Phase B3: pixel-extrapolation ghosts (EXTRAP / Path B)

[mandarwagh9]

What & why

Phase B3 — the third and final ghost-prediction path. When a camera loses a person it was previously tracking (but the world object is still alive via another camera), dead-reckon a red EXTRAP ghost from that camera's last-known pixel position along its pixel velocity, capped by an adaptive budget. All three ghost paths (green/orange/red) are now live.

⚠️ Stacked on the chain (#10 → … → #4). Auto-retargets as the stack merges.

Changes

• World model — refined the prediction skip from "ever seen by this camera" to "seen within a live window" (~2 frames), so a camera that lost a target becomes eligible for a ghost. generate_predictions now tries Path A (homography) → Path B (extrapolation) → Path C (world projection).
• _try_extrapolation_prediction — slides the last pixel by velocity × time × fps, capped at min(250, 80 + 40·t) px; zero-velocity stays put; no pixel history → None (falls through).
• README — flipped EXTRAP / red ghost from 🔭 planned to ✅ implemented (status table, ghost-predictions row, AR-overlay table + paragraph, edge-case caveat). All three paths now documented as active.

Tests (TDD)

6 new in test_extrapolation.py: no-history → None, moves in velocity direction, displacement capped by budget, zero-velocity stays at last pixel, integration emits EXTRAPOLATION for a lost camera, and a live camera is skipped (no ghost).

Verification

• ruff clean · mypy clean (17 files) · 95 tests pass (was 89).
• Coverage 58% (floor 50%).

🤖 Generated with Claude Code

[mandarwagh9]

Superseded — all commits from this phase landed on main via the merge of #13 (the top of the stack, which contains every phase commit). Closing to tidy up; the work is fully merged.