Reconcile the staged-projection and Diagnostics feature work with the CI
quality-gate changes from origin/main. The origin side of every conflict was
lint-only, so the feature imports were kept and the ruff fixes re-applied:
- keep the branch's added imports (timefmt, activity_readout, projection
ladder helpers) and drop the unused ones ruff reported
- restore `noqa: E402` on the post-bootstrap imports in fenris-monitor and
keep the `noqa: F401` re-exports in the TUI that tests assert on
- make the TUI monitoring-glyph test clock-relative instead of pinned to a
fixed date
Record host read and write commands, controller busy time, error log
entries, warning and critical temperature time, and thermal management
transitions from the existing smartctl -a -j acquisition. Schema 7 adds
nullable columns to samples so legacy rows read as unknown, and the
Diagnostics panel shows each counter with its change.
Toggle a Diagnostics panel with i, built on the Activity readout. It swaps
in for the Drive pane and shows the decoded critical_warning, power-on
hours, power cycles, unsafe shutdowns, media errors and available spare
with their change over the selection, and per-hour (Day) or per-day
(History) temperature min/avg/max, active/idle/powered-off/unknown split
and coverage. History bars draw reads beside writes while it is open, and a
set critical_warning shows as a badge in the endurance outlook without
affecting the projection. Counter deltas are withheld across a controller
segment change.
Declare every dashboard key as a Binding on the widget that owns it, and
generate the footer, activity tool chips and help screen from those
bindings so they cannot drift. The footer collapses to "? help · q quit"
under 80x24.
Add Home/End and PgUp/PgDn (one week) navigation, mouse-wheel selection on
the graphs, click-to-focus on every panel, and clickable footer chips.
Tab and Shift+Tab now visit only the dashboard panels.
Move activity data assembly out of tui.py into activity_readout, behind
readout(conn, selection, now). Dashboard and fenris-monitor enable now show
local time with a timezone label, inspected points keep a secondary UTC line,
and the legacy UTC-relabelled day and hour branches are gone. Time formatting
(including freshness_age_human) lives in timefmt. Hour offsets now render as
+05:30 to match day labels.
Add a b key and a no-baseline banner that open a form for rated TBW and an
optional source URL. The value is persisted as an unverified override through
fenris-monitor baseline set under the existing polkit action, and the outcome
is reported synchronously. Verified overrides stay CLI-only (ADR 0012 section 7).
Report a projection stage, an evidence ladder and the sustained-regime
write rate from compute_projection, per ADR 0012. A provisional lifespan
appears after 3 observed hours and carries the hours observed and a
short-horizon spread; the complete-local-day condition becomes a fact
instead of a gate, and the write rate replaces the lifespan when no
baseline applies. The TUI and CLI render the stage, ladder count and
full ladder, and drop their own two-sample pre-gates.
Use lifetime written bytes from the current segment's newest published
sample in the headline formula, select the Percentage-Used-implied
baseline after two increments in the segment (otherwise show the
too-coarse fact), and judge horizon agreement across existing horizons.
Make the vacuous projection tests assert unconditionally and add
coverage for implied-baseline gating and single-horizon agreement.
The _add_unattributed_bytes function was adding the same cross-hour
delta to both days when the interval spanned midnight, violating the
spec requirement to preserve measured volume once as shared boundary
evidence.
Store the unattributed bytes only on the day where the interval starts
(the earlier day). This ensures the total across both days equals the
actual delta without duplication.
Closes#88 (cross-day delta duplication portion)
Add query_local_day_history() with evidence-limit metadata to local_day.py,
including LocalDayHistoryEntry dataclass that annotates each summary with
detail_available and derived_from_surviving flags. Document the 14-day
detail vs indefinite-summary retention policy in both local_day.py and
pruning.py.
Fix query_local_day_summary() to convert SQLite integer booleans to Python
bool, and add deduplication guard to the midnight-spanning hour logic in
derive_local_day_summary().
Add 17 lifecycle tests covering all acceptance criteria:
- AC1: Summaries survive sample pruning, evidence limits visible
- AC2: Boundary anchors preserved before pruning
- AC3: Timezone survives system-timezone change
- AC4: Legacy UTC summaries labelled incomplete
- AC5: Idempotent pruning and repair
- AC6: Volume conservation (no double counting)
- AC7: Real temporary stores
- AC8: User-visible transition from recent to aged
- AC9: Retention policy documented
All 775 tests pass.
xavierk
xavierk