All notable changes to this project will be documented in this file.

• Sidecar mode: compose precision prompts in VectorPad and inject into active agent sessions
• internal/sidecar package: session discovery, JSONL parsing, user entry injection with UUID chain
• Launch target 7 (Sidecar inject): direct inject into most recent session JSONL
• Launch target 8 (Sidecar + deliberate): VectorCourt deliberation on framing before inject
• Two-path flow: DIRECT (compose → inject) and DELIBERATE (compose → VectorCourt → inject)

• Pressure panel: only show bars for non-zero sentences, display "✓ clear" when all safe

• Async submit flow: POST /v1/submit with queue position, poll loop, and live SSE spar events
• Submit(), PollSubmission(), GetCase() client methods
• Transparent fallback to sync Consult() when server returns 404/501

• Escalation support: parse clarification questions from VectorCourt verdicts
• SubmitClarification() client method for POST /v1/cases/:id/clarify
• vectorpad clarify <case-id> CLI command: interactive or --json pipe mode
• CLI submit output shows escalation questions with impact level and defaults
• TUI verdict summary shows clarification hint with case ID

• Instant precedent recognition: 300ms debounced trigram search in risk panel while typing
• InstantPrecedents() client method for GET /v1/precedents
• Prediction debt health display in risk panel: debt ratio and band (healthy/accumulating/critical)
• GetPredictionDebt() client method for GET /v1/health/prediction-debt
• Live spar viewer: vectorpad submit --to vectorcourt --live streams deliberation rounds via SSE
• SSE stream client (internal/vectorcourt/stream.go) for GET /v1/submissions/{id}/stream
• ANSI spar renderer (internal/tui/spar.go) — plain terminal output, works over SSH
• Editor line indicator: L3/18 in dashboard when text overflows visible area
• Adaptive classified view: collapses to 2 lines for long text, hidden for very long text

• Rebrand Oracul to VectorCourt: package internal/oracul renamed to internal/vectorcourt
• Auth header: X-Oracul-Key replaced by X-VC-Key
• Default endpoint: https://oracul.app replaced by https://vectorcourt.com
• Config keys: oracul.api_key / oracul.endpoint replaced by vectorcourt.api_key / vectorcourt.endpoint
• CLI flags: --to oracul / --format oracul replaced by --to vectorcourt / --format vectorcourt
• TUI: launch target "Oracul Council" renamed to "VectorCourt"
• Backward-compatible config migration: existing oracul config keys auto-migrate on load

• vectorpad outcome <claim-id> --result success|failure|partial --note "...": report decision outcomes to Oracul
• ReportOutcome() client method for POST /v1/cases/:id/outcome
• Outcome auto-annotates matching flight log record
• Shell completions updated with outcome subcommand

• vectorpad stash diff <id1> <id2>: compare two verdict entries field-by-field
• Verdict diff logic in internal/stash/diff.go: JSON parsing, field comparison, rendered output
• lookupStashItem helper: find items by ID, claim ID, or prefix match

• vp short alias: same binary built as vp via GoReleaser, included in Homebrew install and release tarballs

• Oracul metadata in flight recorder: tier, filing quality, preflight result, warnings, precedent count
• vectorpad log --stats shows Oracul aggregates: submits, avg filing quality, rejection rate, top warnings
• UpdateOracul() method on Recorder for attaching async verdict data to flight records

• Async Oracul submit: deliberation runs in background, TUI remains interactive
• Elapsed timer display in editor panel during deliberation
• Context cancellation on TUI quit cancels in-progress deliberation
• Rejects duplicate submissions while deliberation is active

• Precedent search: live similar-case lookup in TUI risk panel after 3s idle
• vectorpad precedent CLI subcommand with --limit and --json flags
• PRECEDENTS section in risk panel: similarity scores, outcome dots, ref class summary
• SearchPrecedents() client method for GET /v1/precedents/search
• Debounced async precedent search with independent 3s tick track

• Verdict history in stash: Oracul verdicts auto-stashed after successful submit
• Distinct ◆ symbol for verdict entries in TUI stash panel and CLI list
• SourceVerdict and ItemTypeVerdict types for verdict classification
• stash list --type verdict filters to verdict entries only
• stash show <id> renders full verdict JSON for verdict entries

• Live preflight readiness indicator: READY/WARN/BLOCKED updates in risk panel after 2s idle
• Debounced async preflight check with Bubbletea tick/cmd pattern
• Cached results until text changes — no redundant API calls

• Oracul Council as launch target 6 in ctrl+l overlay: classify, preflight gate, submit, verdict summary
• Target only visible when API key is configured; hidden otherwise

• Oracul account status in TUI risk panel: tier, usage bar, reset time (only when API key configured)
• Account status refreshes on startup and after each launch
• Silent degradation: section hidden when no key or fetch fails

• Oracul integration: vectorpad submit --to oracul sends classified cases for deliberation
• Oracul export: vectorpad export --format oracul emits CaseFiling JSON offline
• Config system: vectorpad config set/get for persistent settings (API key, endpoint)
• Preflight gate: validates cases with Oracul before submission, blocks rejected filings
• Sentence-to-CaseFiling mapper: DECISION, CONSTRAINT, TENTATIVE, SPECULATION, EXPLANATION tags map to filing fields
• Shell completions updated for config, submit, export commands

• ContextSpectre feedback loop: risk panel shows session health grade, context pressure, turns remaining, model, and cost
• Decision economics display: actual CPD/TTC/CDR from contextspectre stats with per-epoch breakdown
• Feedback refreshes on TUI startup and after each launch event
• Graceful degradation: feedback sections hidden when contextspectre is not installed

• Claim registry: SQLite stash with embedding similarity (vectorpad stash)
• SQLite backend replaces JSON for stash persistence (auto-migrates existing JSON)
• Ollama embeddings via nomic-embed-text for semantic similarity search
• Cosine similarity with thresholds: near-duplicate (>0.90), same idea (0.80-0.90), related (0.65-0.80)
• Claim IDs assigned at stash time for tracking idea evolution
• CLI: stash add, stash list, stash compare, stash show, stash cluster, stash evolve, stash reindex
• Graceful degradation: works without Ollama (no embeddings, Jaccard fallback for clustering)

• Vector decomposition (ctrl+b): split high-blast-radius vectors into focused sub-vectors by target groups
• Sub-vectors include shared preamble (constraints, context) plus group-specific sentences
• Decompose suggestion in risk panel when 3+ distinct targets detected
• Decompose output in CLI pipe mode (DECOMPOSE section)

• Pressure heat map: per-sentence risk scoring based on lock policy, classification tag, vague verb presence, and brevity
• Three pressure levels (LOW/MED/HIGH) with visual bar in risk panel
• Pressure output in CLI pipe mode (PRESSURE section)

• Risk panel now displays detected scope markers ("all repos") and target count instead of showing repos: 0 files: 0

• Scope declaration (ctrl+d): declare blast radius before writing, cross-reference against text
• Scope mismatches surface in risk panel (scope vs constraints, operation vs preservation, target mentions)

• Negative space detection: 6 gap types flag missing constraint classes (preservation, success criteria, review, rollback, scope boundary, identity)
• Drift timeline in TUI: risk panel shows meaning drift as you edit (strengthened/weakened/flipped/added/removed)
• Flight recorder: append-only JSONL log of launched vectors with metrics, gaps, and outcome annotation (vectorpad log)
• Constraint pinning: risk panel warns when CONSTRAINT sentences are removed during editing

• Pastewatch detection now checks both pastewatch-cli (Homebrew) and pastewatch binary names

• Launch target picker (ctrl+l): clipboard, Claude for Mac, ChatGPT for Mac, Claude Code CLI, file
• Essence extraction (ctrl+e): collapse stash stack into launchable summary
• Stash uniqueness symbols: ● bright (novel), ○ dim (overlaps), ◌ faint (duplicate)
• Shell completions for bash, zsh, and fish (vectorpad completion <shell>)
• Full README with badges, architecture, philosophy, bond, roadmap

• vectorpad now launches TUI by default (no need for vectorpad tui)
• Added keybinding hint bar at bottom of editor panel
• Clipboard test skips on Linux CI when xsel/xclip unavailable

• File attachment pipeline: detect paths, classify by extension, preview, serialize for copy-out
• Image protocol detection (iTerm2, kitty)
• Object card rendering in editor panel
• Paste interception: dragged files become attachment objects, not inline text
• Excerpt configuration per attachment
• Three serialization modes: path-only, excerpt, evidence

• Three-panel TUI: stash roster, vector editor, risk panel
• Sentence classifier with 6 tags (CONSTRAINT, DECISION, TENTATIVE, QUESTION, SPECULATION, EXPLANATION)
• Meaning drift detection on 6 axes
• Pre-flight metrics: token weight, CPD, TTC, CDR projections
• Ambiguity detection with blast radius, brevity ratio, vague verb flagging
• Nudge protocol for ambiguous vectors
• Stash persistence with Jaccard similarity clustering
• Capability detection for pastewatch and contextspectre
• Pastewatch integration: scan outbound payload before clipboard copy
• CLI pipeline: stdin classification + preflight + ambiguity analysis
• vectorpad tui - interactive three-panel interface
• vectorpad add - quick-add ideas to stash from CLI