- Status line: 2 perception domains → 4
- Components table: add activity_monitor.py + bump test counts
- "Adding a new perception domain": correct describe_signal location
  (augur_advisor.py, not correlator.py) + mention DOMAIN_DESCRIBERS
  registry and the key-parity test
- New section "Optional Windows companion: activity perception"
  covering install, privacy default, drop-log semantics
- Status/roadmap: workstation activity Phase 1 → Shipped;
  Phase 2 (system) and Phase 3 (biometric) listed under Future
- Dependencies: requirements-windows.txt block

Co-Authored-By: Rooty

Prior intensity prompt menu — "high-energy work, fatigue,
automation, or distraction" — was 3/4 pathological. Focus was
equally biased. Live runs against click-heavy consumer apps
consistently produced "automation" advice for ordinary browsing.

New prompts add "normal heavy use" / "normal varied task" as
first-class options, invite weighing the keystroke/click
breakdown, and warn that early-session deviation alone is weak
signal.

Co-Authored-By: Rooty

asyncio on Windows resolves "localhost" to ::1 first but Docker
binds IPv4-only, so the activity daemon's NATS client silently
timed out while a TCP probe to "localhost" succeeded. Env override
still works.

Detection baselines forming in 3 samples scored the 4th observation
as anomalous on freshly-seen apps. Bumped to 15; locked detector
default to AugurConfig with a parity test; adaptive warmup in the
graph-flush integration test.

Co-Authored-By: Rooty

The 3→15 bump in 693ef99 broke 4 integration tests on CI that
hardcoded 10-sample warmup arrays. Locally they passed because Redis
held trained baselines from prior runs.

Adopt the same adaptive pattern already used elsewhere — pool cycled
to config.min_observations + 5 — so future detector tuning doesn't
require touching every test.

Co-Authored-By: Rooty

CodeQL runs on push, PR, and weekly with the security-extended and
security-and-quality query packs. pip-audit runs in the lint job to
catch CVEs in pinned deps; verified clean locally before wiring up.

Co-Authored-By: Rooty

- Docker build smoke job builds the deploy image with buildx cache
  and verifies core modules import. Catches Dockerfile drift when
  requirements.txt changes.
- Unit test step now emits coverage XML uploaded as an artifact.
- mypy step runs in report-only mode (~40 errors today, mostly
  redis-py / nats-py library-typing false positives — tighten
  module-by-module later).

Co-Authored-By: Rooty

One-time install: `pre-commit install`. Same ruff check + ruff
format checks as the CI lint job, run on every git commit so
format drift is caught locally instead of after a CI cycle.

Co-Authored-By: Rooty

pytest --cov and mypy now run in CI and may run locally; their
output (.coverage, coverage.xml, htmlcov/, .mypy_cache/) should
never land in the tree.

Co-Authored-By: Rooty

- augur_advisor.py: drop dead `system_prompt = None` in the
  correlation branch; the variable was never used after assignment.
- typing_monitor.py: document the deliberate drop-on-queue-full
  pattern in the keyboard hook to satisfy py/empty-except.

Co-Authored-By: Rooty

Visible signal of build health and basic project facts at the top
of the README. CI + CodeQL badges link to their workflow pages.

Co-Authored-By: Rooty

upload-artifact@v4 → @v7, build-push-action@v6 → @v7,
setup-buildx-action@v3 → @v4. All three were still on Node 20;
new majors default to Node 24. No breaking changes affect current
usage (verified inputs against changelogs).

Co-Authored-By: Rooty

Redis-hash app->descriptor map on PersistenceManager (HSET for
authoritative writes, HSETNX so fallbacks never clobber; byte-decode;
2000-entry cap). New ollama_classifier_model/_enabled/_timeout config.

Co-Authored-By: Rooty

Resolve a short descriptor per app from event-borne OS identity, with a
dedicated small-model LLM fallback on a single-flight lane that runs
parallel to the advice call (ASCII-guarded entity names, bounded
timeout). Inject it into activity and correlation prompts; expose a
read-only MCP tool.

Co-Authored-By: Rooty

Resolve each foreground app's FileDescription (cached per exe, with a
pid+exe short-circuit) and attach it to focus/intensity events: the
previous app for focus, the current app for intensity.

Co-Authored-By: Rooty

Bumps [github/codeql-action](https://github.com/github/codeql-action) from 3 to 4.
- [Release notes](https://github.com/github/codeql-action/releases)
- [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md)
- [Commits](github/codeql-action@v3...v4)

---
updated-dependencies:
- dependency-name: github/codeql-action
  dependency-version: '4'
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

Updates the requirements on [pywin32](https://github.com/mhammond/pywin32) to permit the latest version.
- [Release notes](https://github.com/mhammond/pywin32/releases)
- [Changelog](https://github.com/mhammond/pywin32/blob/main/CHANGES.md)
- [Commits](https://github.com/mhammond/pywin32/commits)

---
updated-dependencies:
- dependency-name: pywin32
  dependency-version: '311'
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

Updates the requirements on [nats-py](https://github.com/nats-io/nats.py) to permit the latest version.
- [Release notes](https://github.com/nats-io/nats.py/releases)
- [Commits](nats-io/nats.py@v2.6.0...v2.14.0)

---
updated-dependencies:
- dependency-name: nats-py
  dependency-version: 2.14.0
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

Updates the requirements on [pynput](https://github.com/moses-palmer/pynput) to permit the latest version.
- [Changelog](https://github.com/moses-palmer/pynput/blob/master/CHANGES.rst)
- [Commits](moses-palmer/pynput@v1.7.0...v1.8.2)

---
updated-dependencies:
- dependency-name: pynput
  dependency-version: 1.8.2
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

Updates the requirements on [psutil](https://github.com/giampaolo/psutil) to permit the latest version.
- [Changelog](https://github.com/giampaolo/psutil/blob/master/docs/changelog.rst)
- [Commits](giampaolo/psutil@v5.9.0...v7.2.2)

---
updated-dependencies:
- dependency-name: psutil
  dependency-version: 7.2.2
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

Updates the requirements on [redis](https://github.com/redis/redis-py) to permit the latest version.
- [Release notes](https://github.com/redis/redis-py/releases)
- [Changelog](https://github.com/redis/redis-py/blob/master/CHANGES)
- [Commits](redis/redis-py@v5.0.0...v8.0.0)

---
updated-dependencies:
- dependency-name: redis
  dependency-version: 7.4.0
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

Adds all gate_* fields to AugurConfig (§7): master + per-arm enable
flags, every tuning knob with spec defaults, and gate_tier1_mode with a
validating _coerce_gate_tier1_mode assigned after the auto-build loop.
__post_init__ asserts 0 <= gate_behavioral_weight <= gate_explicit_weight.

Co-Authored-By: Rooty

Add fakeredis>=2.0 to requirements.txt; pip-install into .venv.
Add _CountingFakeRedis / _InstrumentedPM wrappers and the fake_pm,
cfg, cfg_disabled, fake_pm_at_cap fixtures plus six payload
constants (SINGLE_MEDIUM, SINGLE_MEDIUM_TYPING, SINGLE_HIGH_TYPING,
EXEMPT_PAYLOAD, CORRELATION_MEDIUM,
SINGLE_MEDIUM_NEWKEY_THAT_WOULD_SUPPRESS) used across gate unit tests.
Fixtures guard missing Phase-1 constants (MAX_GATE_STATE_KEYS) so
test collection stays clean until persistence layer is implemented.

Co-Authored-By: Rooty

- fake_pm_at_cap: reset write_calls/read_calls to 0 after setup pre-fill
  so cap-fail-open tests that assert write_calls==0 start from a clean
  baseline (the three hset pre-fills were incrementing the counter)
- _WRITE_CMDS/_READ_CMDS: add lpop (write) and exists (read) to match the
  full PersistenceManager Redis surface; add a comment explaining why
  pipeline/MULTI-EXEC is not intercepted per-command
- AugurConfig.__post_init__: validate gate_tier1_mode membership on every
  construction path, not just from_env(); direct construction now raises
  ValueError for invalid values, consistent with other validated fields
- test_gate_tier1_mode_garbage_falls_back: assert the WARNING is logged by
  augur.config (caplog was previously unused/misleading); add a separate
  test_gate_tier1_mode_invalid_direct_construction for the new path

Co-Authored-By: Rooty

…very_failures)

Add MAX_GATE_SILENCES/EMISSIONS/OBSERVED=2000, MAX_GATE_DELIVERY_FAILURES=500,
MAX_GATE_STATE_KEYS=2000 constants and four save_/load_ pairs per spec §6.
Each list uses lpush+ltrim (mirrors save_feedback) and loads with a guarded
comprehension returning [] on any decode error (mirrors load_rule_window_state).
save_delivery_failure builds {ts, decision_id, state_key, domain, entity, reason}
from a signature object.

Co-Authored-By: Rooty

Co-Authored-By: Rooty

Co-Authored-By: Rooty

Co-Authored-By: Rooty

Co-Authored-By: Rooty

Co-Authored-By: Rooty

The cell guard's NATS arm was an exact-string denylist of three
spellings, open by construction: any unlisted spelling (a portless
host, a trailing slash, a bare IP) passed and then connected to the
live bus, since nats-py fills in the default port itself. Replace it
with a port-based predicate that mirrors nats-py's own resolution, so
the arm is closed by construction like the Redis arm already was. A
malformed port refuses rather than raising or falling open.

Also: type check_test_cell's config param as AugurConfig instead of
Any, and fix redis_db's non-numeric-path check (isdigit accepts
Unicode superscript digits that int() then rejects; isdecimal doesn't).

Co-Authored-By: Rooty

tests/test_cell_guard.py only exercises check_test_cell() as a pure
function; nothing proved the interlock is actually consulted before
integration fixtures run. Deleting autouse=True from the session
fixture would leave the rest of the suite green while the guard
silently stopped protecting the live cell.

Adds a pytester-driven replay of the fixture's wiring shape (asserts
a live-cell config aborts with exit code 2, zero tests run, across
several live spellings including portless nats://localhost and Redis
db 0), plus a direct attribute check on the real fixture object's
scope/autouse marker — the assertion that actually fails if autouse
is removed.

Co-Authored-By: Rooty

Classifies every augur:* key and, on --confirm, deletes learned-policy,
session-scoped, and synthetic keys so faculties revert to config defaults.
Keeps real perception baselines and user-taught semantic memories; unknown
keys are kept and flagged. Dry-run by default; refuses non-db-0 without
--force.

Co-Authored-By: Rooty

Co-Authored-By: Rooty

Co-Authored-By: Rooty

Co-Authored-By: Rooty

Co-Authored-By: Rooty

Co-Authored-By: Rooty

Co-Authored-By: Rooty

Registration is inert and must never crash the user's dialogue. A Redis
failure at startup now logs and continues (the session degrades to
non-learnable) instead of taking the console down before the greeting.

Co-Authored-By: Rooty

The six perception-injecting drivers run against the live stack, so their
injected sessions must be explicitly non-learnable, not merely unprovenanced.
Each now records augur:session:meta with origin="synthetic" before it injects,
and an AST guard fails if a perception driver ever omits the mint.

Co-Authored-By: Rooty

Introduce the provenance primitives: LearnContext (session id, learnable,
origin) and a single fail-closed resolve_learn_context that is_learnable_session
delegates to, plus a top-level session_id on the nexus and foreseen payloads so
an event's session is available downstream. Add the learned_write /
non_learning_write decorators with an OFF/REPORT/ENFORCE mode, and a CL10 AST
guard requiring every Redis-mutating function to carry a marker. All inert.

Co-Authored-By: Rooty

Mark the 24 non-learning writers (bookkeeping, audit, and the filter-at-read
group) with @non_learning_write and the 38 learned writers with @learned_write.
Make the learned_write ctx an optional keyword — lenient under OFF/REPORT,
mandatory under ENFORCE — so the existing storage tests are unchanged while
production cannot forget provenance where it is enforced. CL10 backlog is empty.

Co-Authored-By: Rooty

Resolve provenance at all 54 production call sites and hand it to the learned
writers: the event's session for Vigil/Praesagium/Limen, the open prediction's
origin for resolves, the reflection's session for tuning, the dialogue session
for teaching, and LearnContext.system() for bootstrap and operator writes. A
CL13 AST guard fails if any learned-write call omits ctx. No behaviour change.

Co-Authored-By: Rooty

Under ENFORCE, exclude non-learnable sessions where records are read back into
learning: the cross-session feedback pool and the Imperator reflection
read-model. Initialize the mode from AUGUR_PROVENANCE_MODE (off|report|enforce,
default off) at import, so a single env var activates enforcement across every
faculty and an unknown value fails safe to off.

Co-Authored-By: Rooty

Updates the requirements on [fakeredis](https://github.com/cunla/fakeredis-py) to permit the latest version.
- [Release notes](https://github.com/cunla/fakeredis-py/releases)
- [Commits](cunla/fakeredis-py@v2.36.2...v2.37.0)

---
updated-dependencies:
- dependency-name: fakeredis
  dependency-version: 2.37.0
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>