Internal Engine (Trajectory)¶
Predict likely next features from recent development patterns. Archobs exposes per-cluster velocity, edge relationships, and commit activity natively — use these as the primary data source.
The data is deterministic and structured. Feature adjacency reasoning ("export features suggest reports are coming next") is your job as the LLM — the tools give you the evidence.
Same-session fast path¶
When running in the same session as archobs (data already loaded):
archobs show velocity --window 30 --compare --format json— includesadded_paths,external_inbound_weight,recent_file_changes_30d,recent_file_changes_90d, andis_emerging(when--compareis used).archobs show edges --top-active 3 --max-neighbors 10 --format json— auto-selects top-3 clusters by file_change_count.git branch -r --sort=-committerdate | head -20git log --since="30 days ago" --format="%s" --no-merges | head -40- Theme extraction + feature adjacency reasoning (see below)
Run commands 1-2 in parallel (independent artifacts). Then 3-4 in parallel (independent git queries).
Primary path: archobs-native queries¶
-
Get velocity data — per-cluster commit activity with growth/churn ratios:
Recently added file paths (added_paths) are included by default in JSON output — they are the highest-signal data for feature prediction. Use--compareto get acceleration relative to the prior window. -
Inspect cluster relationships — which clusters are connected and how strongly:
-
Get full context — run individual queries in parallel:
-
Collect branch signals — active branches are the highest-confidence trajectory signal:
Branch names are direct feature declarations. Extract ticket ID prefixes (e.g.OIQ-516,CR-02) and cross-reference with commit message prefixes to group related branches into feature initiatives.
Ticket series grouping: When 3+ branches share a numeric prefix series or a domain keyword appears in both branch names and commit messages, group them as a single initiative. Report the initiative name, branch count, and combined cluster footprint.
4b. Collect commit message themes:
Parse conventional commit prefixes to identify which domains are receiving feature work vs maintenance.- Interpret velocity signals:
| Signal | Suggests |
|---|---|
High growth_ratio |
New capability being built |
High churn_ratio |
Feature refinement/iteration |
High acceleration (with --compare) |
Active development push |
Low acceleration |
Work winding down |
High recent_file_changes_30d in cluster |
Focused sprint in one area |
| Cross-cluster edges (show edges) | Feature adjacency — what depends on what |
High external_inbound_weight |
Gravitational center — other clusters pull toward this one |
For detailed interpretation rules (compound velocity matrix, per-file intensity, convergent hub pattern, is_emerging, test-only clusters, cross-cluster initiative detection), see velocity-interpretation.md.
- Reason about feature adjacency using the patterns in
feature-adjacency.mdand your domain knowledge.
Manual fallback: When archobs artifacts are not available, see manual-fallback.md for git-only trajectory extraction.
Combined archobs + trajectory workflow¶
When running both archobs and trajectory in the same session (the most common case):
archobs report(blocking — wait for completion)- Parallel: all archobs
showcommands + trajectory git commands:# Archobs queries (parallel): archobs show risks --top 10 --format json archobs show clusters --sort leakage --format json archobs show drift --format json archobs show summary --format json archobs show velocity --window 30 --compare --format json archobs show edges --top-active 3 --max-neighbors 10 --format json archobs show suggestions --format json # Git queries (parallel with above): git branch -r --sort=-committerdate | head -20 git log --since="30 days ago" --format="%s" --no-merges | head -40 - Synthesize into a combined report.