0.8.2 (part A): Activity Monitor, lint unproven state, fuzz grid thresholds, installer hint

[BitmapAsset]

Part A of 0.8.2. No version bump, tag or release here; CHANGELOG entries sit under ## 0.8.2 - unreleased.

1. Activity Monitor (probbit monitor STRAND --open / --serve)

A compact, read-only page for one individual: a sortable, filterable table of the persona's declared traits, moods and (when present) drive channels, event-history traces, Δ against exactly the preceding event (a gap when its document is missing), and an inspector with the full distribution, units, range, source field and the event's why. Hold/follow, and separate states for replay divergence, stale receiver age, held view, paused/retired strand, persona-confidence hold, engine-unreleased trait and lost connection. Works for any persona read from the strand. Embedded in the binary, no external assets. Additive layout metadata caps.prediction_error (from expectation.pe_cap). The terminal view is unchanged.

• Tests: 13 monitor unit tests (monitor::tests, incl. drive_metadata_preserves_declared_scales), 10 integration tests (probbit-cli/tests/monitor.rs), real-browser suite probbit-cli/tests/monitor_browser.mjs (run by the wasm job): sort/filter/search, keyboard, generic and traits-only personas, hold vs unreleased, declared caps, append, hold/follow, stale clock, pause/retire, disconnect/reconnect, missing documents, checkpoint bounds, identity reset, narrow layout.

2. persona lint: "unproven" state (exit 3)

When prove leaves a rule unknown and the default fuzz search finds no counterexample, lint used to report ok: true and exit 0. Now the document gains "unproven": [{id, rule, seeds_unknown, searched}], ok is false, stderr says "unproven, not broken", and the exit code is 3. 0 = every rule held or proved; 1 = unresolved contradiction or a counterexample (wins over 3); 2 = bad input, as before. In-repo callers: no script, Python, npm, MCP or CI caller of persona lint; tests and docs (README, docs/agents.md, docs/persona.md §5.5/§5.6 and the command table) updated.

• Tests: lint_reports_an_unknown_unbroken_rule_as_unproven (unknown + unbroken → 3 with the list; proved → 0 with unproven: []; broken next to unproven → 1, both reported); lint_with_rules_proves_then_fuzzes_the_unknown_ones also asserts unproven: [].

3. Fuzz grid includes the thresholds

Every number literal in the rule's when and the habits' when on a number input is added to the search grid (as a fraction of that input's max, after the given points, each once) for persona fuzz, MCP probbit_persona_fuzz and lint's search; search.grid reports it. No schema change. Personas whose rules and habits read no off-grid thresholds get the same search byte for byte (the pinned fuzz goldens pass unchanged).

• Test: fuzz_grid_includes_the_thresholds_of_the_rule_and_the_habits. A rule in force only at stakes 0.7 to 0.8 (between the default points 0, 0.5, 1) with --beam 0: grid [0,0.5,1,0.7,0.8], counterexample [{"stakes":0.7}] (before this change: grid [0,0.5,1], none found). Also covers dedup of an on-grid threshold and a max-100 input (habit 65 and rule 42 → 0.65 and 0.42).

4. install.sh PATH hint

When the install directory is not on PATH, the hint names the rc file of the user's shell: zsh ~/.zshrc; bash ~/.bash_profile on macOS, ~/.bashrc elsewhere; any other shell ~/.profile. Still POSIX sh. install.ps1 has no rc-file hint (it offers -AddToPath), so it is unchanged.

• Test: scripts/bench/test_install_sh.sh step 8 checks five SHELL values (passes locally under sh and dash); scripts/tests/test_install_contract.py unchanged and passing.

5. Docs

CHANGELOG ## 0.8.2 - unreleased (Added / Changed / Fixed); README monitor section keeps the no-strand demo line; cheatsheet monitor rows updated.

Local: cargo test --workspace --locked 290 passed, 0 failed, 1 ignored; release build --locked; monitor browser suite ok; Python contracts 44 OK; npm wrapper 13 pass. MSRV 1.78 is checked by the msrv job.