navigator.clipboard.readText() only exists in a secure context, so on plain http
over a LAN address it is absent, not merely refused — the previous one-click
paste could never work there. A keyboard paste into a focused textarea is an
ordinary user gesture and is not gated, so paste now opens a dialog: the YAML
goes in a textarea, prefilled only when the browser allows the read.
Nothing is written until Apply, and Apply stays disabled while the text does not
parse, so a typo cannot be clicked through. A live line names the classes the
block matched, the container flags it would flip and the classes it would ignore.
Esc closes and returns focus to the button; Enter is left to the textarea,
because YAML needs newlines.
planPaste() produces that review line and applyPasted() does the writing, keeping
the selected-classes-only rule and the single batched container PATCH.
Copy now emits YAML instead of the one-line "class <name> conf ..." text: one
block per selected class, fields named after the row labels, globals resolved
into the block, a name quoted only when YAML would misread it.
Paste reads the clipboard back and fills the overrides for the classes selected
in that modal, naming any it ignored — an override for a deselected class would
otherwise ride along in class_params unseen. container diffs go out as one
PATCH /projects/{id} and revert together if the save is rejected.
Parsing is parseClassYaml, a strict subset reader for exactly what Copy emits,
so no YAML dependency: any malformed line throws with its line number and
applies nothing. Values must look like numbers, not merely start like them —
parseFloat would have taken "0.5abc" as 0.5.
Both are real buttons with Lucide icons plus a status line reporting the
outcome, and both live in the shared table so the mass modal carries them too.
Track 5 Frames becomes Track Forward with a 1-100 count input, remembered in
localStorage (default 5). A separate -> End action runs to the last frame
behind a confirm and can be cancelled mid-flight.
Frames that already hold a shape of the tracked class are skipped instead of
re-annotated: hidden and dimmed shapes count as present. Skips are reported in
the result banner and marked with a dot plus a class name on the filmstrip
thumbnail. frames() now returns the class ids each frame holds, the same query
row, no migration.
Chaining and seed growth stay out: every frame is still seeded from the drawn
box only.
Reverts 1530bbc and 9c95a4a. Track 5 Frames sends the drawn box unchanged to
all five frames - no chaining, no growth. Both were rejected: chaining let a
wrong read on one frame poison the next, and growth was an attempt to follow a
moving object that the user did not want.
REQ-189, ui-spec and tasks.md return to the b71f1d4 wording.
The chained attempt (1530bbc) fed each run's result into the next frame's
seed, so a wrong read on one frame carried into the ones after it. Runs are
independent again: every frame is seeded from the drawn shape, frames 2-5
with it grown x1.5 about its centre for motion tolerance. REQ-189, ui-spec
and tasks.md updated to the independent contract.
trackForward sent the identical box to all five target frames. On batch 19
(extracted at 0.5 fps - one frame every 2 s) a truck entering the frame
outruns a static seed, so the action either missed it or grabbed the wrong
part.
- geometry handling lifted to module-level shapeBox(), reused for the drawn
shape and for each assist result
- first frame still uses the drawn box exactly; frames 2-5 are seeded with
the previous run's shape grown x1.5 about its centre (SEED_GROWTH)
- a frame SAM3 refuses keeps the last good seed for the next attempt
- REQ-189 rewritten to the chaining contract, including the accepted cost:
a bad read can carry forward; this is not tracking, no motion model
trackForward read geometry.coordinates, a key the backend never emits, and
had no bbox branch - the action silently no-op'd for every shape (flagged
dead in docs/audit-2026-08-07.md:103).
- bbox points used as-is, polygon reduced to its bounding box
- guards speak instead of returning silently
- a frame SAM3 refuses no longer aborts the remaining frames
- a press during an in-flight run is ignored (would double-write)
- REQ-189 added; non-goal requirements.md:22 amended to keep exemplar
propagation out while admitting this one-shot hand-off
The width:100% table plus the new MaxBox column let the Class column absorb
the slack, stranding the Container checkbox away from its class. One block
per class now: name + Container checkbox on the first line, Conf/IoU/MinBox/
MaxBox inputs in a wrapping grid below. Behavior identical - KEYS,
buildClassParams, compose() copy line, toggleContainer revert, aria labels.
- AutoAnnotateModal: render merges last full-set detections for
non-exemplared classes with conditioned results for exemplared ones
(selected-not-exemplared filter); drawing an example no longer clears
the other classes' detections
- Run Preview now always refreshes full-set first (exemplars: []) then
re-runs exemplared classes
- docs: REQ-182 amended (merge semantics), design/ui-spec/tasks synced
- ClassParamsTable: Copy button above per-class overrides table —
clipboard text, one line per class (class/conf/iou/minbox/container),
effective values (override else global), present in both the
per-batch and mass auto-annotate modals via the shared component
- clipboard.js: extract legacyCopy + copyText from ArchiveControls
(insecure-context execCommand fallback), ArchiveControls now imports it
- docs: REQ-186 added; design/ui-spec/tasks synced
- REQ-181: class_params {name: {threshold?, iou_threshold?, min_box_frac?}}
on /preview and /autolabel, per-class override table in both modals;
empty overrides take the unchanged global path
- REQ-180: x button on each review sidebar class row clears that class on
the current frame only via bulk-delete, no confirmation
- includes REQ-178 empty date-folder cycle fix (archive_index.py)
- Auto-generate model names: {arch}-{labelType}-{epochs}ep-{classNames}-{YYYYMMDD}
- Add PATCH /api/models/{id}/rename endpoint
- Inline rename UI on Models & Training page
- Download filename uses model name instead of v{N}
- DB migration: add name column to model_versions
- Update all docs to reflect new naming convention
asus
Andrew-AAAA