cjm-transcript-correction-core 0.0.10__tar.gz → 0.0.30__tar.gz

{cjm_transcript_correction_core-0.0.10/cjm_transcript_correction_core.egg-info → cjm_transcript_correction_core-0.0.30}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cjm-transcript-correction-core
-Version: 0.0.10
+Version: 0.0.30
 Summary: A frontend-agnostic core for the transcript correction workflow — the first downstream graph-extension core; composes the graph-storage capability worker into a headless pipeline that applies unified, non-destructive corrections (text, punctuation, segmentation) as a supersede-able overlay on a committed decomposition spine, recomputes the review worklist from deterministic signals, and exposes a CLI as its first driver.
 Author-email: "Christian J. Mills" <9126128+cj-mills@users.noreply.github.com>
 License: Apache-2.0
@@ -14,8 +14,11 @@ Requires-Dist: cjm_substrate>=0.0.61
 Requires-Dist: cjm_context_graph_primitives>=0.0.15
 Requires-Dist: cjm-capability-primitives>=0.0.12
 Requires-Dist: cjm-context-graph-layer>=0.0.11
-Requires-Dist: cjm-transcript-graph-schema>=0.0.7
+Requires-Dist: cjm-transcript-graph-schema>=0.0.11
 Requires-Dist: numpy
+Provides-Extra: test
+Requires-Dist: soundfile; extra == "test"
+Requires-Dist: pytest; extra == "test"
 Dynamic: license-file
 
 # cjm-transcript-correction-core
@@ -27,21 +30,34 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 ## Modules
 
 - **`cjm_transcript_correction_core.__init__`**
+- **`cjm_transcript_correction_core.cleanread`** — The clean read — L1 of the reading ladder (design 6752db0a; the subtraction
 - **`cjm_transcript_correction_core.cli`** — The CLI driver — the correction core's first (and currently only) frontend. run <decomp-manifest> corrects the committed spine in the decomp graph DB, pointing the graph worker at that shared DB via load-time config, with optional session resume/reopen; review runs the interactive text-correction loop (the cross-transcriber diff is intra-graph since stage 5).
 - **`cjm_transcript_correction_core.graph`** — The correction overlay's graph I/O: targeted (scale-shaped) reads of a committed spine via the graph-storage query action, construction of Correction / CorrectionSession nodes + CORRECTS / SUPERSEDES / DERIVED_FROM / REVIEWED edges, the in-core effective-spine projection (layer-0 + applied corrections), and commit through the job queue. Hand-rolled (revolution-1) = direct CR-18 spec material; append-only on layer-0 (never update/delete a Segment).
 - **`cjm_transcript_correction_core.journal`** — Live append-through for the correction verbs — the workflow journal's domain half.
 - **`cjm_transcript_correction_core.launch`** — The shared launch surface every correction shell drives through: the
 - **`cjm_transcript_correction_core.models`** — Overlay data shapes for the transcript-correction workflow: the Correction / CorrectionSession graph nodes + their relation registry, the read view of a committed spine segment, the worklist item, run configuration, and the correction run manifest (proto-bundle that chains decomp -> correction).
 - **`cjm_transcript_correction_core.pipeline`** — The headless correction workflow: load a decomp run manifest, resolve the shared graph DB, start/resume/reopen a CorrectionSession, recompute the worklist from deterministic signals + persisted review state, run the D14 empty-segment prune (first operation), and record a chainable correction run manifest — with a cheapest-form HITL approval seam.
+- **`cjm_transcript_correction_core.respine_transfer`** — The chunk-scoped transfer a chunk respine calls (work item 7a5e9c84; ruling 0b4d5cfa (5),
 - **`cjm_transcript_correction_core.signals`** — Pure deterministic Tier-1 signal functions (no capability calls): empty-segment detection, bidirectional boundary punctuation/capitalization heuristics, forced-alignment coverage flags, positional cross-transcriber diff, phonetic + edit-distance variant clustering, and the event-proposal overlay (leg 4: the finetuned detector's spans anchored onto the spine). The worklist is recomputed from these each session; revolution-1 builds ZERO new capabilities.
+- **`cjm_transcript_correction_core.spans`** — Spans — the overlay-proposal lane's domain half (design bbf8bafd, extending the
 - **`cjm_transcript_correction_core.spine`**
 - **`cjm_transcript_correction_core.state`** — Sidecar view-state + spine-picker helpers — the correction TUI's pure
 - **`cjm_transcript_correction_core.strata`** — Strata — the filtering lane's domain half (DECs 304fd984 + 9d4c0a38; work items
 
 ## API
 
+### `cjm_transcript_correction_core.cleanread`
+
+- `clean_read` _function_ — L1 — the clean read (pure). Kept lines in spine order; each carries its
+- `clean_read_pack_read` _function_ — The `read` record an L1 pack carries: the layer, the marker, the
+- `clean_read_segments` _function_ — Project the clean read back onto SpineSegments (clean text, same
+- `clean_read_summary` _function_
+- `repeat_survivor` _function_ — Locate the last repeating unit of a word-repeat span (pure): the
+- `subtract_spans` _function_ — Cut the ranges out of the text, leaving `marker` at each seam (pure).
+
 ### `cjm_transcript_correction_core.cli`
 
+- `attention_command` _function_ — Execute `attention`: the walk-lane ATTENTION TIER (item 3758f6cb, ruling f400d2c3)
 - `bench_command` _function_ — Execute the `bench` subcommand: the reserved-tail verdict join.
 - `build_parser` _function_ — Build the CLI parser (subcommands: run, review).
 - `commit_wordless_transfer` _function_ — COMMIT a planned transfer — the engine's second half. One
@@ -49,6 +65,7 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `extract_command` _function_ — Execute the `extract` subcommand: fold the gated overlay into a manifested dataset.
 - `filter_confirm_command` _function_ — Execute `filter-confirm`: the HEADLESS HITL worklist (bc8dbbdd pass-1
 - `filter_ingest_command` _function_ — Execute `filter-ingest`: validate proposer rows against their pack and
+- `filter_merge_command` _function_ — Execute `filter-merge`: fold several proposal sets over one spine into
 - `filter_pack_command` _function_ — Execute `filter-pack`: write one spine window's effective text-bearing
 - `gate_command` _function_ — Execute the `gate` subcommand: show or assert per-spine extraction gates.
 - `load_capabilities` _function_ — Discover manifests + load each capability, passing per-capability config (CR-2 caller-wins).
@@ -62,6 +79,11 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `run_command` _function_ — Execute the `run` subcommand: correct a decomp manifest's committed spine.
 - `run_extract` _function_ — The extract fold on an ALREADY-OPEN graph seat (flywheel build leg 2,
 - `scan_command` _function_ — Execute `scan-mishomed`: flag authoritative FA words stranded outside
+- `span_confirm_command` _function_ — Execute `span-confirm`: the HEADLESS HITL worklist over one span
+- `span_ingest_command` _function_ — Execute `span-ingest`: resolve a proposer's quoted-word rows against
+- `span_lexicon_command` _function_ — Execute `span-lexicon`: the pattern tier (design bbf8bafd (g)) — bare
+- `span_merge_command` _function_ — Execute `span-merge`: fold several span sets over one spine into ONE
+- `span_pack_command` _function_ — Execute `span-pack`: the filter pack's lines as a SPEECH-OVERLAY pack
 - `stats_command` _function_ — Execute the `stats` subcommand: flywheel accounting over the shared graph.
 - `transfer_command` _function_ — Execute `transfer-wordless`: replay wordless event inserts (and speaker-
 - `wordless_donors` _function_ — The EFFECTIVE wordless layer of a spine: labeled, effectively wordless
@@ -73,6 +95,7 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `active_speaker_assignments` _function_ — Project the ACTIVE speaker assignment per segment (latest-wins).
 - `active_speech_overlays` _function_ — The surviving speech-overlay corrections (supersession applied; pure).
 - `aggregate_session_purposes` _function_ — Fold sessions into a per-source purpose mix (pure; d915d545 picker rung).
+- `annotate_retired` _function_ — Mark spine rows with their retirement state (pure).
 - `apply_chunk_inserts` _function_ — Synthesize inserted chunks into the effective spine (DEC 3d3fa2a8).
 - `apply_time_nudges` _function_ — Apply timing corrections onto segment times (latest-wins per edge).
 - `bench_event_proposals` _function_ — The reserved-tail bench join (leg 4, DECs 8e05b87b + 8cf12c22) — pure.
@@ -111,6 +134,8 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `correction_stats` _function_ — Fold one Source's ACTIVE overlay into flywheel-accounting counts (pure).
 - `corrections_to_edits` _function_ — Map this core's Correction payloads onto the layer's spine-edit vocabulary.
 - `count_source_segments` _function_ — Count a Source's segments server-side under its chosen rendition + skeleton (typed count mode).
+- `default_spine` _function_ — Which spine opens by default (pure; ruling a7617bd4 rule (c)): preference is a
+- `discharge_mark` _function_ — Discharge an open mark BY THE CORRECTION THAT ADDRESSED IT — the closure
 - `extract_spine_dataset` _function_ — Fold ONE spine's overlay into its v1 insert-span dataset slice (pure).
 - `fa_words_for_transcript` _function_ — One transcript's FA words in source coordinates (the scan-mishomed join,
 - `find_active_text_correction` _function_ — Single-segment convenience over the batch read (cross-session; latest wins).
@@ -143,6 +168,7 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `set_session_status` _function_ — Update a session's status + updated_at.
 - `skeleton_hash_for` _function_ — Resolve a skeleton selector to the chosen spine's HASH (pure) — the gate's
 - `source_audio_segment_ids` _function_ — The Source's coarse spine (one small typed read; ordered by index).
+- `speaker_assignment_sources` _function_ — Where each speaker Entity has been assigned — the picker's COLLECTION
 - `speech_overlay_spans` _function_ — Fold the ACTIVE speech overlays into extraction-facing span records (pure).
 - `spine_where_for` _function_ — Resolve a skeleton selector against the observed spine set (pure).
 - `start_session` _function_ — Create + commit a new CorrectionSession node.
@@ -186,8 +212,22 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `run_correction` _function_ — Correct every source in a decomp run manifest (prune + worklist surfacing).
 - `run_review` _function_ — Interactive review pass over a decomp manifest's flagged worklist (text corrections).
 
+### `cjm_transcript_correction_core.respine_transfer`
+
+- `chunk_respine_transfer` _function_ — The registered chunk-scoped transfer (the entry point decomp-core discovers).
+- `group_speaker_rows` _function_ — Fold per-segment rows into turn-shaped commits (pure): contiguous same-entity
+- `live_neighbour` _function_ — The live segment right before the chunk — the flank a donor that starts before
+- `load_segments_by_ids` _function_ — Segments by id — the ONE read that bypasses the live predicate on purpose: the
+- `plan_speaker_carry` _function_ — Which new segment takes which speaker (pure; 4a7ec4f8 (2)/(3)): a segment whose
+- `speaker_spans` _function_ — The per-span speaker map the carry reads (pure).
+
 ### `cjm_transcript_correction_core.signals`
 
+- `attention_boundary_marks` _function_ — The boundary signals of the attention tier (item 3758f6cb signals 1+3 and the folded
+- `attention_divergence_marks` _function_ — The two-transcriber disagreement signal of the attention tier (item 3758f6cb signal
+- `attention_fa_marks` _function_ — The forced-alignment signals of the attention tier (item 3758f6cb signal 1), pure.
+- `attention_marks` _function_ — Compose the attention tier (item 3758f6cb): every enabled signal's mark rows,
+- `attention_speaker_marks` _function_ — The next-speaker-change landmark of the folded item 65791933, pure: `speaker-change`
 - `boundary_punct_caps_flags` _function_ — Bidirectional boundary punctuation/capitalization heuristics (in-segment only).
 - `cluster_variants` _function_ — Cluster word variants by phonetic key + edit distance (fix-one-fix-all).
 - `compute_signal_flags` _function_ — Combine all deterministic Tier-1 signals into per-segment flags.
@@ -200,14 +240,36 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `speaker_turn_proposals` _function_ — Dominant diarization cluster per segment — the assign lane's proposal paint.
 - `variant_divergence` _function_ — Within-segment cross-transcriber divergence (stage 5: intra-graph).
 
+### `cjm_transcript_correction_core.spans`
+
+- `bench_span_proposals` _function_ — Derive the span verdicts (design bbf8bafd (f), the bench_filter_proposals
+- `build_span_pack` _function_ — Build the span proposer's input: the filter pack's numbered lines with
+- `find_token_span` _function_ — Resolve quoted words to WHOLE word tokens of a line (pure): the quote is
+- `is_span_pack` _function_
+- `lexicon_span_rows` _function_ — The LEXICON TIER (design bbf8bafd (g)): a bare um / uh token is a
+- `load_span_proposal_sets` _function_ — Every SPAN proposal set for a source (and optionally one spine), newest
+- `locate_span_tokens` _function_ — Where a proposal's words sit on the CURRENT line, as a character range
+- `merge_span_proposals` _function_ — Fold several SPAN sets over one spine into one walkable set — the
+- `overlay_label_slate` _function_ — The slate a span pack carries: every label with its gloss and whether a
+- `pending_span_proposals` _function_ — The headless worklist: proposals with NO active overlay carrying their id
+- `render_span_pack` _function_ — Render a span pack as the brief a proposer reads: identity + window, the
+- `render_span_propset_markdown` _function_ — Project a span set for a human: one line per proposal — tier, label,
+- `segment_word_tokens` _function_
+- `snap_span_proposal` _function_ — Resolve a proposal against the CURRENT line at accept (design bbf8bafd
+- `span_proposals_from_rows` _function_ — Resolve validated rows to span proposal-set rows: proposal id, label
+- `validate_span_rows` _function_ — Validate + resolve proposer rows against their span pack — loud on the
+- `write_span_propset` _function_ — Write one SPAN proposal set — the filter set's layout under the span
+
 ### `cjm_transcript_correction_core.spine`
 
 - `ChunkRef` _class_ — Where one Segment's VAD-chunk audio lives: the model-input WAV + the chunk-local span.
 - `SeamRef` _class_ — A source-coordinate audio span across one fine-spine boundary (the g/G
 - `SpineView` _class_ — One Source's effective correction spine, cursor-windowed for the TUI.
+- `layer_speaker_menu` _function_ — Layer the assign-lane digit menu (pure; DEC 774dbe40). Tier 1 = the
 - `list_sources` _function_ — Enumerate the graph's Source nodes (the discovery corpus, 2ce81638).
 - `load_source_slice` _function_ — Decode a source-coordinate slice of the ORIGINAL media to playable samples.
 - `match_sources` _function_ — The --source selector (pure; shared by direct open and the picker's seed).
+- `match_speaker_query` _function_ — Narrow the registry by a typed query (pure; DEC 774dbe40 (4)). Rank 0 =
 - `neighbor_word_bound` _function_ — The adjacent word's FA boundary facing an overlay span (pure).
 - `open_stack` _function_ — Bootstrap the graph capability stack, resolving the db path (2ce81638).
 - `parse_entity_input` _function_ — Parse the new-speaker editor line (pure). A leading `?` marks the entity
@@ -238,11 +300,18 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `build_filter_pack` _function_ — Build the proposer's input: one source window's text-bearing effective
 - `exclude_strata` _function_ — The per-consumer filtered projection (DEC 9d4c0a38): "filtered" is a
 - `load_filter_proposal_sets` _function_ — Every filtering proposal set for a source (and optionally one spine),
+- `materialized_fix_ids` _function_ — The fidelity-edit apply path's materialization: a proposer row whose
+- `materialized_mark_ids` _function_ — Class-family routing's other half: a proposer's mark-family row (an ASR
+- `merge_filter_proposals` _function_ — Fold several proposal sets over ONE spine into one walkable set (design
 - `new_pack_id` _function_ — Generate a unique, sortable pack id.
 - `pack_digest` _function_ — Digest the READ content (source binding + window + numbered segments) —
 - `pending_filter_proposals` _function_ — The headless worklist: proposals with NO live stratum carrying their id and
+- `plan_pack_windows` _function_ — Cut a spine into `count` windows of near-equal TEXT-segment count at
 - `proposals_from_rows` _function_ — Resolve validated rows to proposal-set rows: proposal id, category, source
 - `render_filter_pack` _function_ — Render a pack as the brief a proposer reads: identity + window, the class
+- `render_filter_propset_markdown` _function_ — Project a proposal set for a HUMAN to check against the source in the
+- `resolve_slate` _function_ — Resolve a NAMED class slate (ruling c5b6cf42) to the two vocabularies a
+- `select_span_segments` _function_ — The span-EDIT gesture's resolver (agreed 2026-09-01 on the ch04 confirm
 - `strata_index` _function_ — Segment-keyed view of the live strata (the consumer query's index).
 - `validate_proposal_rows` _function_ — Validate + normalize proposer rows against their pack — loud on the first
 - `write_filter_propset` _function_ — Write one filtering proposal set: `<out_root>/<set_id>/manifest.json` +
@@ -250,4 +319,4 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 ## Dependencies
 
 **Depends on:** `cjm-capability-primitives`, `cjm-context-graph-layer`, `cjm-context-graph-primitives`, `cjm-substrate`, `cjm-transcript-graph-schema`, `numpy`
-**Used by:** `cjm-transcript-correction-qt`, `cjm-transcript-correction-tui`, `cjm-transcription-core`, `cjm-workflow-hub-qt`, `cjm-workflow-hub-tui`
+**Used by:** `cjm-context-graph-projection`, `cjm-transcript-correction-qt`, `cjm-transcription-core`, `cjm-workflow-hub-qt`

{cjm_transcript_correction_core-0.0.10 → cjm_transcript_correction_core-0.0.30}/README.md

@@ -7,21 +7,34 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 ## Modules
 
 - **`cjm_transcript_correction_core.__init__`**
+- **`cjm_transcript_correction_core.cleanread`** — The clean read — L1 of the reading ladder (design 6752db0a; the subtraction
 - **`cjm_transcript_correction_core.cli`** — The CLI driver — the correction core's first (and currently only) frontend. run <decomp-manifest> corrects the committed spine in the decomp graph DB, pointing the graph worker at that shared DB via load-time config, with optional session resume/reopen; review runs the interactive text-correction loop (the cross-transcriber diff is intra-graph since stage 5).
 - **`cjm_transcript_correction_core.graph`** — The correction overlay's graph I/O: targeted (scale-shaped) reads of a committed spine via the graph-storage query action, construction of Correction / CorrectionSession nodes + CORRECTS / SUPERSEDES / DERIVED_FROM / REVIEWED edges, the in-core effective-spine projection (layer-0 + applied corrections), and commit through the job queue. Hand-rolled (revolution-1) = direct CR-18 spec material; append-only on layer-0 (never update/delete a Segment).
 - **`cjm_transcript_correction_core.journal`** — Live append-through for the correction verbs — the workflow journal's domain half.
 - **`cjm_transcript_correction_core.launch`** — The shared launch surface every correction shell drives through: the
 - **`cjm_transcript_correction_core.models`** — Overlay data shapes for the transcript-correction workflow: the Correction / CorrectionSession graph nodes + their relation registry, the read view of a committed spine segment, the worklist item, run configuration, and the correction run manifest (proto-bundle that chains decomp -> correction).
 - **`cjm_transcript_correction_core.pipeline`** — The headless correction workflow: load a decomp run manifest, resolve the shared graph DB, start/resume/reopen a CorrectionSession, recompute the worklist from deterministic signals + persisted review state, run the D14 empty-segment prune (first operation), and record a chainable correction run manifest — with a cheapest-form HITL approval seam.
+- **`cjm_transcript_correction_core.respine_transfer`** — The chunk-scoped transfer a chunk respine calls (work item 7a5e9c84; ruling 0b4d5cfa (5),
 - **`cjm_transcript_correction_core.signals`** — Pure deterministic Tier-1 signal functions (no capability calls): empty-segment detection, bidirectional boundary punctuation/capitalization heuristics, forced-alignment coverage flags, positional cross-transcriber diff, phonetic + edit-distance variant clustering, and the event-proposal overlay (leg 4: the finetuned detector's spans anchored onto the spine). The worklist is recomputed from these each session; revolution-1 builds ZERO new capabilities.
+- **`cjm_transcript_correction_core.spans`** — Spans — the overlay-proposal lane's domain half (design bbf8bafd, extending the
 - **`cjm_transcript_correction_core.spine`**
 - **`cjm_transcript_correction_core.state`** — Sidecar view-state + spine-picker helpers — the correction TUI's pure
 - **`cjm_transcript_correction_core.strata`** — Strata — the filtering lane's domain half (DECs 304fd984 + 9d4c0a38; work items
 
 ## API
 
+### `cjm_transcript_correction_core.cleanread`
+
+- `clean_read` _function_ — L1 — the clean read (pure). Kept lines in spine order; each carries its
+- `clean_read_pack_read` _function_ — The `read` record an L1 pack carries: the layer, the marker, the
+- `clean_read_segments` _function_ — Project the clean read back onto SpineSegments (clean text, same
+- `clean_read_summary` _function_
+- `repeat_survivor` _function_ — Locate the last repeating unit of a word-repeat span (pure): the
+- `subtract_spans` _function_ — Cut the ranges out of the text, leaving `marker` at each seam (pure).
+
 ### `cjm_transcript_correction_core.cli`
 
+- `attention_command` _function_ — Execute `attention`: the walk-lane ATTENTION TIER (item 3758f6cb, ruling f400d2c3)
 - `bench_command` _function_ — Execute the `bench` subcommand: the reserved-tail verdict join.
 - `build_parser` _function_ — Build the CLI parser (subcommands: run, review).
 - `commit_wordless_transfer` _function_ — COMMIT a planned transfer — the engine's second half. One
@@ -29,6 +42,7 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `extract_command` _function_ — Execute the `extract` subcommand: fold the gated overlay into a manifested dataset.
 - `filter_confirm_command` _function_ — Execute `filter-confirm`: the HEADLESS HITL worklist (bc8dbbdd pass-1
 - `filter_ingest_command` _function_ — Execute `filter-ingest`: validate proposer rows against their pack and
+- `filter_merge_command` _function_ — Execute `filter-merge`: fold several proposal sets over one spine into
 - `filter_pack_command` _function_ — Execute `filter-pack`: write one spine window's effective text-bearing
 - `gate_command` _function_ — Execute the `gate` subcommand: show or assert per-spine extraction gates.
 - `load_capabilities` _function_ — Discover manifests + load each capability, passing per-capability config (CR-2 caller-wins).
@@ -42,6 +56,11 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `run_command` _function_ — Execute the `run` subcommand: correct a decomp manifest's committed spine.
 - `run_extract` _function_ — The extract fold on an ALREADY-OPEN graph seat (flywheel build leg 2,
 - `scan_command` _function_ — Execute `scan-mishomed`: flag authoritative FA words stranded outside
+- `span_confirm_command` _function_ — Execute `span-confirm`: the HEADLESS HITL worklist over one span
+- `span_ingest_command` _function_ — Execute `span-ingest`: resolve a proposer's quoted-word rows against
+- `span_lexicon_command` _function_ — Execute `span-lexicon`: the pattern tier (design bbf8bafd (g)) — bare
+- `span_merge_command` _function_ — Execute `span-merge`: fold several span sets over one spine into ONE
+- `span_pack_command` _function_ — Execute `span-pack`: the filter pack's lines as a SPEECH-OVERLAY pack
 - `stats_command` _function_ — Execute the `stats` subcommand: flywheel accounting over the shared graph.
 - `transfer_command` _function_ — Execute `transfer-wordless`: replay wordless event inserts (and speaker-
 - `wordless_donors` _function_ — The EFFECTIVE wordless layer of a spine: labeled, effectively wordless
@@ -53,6 +72,7 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `active_speaker_assignments` _function_ — Project the ACTIVE speaker assignment per segment (latest-wins).
 - `active_speech_overlays` _function_ — The surviving speech-overlay corrections (supersession applied; pure).
 - `aggregate_session_purposes` _function_ — Fold sessions into a per-source purpose mix (pure; d915d545 picker rung).
+- `annotate_retired` _function_ — Mark spine rows with their retirement state (pure).
 - `apply_chunk_inserts` _function_ — Synthesize inserted chunks into the effective spine (DEC 3d3fa2a8).
 - `apply_time_nudges` _function_ — Apply timing corrections onto segment times (latest-wins per edge).
 - `bench_event_proposals` _function_ — The reserved-tail bench join (leg 4, DECs 8e05b87b + 8cf12c22) — pure.
@@ -91,6 +111,8 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `correction_stats` _function_ — Fold one Source's ACTIVE overlay into flywheel-accounting counts (pure).
 - `corrections_to_edits` _function_ — Map this core's Correction payloads onto the layer's spine-edit vocabulary.
 - `count_source_segments` _function_ — Count a Source's segments server-side under its chosen rendition + skeleton (typed count mode).
+- `default_spine` _function_ — Which spine opens by default (pure; ruling a7617bd4 rule (c)): preference is a
+- `discharge_mark` _function_ — Discharge an open mark BY THE CORRECTION THAT ADDRESSED IT — the closure
 - `extract_spine_dataset` _function_ — Fold ONE spine's overlay into its v1 insert-span dataset slice (pure).
 - `fa_words_for_transcript` _function_ — One transcript's FA words in source coordinates (the scan-mishomed join,
 - `find_active_text_correction` _function_ — Single-segment convenience over the batch read (cross-session; latest wins).
@@ -123,6 +145,7 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `set_session_status` _function_ — Update a session's status + updated_at.
 - `skeleton_hash_for` _function_ — Resolve a skeleton selector to the chosen spine's HASH (pure) — the gate's
 - `source_audio_segment_ids` _function_ — The Source's coarse spine (one small typed read; ordered by index).
+- `speaker_assignment_sources` _function_ — Where each speaker Entity has been assigned — the picker's COLLECTION
 - `speech_overlay_spans` _function_ — Fold the ACTIVE speech overlays into extraction-facing span records (pure).
 - `spine_where_for` _function_ — Resolve a skeleton selector against the observed spine set (pure).
 - `start_session` _function_ — Create + commit a new CorrectionSession node.
@@ -166,8 +189,22 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `run_correction` _function_ — Correct every source in a decomp run manifest (prune + worklist surfacing).
 - `run_review` _function_ — Interactive review pass over a decomp manifest's flagged worklist (text corrections).
 
+### `cjm_transcript_correction_core.respine_transfer`
+
+- `chunk_respine_transfer` _function_ — The registered chunk-scoped transfer (the entry point decomp-core discovers).
+- `group_speaker_rows` _function_ — Fold per-segment rows into turn-shaped commits (pure): contiguous same-entity
+- `live_neighbour` _function_ — The live segment right before the chunk — the flank a donor that starts before
+- `load_segments_by_ids` _function_ — Segments by id — the ONE read that bypasses the live predicate on purpose: the
+- `plan_speaker_carry` _function_ — Which new segment takes which speaker (pure; 4a7ec4f8 (2)/(3)): a segment whose
+- `speaker_spans` _function_ — The per-span speaker map the carry reads (pure).
+
 ### `cjm_transcript_correction_core.signals`
 
+- `attention_boundary_marks` _function_ — The boundary signals of the attention tier (item 3758f6cb signals 1+3 and the folded
+- `attention_divergence_marks` _function_ — The two-transcriber disagreement signal of the attention tier (item 3758f6cb signal
+- `attention_fa_marks` _function_ — The forced-alignment signals of the attention tier (item 3758f6cb signal 1), pure.
+- `attention_marks` _function_ — Compose the attention tier (item 3758f6cb): every enabled signal's mark rows,
+- `attention_speaker_marks` _function_ — The next-speaker-change landmark of the folded item 65791933, pure: `speaker-change`
 - `boundary_punct_caps_flags` _function_ — Bidirectional boundary punctuation/capitalization heuristics (in-segment only).
 - `cluster_variants` _function_ — Cluster word variants by phonetic key + edit distance (fix-one-fix-all).
 - `compute_signal_flags` _function_ — Combine all deterministic Tier-1 signals into per-segment flags.
@@ -180,14 +217,36 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `speaker_turn_proposals` _function_ — Dominant diarization cluster per segment — the assign lane's proposal paint.
 - `variant_divergence` _function_ — Within-segment cross-transcriber divergence (stage 5: intra-graph).
 
+### `cjm_transcript_correction_core.spans`
+
+- `bench_span_proposals` _function_ — Derive the span verdicts (design bbf8bafd (f), the bench_filter_proposals
+- `build_span_pack` _function_ — Build the span proposer's input: the filter pack's numbered lines with
+- `find_token_span` _function_ — Resolve quoted words to WHOLE word tokens of a line (pure): the quote is
+- `is_span_pack` _function_
+- `lexicon_span_rows` _function_ — The LEXICON TIER (design bbf8bafd (g)): a bare um / uh token is a
+- `load_span_proposal_sets` _function_ — Every SPAN proposal set for a source (and optionally one spine), newest
+- `locate_span_tokens` _function_ — Where a proposal's words sit on the CURRENT line, as a character range
+- `merge_span_proposals` _function_ — Fold several SPAN sets over one spine into one walkable set — the
+- `overlay_label_slate` _function_ — The slate a span pack carries: every label with its gloss and whether a
+- `pending_span_proposals` _function_ — The headless worklist: proposals with NO active overlay carrying their id
+- `render_span_pack` _function_ — Render a span pack as the brief a proposer reads: identity + window, the
+- `render_span_propset_markdown` _function_ — Project a span set for a human: one line per proposal — tier, label,
+- `segment_word_tokens` _function_
+- `snap_span_proposal` _function_ — Resolve a proposal against the CURRENT line at accept (design bbf8bafd
+- `span_proposals_from_rows` _function_ — Resolve validated rows to span proposal-set rows: proposal id, label
+- `validate_span_rows` _function_ — Validate + resolve proposer rows against their span pack — loud on the
+- `write_span_propset` _function_ — Write one SPAN proposal set — the filter set's layout under the span
+
 ### `cjm_transcript_correction_core.spine`
 
 - `ChunkRef` _class_ — Where one Segment's VAD-chunk audio lives: the model-input WAV + the chunk-local span.
 - `SeamRef` _class_ — A source-coordinate audio span across one fine-spine boundary (the g/G
 - `SpineView` _class_ — One Source's effective correction spine, cursor-windowed for the TUI.
+- `layer_speaker_menu` _function_ — Layer the assign-lane digit menu (pure; DEC 774dbe40). Tier 1 = the
 - `list_sources` _function_ — Enumerate the graph's Source nodes (the discovery corpus, 2ce81638).
 - `load_source_slice` _function_ — Decode a source-coordinate slice of the ORIGINAL media to playable samples.
 - `match_sources` _function_ — The --source selector (pure; shared by direct open and the picker's seed).
+- `match_speaker_query` _function_ — Narrow the registry by a typed query (pure; DEC 774dbe40 (4)). Rank 0 =
 - `neighbor_word_bound` _function_ — The adjacent word's FA boundary facing an overlay span (pure).
 - `open_stack` _function_ — Bootstrap the graph capability stack, resolving the db path (2ce81638).
 - `parse_entity_input` _function_ — Parse the new-speaker editor line (pure). A leading `?` marks the entity
@@ -218,11 +277,18 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 - `build_filter_pack` _function_ — Build the proposer's input: one source window's text-bearing effective
 - `exclude_strata` _function_ — The per-consumer filtered projection (DEC 9d4c0a38): "filtered" is a
 - `load_filter_proposal_sets` _function_ — Every filtering proposal set for a source (and optionally one spine),
+- `materialized_fix_ids` _function_ — The fidelity-edit apply path's materialization: a proposer row whose
+- `materialized_mark_ids` _function_ — Class-family routing's other half: a proposer's mark-family row (an ASR
+- `merge_filter_proposals` _function_ — Fold several proposal sets over ONE spine into one walkable set (design
 - `new_pack_id` _function_ — Generate a unique, sortable pack id.
 - `pack_digest` _function_ — Digest the READ content (source binding + window + numbered segments) —
 - `pending_filter_proposals` _function_ — The headless worklist: proposals with NO live stratum carrying their id and
+- `plan_pack_windows` _function_ — Cut a spine into `count` windows of near-equal TEXT-segment count at
 - `proposals_from_rows` _function_ — Resolve validated rows to proposal-set rows: proposal id, category, source
 - `render_filter_pack` _function_ — Render a pack as the brief a proposer reads: identity + window, the class
+- `render_filter_propset_markdown` _function_ — Project a proposal set for a HUMAN to check against the source in the
+- `resolve_slate` _function_ — Resolve a NAMED class slate (ruling c5b6cf42) to the two vocabularies a
+- `select_span_segments` _function_ — The span-EDIT gesture's resolver (agreed 2026-09-01 on the ch04 confirm
 - `strata_index` _function_ — Segment-keyed view of the live strata (the consumer query's index).
 - `validate_proposal_rows` _function_ — Validate + normalize proposer rows against their pack — loud on the first
 - `write_filter_propset` _function_ — Write one filtering proposal set: `<out_root>/<set_id>/manifest.json` +
@@ -230,4 +296,4 @@ A frontend-agnostic core for the transcript correction workflow — the first do
 ## Dependencies
 
 **Depends on:** `cjm-capability-primitives`, `cjm-context-graph-layer`, `cjm-context-graph-primitives`, `cjm-substrate`, `cjm-transcript-graph-schema`, `numpy`
-**Used by:** `cjm-transcript-correction-qt`, `cjm-transcript-correction-tui`, `cjm-transcription-core`, `cjm-workflow-hub-qt`, `cjm-workflow-hub-tui`
+**Used by:** `cjm-context-graph-projection`, `cjm-transcript-correction-qt`, `cjm-transcription-core`, `cjm-workflow-hub-qt`

cjm_transcript_correction_core-0.0.30/cjm_transcript_correction_core/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.0.30"

cjm_transcript_correction_core-0.0.30/cjm_transcript_correction_core/cleanread.py

@@ -0,0 +1,220 @@
+"""The clean read — L1 of the reading ladder (design 6752db0a; the subtraction
+shape from bbf8bafd (a)).
+
+L0 is the spine as spoken. L1 is the same spine MINUS what a reader never
+needs, as a MECHANICAL projection with VISIBLE elisions — it subtracts twice:
+
+1. FILLER LINES first: every text segment under an excluded stratum (default
+   `filler` — a wholly elidable line or run; the notes types' exclude_strata
+   name more) drops as a line;
+2. then the FILTERABLE OVERLAY SPANS inside the lines that remain — the
+   annotate lane's speech overlays whose label a reader filters (hesitation-
+   marker / discourse-marker / word-repeat / false-start); the KEEP labels
+   (emphasis-repeat / coincidental-repeat) never subtract. A line emptied by
+   its overlays drops too.
+
+The projection is pure over the effective spine + the active strata + the
+active overlays; nothing is stored. Every elision is visible: a subtracted
+span leaves `marker` in the line (default `[…]`; "" for a clean-verbatim
+consumer), a dropped line is recorded in the gap before the next kept line,
+and each kept line carries what was cut. A wrong exclusion must never HIDE
+content silently (ruling 6752db0a (9)) — the reader can always see that
+something was elided and audit it against L0.
+
+WORD-REPEAT KEEPS ITS LAST UNIT: the overlay quotes the whole run ('the the',
+'we, we', 'and and and') because the run is the detector sample; the clean
+read keeps the speaker's final instance — the span's tokens are split into
+the shortest repeating unit and only the last unit survives; a run that is
+not a clean repetition keeps its last token (a mechanical rule, auditable
+through the marker; refine per label as drives show).
+"""
+
+import re
+from dataclasses import replace
+from typing import Any, Dict, List, Sequence, Tuple
+
+from cjm_transcript_correction_core.graph import reanchor_span
+from cjm_transcript_correction_core.models import SpineSegment
+from cjm_transcript_correction_core.strata import strata_index
+
+CLEAN_READ_EXCLUDE_STRATA = ("filler",)   # The default line-level subtraction (the notes types name more)
+CLEAN_READ_FILTER_LABELS = ("hesitation-marker", "discourse-marker", "word-repeat", "false-start")
+CLEAN_READ_KEEP_LAST_UNIT = ("word-repeat",)   # Labels whose span keeps its final repeated unit
+ELISION_MARKER = "[…]"
+
+
+def _norm(t: str) -> str:  # Case/punctuation-insensitive token form (the span lane's rule)
+    return "".join(ch for ch in t.lower() if ch.isalnum())
+
+
+def repeat_survivor(
+    span_text: str,  # The overlay's words (a whole repeated run)
+) -> Tuple[int, int]:  # (char offset, char end) INSIDE span_text of the surviving last unit
+    """Locate the last repeating unit of a word-repeat span (pure): the
+    shortest unit u with the span = u × n (normalized) wins; no clean
+    repetition = the last token. Returned offsets are relative to the span."""
+    toks = [(m.start(), m.end(), m.group()) for m in re.finditer(r"\S+", span_text or "")]
+    if not toks:
+        return 0, len(span_text or "")
+    norm = [_norm(t) for _, _, t in toks]
+    k = len(norm)
+    for unit in range(1, k // 2 + 1):
+        if k % unit == 0 and all(norm[i:i + unit] == norm[:unit] for i in range(0, k, unit)):
+            return toks[k - unit][0], toks[-1][1]
+    return toks[-1][0], toks[-1][1]
+
+
+def _cut_ranges(
+    text: str,                          # The line's effective text
+    overlays: List[Dict[str, Any]],     # Active overlay correction dicts anchored on this line
+    labels: Sequence[str],              # Filterable labels
+    keep_last: Sequence[str],           # Labels that keep their last repeated unit
+) -> Tuple[List[Tuple[int, int]], List[Dict[str, Any]]]:  # (merged char ranges to cut, the cuts made)
+    cuts: List[Tuple[int, int]] = []
+    made: List[Dict[str, Any]] = []
+    for c in overlays:
+        p = c.get("payload") or {}
+        lab = p.get("label")
+        if lab not in labels:
+            continue
+        a = p.get("anchor") or {}
+        located = reanchor_span(a, text)
+        if located is None:
+            made.append({"overlay_id": c.get("id"), "label": lab, "text": a.get("text_snapshot"),
+                         "cut": None, "note": "words no longer on the line"})
+            continue
+        cs, ce = located
+        if lab in keep_last:
+            ks, ke = repeat_survivor(text[cs:ce])
+            rng = (cs, cs + ks)          # everything before the surviving unit
+            tail = (cs + ke, ce)         # trailing punctuation after it, if any
+            cuts.append(rng)
+            if tail[1] > tail[0]:
+                cuts.append(tail)
+            made.append({"overlay_id": c.get("id"), "label": lab, "text": text[cs:ce],
+                         "cut": text[cs:cs + ks], "kept": text[cs + ks:cs + ke]})
+        else:
+            cuts.append((cs, ce))
+            made.append({"overlay_id": c.get("id"), "label": lab, "text": text[cs:ce],
+                         "cut": text[cs:ce]})
+    cuts.sort()
+    merged: List[Tuple[int, int]] = []
+    for s, e in cuts:
+        if e <= s:
+            continue
+        if merged and s <= merged[-1][1]:
+            merged[-1] = (merged[-1][0], max(merged[-1][1], e))
+        else:
+            merged.append((s, e))
+    return merged, made
+
+
+def _tidy(s: str) -> str:  # Whitespace + punctuation seams after a cut
+    s = re.sub(r"\s+", " ", s).strip()
+    # ", ," -> "," — a cut's seam always leaves WHITESPACE between the two marks, so the rule
+    # demands it and runs BEFORE the space-before-punctuation rule closes that gap: a `::`
+    # the source wrote (thrust::reduce) is adjacent and never a seam (finding: the clean
+    # read handed a proposer `thrust:reduce`).
+    s = re.sub(r"([,;:])(?:\s+[,;:])+", r"\1", s)
+    s = re.sub(r"\s+([,.;:?!])", r"\1", s)          # "larger, than" — a space before punctuation
+    s = re.sub(r"^[,;:](?![,;:])\s*", "", s)         # a leading comma left by a cut at the line start (never a leading `::`)
+    return s
+
+
+def subtract_spans(
+    text: str,                        # The line's effective text
+    ranges: List[Tuple[int, int]],    # Merged char ranges to cut (from _cut_ranges)
+    marker: str = ELISION_MARKER,     # What a cut leaves behind ("" = nothing)
+) -> str:  # The clean text (tidied)
+    """Cut the ranges out of the text, leaving `marker` at each seam (pure)."""
+    if not ranges:
+        return text
+    out: List[str] = []
+    at = 0
+    for s, e in ranges:
+        out.append(text[at:s])
+        if marker:
+            out.append(f" {marker} ")
+        at = e
+    out.append(text[at:])
+    joined = "".join(out)
+    if marker:   # never let the marker glue to punctuation-less neighbours: "we [...] have"
+        joined = re.sub(r"\s*" + re.escape(marker) + r"\s*", f" {marker} ", joined)
+    return _tidy(joined)
+
+
+def clean_read(
+    segments: Sequence[SpineSegment],     # The EFFECTIVE spine (corrections applied), index order
+    strata: List[Dict[str, Any]],         # active_strata output
+    overlays: List[Dict[str, Any]],       # active_speech_overlays output
+    *,
+    exclude_strata: Sequence[str] = CLEAN_READ_EXCLUDE_STRATA,  # Line-level subtraction classes
+    filter_labels: Sequence[str] = CLEAN_READ_FILTER_LABELS,    # Overlay labels a reader filters
+    keep_last_unit: Sequence[str] = CLEAN_READ_KEEP_LAST_UNIT,  # Labels whose span keeps its last unit
+    marker: str = ELISION_MARKER,          # In-line elision marker ("" = clean-verbatim)
+) -> List[Dict[str, Any]]:  # Kept lines: {id, index, start, end, text, raw_text, elisions, dropped_before}
+    """L1 — the clean read (pure). Kept lines in spine order; each carries its
+    clean `text`, the L0 `raw_text`, the `elisions` cut inside it, and
+    `dropped_before`: the lines elided between the previous kept line and
+    this one (segment id, index, why). Wordless segments are not lines and
+    are never counted as dropped."""
+    idx = strata_index(strata)
+    drop = set(exclude_strata)
+    by_seg: Dict[str, List[Dict[str, Any]]] = {}
+    for c in overlays:
+        sid = ((c.get("payload") or {}).get("anchor") or {}).get("segment_id")
+        if sid:
+            by_seg.setdefault(sid, []).append(c)
+    out: List[Dict[str, Any]] = []
+    pending_drops: List[Dict[str, Any]] = []
+    for s in sorted(segments, key=lambda x: x.index):
+        if s.is_empty:
+            continue
+        cats = drop & set(idx.get(s.id, []))
+        if cats:
+            pending_drops.append({"id": s.id, "index": s.index, "why": sorted(cats)})
+            continue
+        text = s.text or ""
+        ranges, made = _cut_ranges(text, by_seg.get(s.id, []), filter_labels, keep_last_unit)
+        clean = subtract_spans(text, ranges, marker) if ranges else text
+        bare = clean.replace(marker, "").strip() if marker else clean.strip()
+        if not bare:
+            pending_drops.append({"id": s.id, "index": s.index,
+                                  "why": ["emptied by " + ", ".join(sorted({m["label"] for m in made}))]})
+            continue
+        out.append({"id": s.id, "index": s.index,
+                    "start": float(s.start_time) if s.start_time is not None else None,
+                    "end": float(s.end_time) if s.end_time is not None else None,
+                    "text": clean, "raw_text": text, "elisions": made,
+                    "dropped_before": pending_drops})
+        pending_drops = []
+    if pending_drops and out:   # a trailing run of dropped lines hangs off the last kept line
+        out[-1]["dropped_after"] = pending_drops
+    return out
+
+
+def clean_read_summary(lines: List[Dict[str, Any]]) -> Dict[str, int]:  # Counts for a pack / a report
+    dropped = sum(len(ln.get("dropped_before") or []) + len(ln.get("dropped_after") or []) for ln in lines)
+    cut = sum(1 for ln in lines for m in (ln.get("elisions") or []) if m.get("cut") is not None)
+    return {"lines_kept": len(lines), "lines_dropped": dropped, "spans_cut": cut}
+
+
+def clean_read_pack_read(
+    lines: List[Dict[str, Any]],  # clean_read output
+    marker: str = ELISION_MARKER,  # The marker the lines carry
+) -> Dict[str, Any]:  # build_filter_pack's `read` argument for an L1 pack
+    """The `read` record an L1 pack carries: the layer, the marker, the
+    counts, and the gaps (segment id -> the lines dropped before it) the
+    pack builder attaches to its rows so the brief shows every elision."""
+    return {"layer": "clean", "marker": marker, **clean_read_summary(lines),
+            "gaps": {ln["id"]: ln["dropped_before"] for ln in lines if ln.get("dropped_before")}}
+
+
+def clean_read_segments(
+    lines: List[Dict[str, Any]],  # clean_read output
+    segments: Sequence[SpineSegment],  # The same effective spine (for the fields a line does not carry)
+) -> List[SpineSegment]:  # SpineSegments carrying the CLEAN text — what a pack builder reads as L1
+    """Project the clean read back onto SpineSegments (clean text, same
+    identity and times), so every pack builder reads L1 unchanged."""
+    by_id = {s.id: s for s in segments}
+    return [replace(by_id[ln["id"]], text=ln["text"]) for ln in lines]