Stabilize suggestion acceptance and passive calibration

[RMANOV]

What changed

• make Tab accept only the exact visible completion
• keep Space, Enter, punctuation, and digits as deterministic typed-word boundaries
• preserve literal punctuation, including !
• reject mixed-version ghost updates and bound repeated rejected suggestions in process memory
• add an isolated passive calibration path with owner-thread safety
• refresh PyO3/Rand/Crossbeam dependencies
• keep internal diagnostic guidance out of the public tree

Why

The input path could diverge across clients and, under a stale suggestion race, display or commit text different from the visible candidate. This makes the acceptance contract deterministic and keeps learning/calibration isolated from the typing hot path.

Validation

• Python: 76 passed
• full Rust workspace: passed
• clippy: passed
• cargo audit: no vulnerabilities (allowed warnings only)
• public diff: no newly added Markdown or private artifacts

[Copilot]

Pull request overview

This PR makes suggestion acceptance deterministic (Tab accepts only the visible completion; boundaries like Space/Enter/punctuation/digits remain typed-word boundaries), adds a fail-isolated passive calibration collector + Phase-A calibration tooling, and introduces privacy-guarded keystroke tracing to diagnose input-path regressions without leaking content by default.

Changes:

• Enforce Tab-only acceptance semantics and harden boundary handling to prevent “display/commit mismatch” and delimiter/preedit races.
• Add passive calibration collection (O1 shim) plus Phase-A harness (logging, validity selection, analysis gate, sweeps/watchdog, reconciliation report) with isolation/guardrails.
• Add keystroke diagnostic tracing (OFF/structural/full) with strict privacy constraints, plus session-scoped ghost rejection memory to suppress repeatedly rejected completions.

Reviewed changes

Copilot reviewed 32 out of 34 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/test_o1_shim.py	Adds focused tests for the passive calibration collector (DB schema/privacy, kill-switch, failure isolation, adapter integration).
tests/test_o1_phasea.py	Adds pytest surface for Phase-A self-tests and key invariants (scope firewall, corpus pinning, guards).
tests/conftest.py	Ensures repo root is on sys.path for consistent imports under pytest.
smartkey-debug	Adds a small CLI wrapper to launch the keystroke debug tool.
shim_report.py	Adds reconciliation reporting that shares the canonical row selector with analysis.
phase_a/validity.py	Introduces canonical campaign row selection + exclusion reasons shared across tools.
phase_a/sweep.py	Adds daily sweep receipt + watchdog logic for operational gating signals.
phase_a/selftest.py	Adds an end-to-end self-test suite for the Phase-A harness and gate reachability.
phase_a/paths.py	Adds artifact isolation guardrails and corpus pin verification utilities.
phase_a/harness.py	Adds SQLite event logging schema + HMAC context hashing and logger API.
phase_a/freqmodel.py	Adds a pure n-gram frequency model used as the measured predictor in Phase-A.
phase_a/corpus_replay.py	Adds a corpus-replay driver to generate synthetic calibration events for lab gating.
phase_a/constants.py	Defines shared thresholds, scope/firewall lines, and gate constants.
phase_a/analyze.py	Adds offline analysis and PASS/FAIL/INCONCLUSIVE gate using isotonic calibration and bucketed scoring.
phase_a/init.py	Exposes the Phase-A public surface and documents the measured scope/firewall.
ibus/test_smartkey_debug.py	Adds integration and privacy tests for the keystroke trace (OFF/structural/full behavior, guardrails, budgets).
ibus/test_rejection_memory_adapter.py	Adds adapter-level tests for session-scoped ghost rejection feedback into core memory.
ibus/test_accept_backspace.py	Extends adapter integration tests for backspace forwarding and Tab-only accept/boundary behavior.
ibus/smartkey_debug.py	Implements keystroke tracing + CLI with privacy levels, repo-dir refusal, purge/wipe, and deferred flushing.
ibus/o1_shim.py	Implements the passive calibration collector with hot-path isolation and background SQLite writing.
ibus/main.py	Gates DEBUG logging on SMARTKEY_DEBUG=full to keep structural tracing content-free.
crates/smartkey-py/src/lib.rs	Extends PyO3 core surface with O1 snapshot/resolve/abandon + debug state + rejection recording APIs.
crates/smartkey-py/Cargo.toml	Bumps PyO3 dependency version for the Python extension.
crates/smartkey-core/tests/virtual_typer.rs	Updates tests around BG phonetic mapping expectations (“zdrawej” cases).
crates/smartkey-core/src/rejection_memory.rs	Adds bounded, process-local rejection memory with LRU eviction and suppression threshold.
crates/smartkey-core/src/o1_shim.rs	Adds Rust-side n-gram-only snapshot implementation with parity tests vs Phase-A model semantics.
crates/smartkey-core/src/master_loop.rs	Integrates rejection suppression, ghost attribution, passive calibration plumbing, and debug state wiring.
crates/smartkey-core/src/lib.rs	Exposes new o1_shim and rejection_memory modules.
crates/smartkey-core/src/keymap.rs	Adjusts BG traditional phonetic mapping and adds targeted regression tests.
crates/smartkey-core/src/input.rs	Enforces Tab-only acceptance and removes boundary auto-accept paths; hardens delimiter ordering and ghost behavior.
crates/smartkey-core/src/ensemble.rs	Wires engine-level O1 n-gram snapshot helpers for the passive calibration tap.
Cargo.lock	Updates dependency lockfile (PyO3/crossbeam/rand).
.gitignore	Adds ignore patterns for debug traces and private inputs.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 930b34cc94

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Scope watchdog sweeps to the selected campaign

When real and synthetic data (or two real campaigns) share a DB, this query treats every sweeps row as coverage for the campaign being analyzed. run_sweep(..., synthetic=True) filters its event population but writes an indistinguishable sweep row, and the schema has no run/synthetic field to filter on; therefore synthetic replay sweeps can make a live campaign appear to have zero missed watchdog days and allow a PASS despite no live watchdog execution. Store campaign/scope on sweeps and filter both analysis and watchdog queries by it.

Useful? React with 👍 / 👎.