Changelog
v1.23.0 — 2026-11-06
New Feature
- Team Architecture — Reorg Validator
A new Drift & Reorg tab on the Team Architecture page lets you enter any reorganisation date and compare cross-team coupling before and after:
Coordination Events — sum of shared commits across all cross-team module pairs in each window. Events are strictly additive: Before + After = all-time total.
Cross-Team Coupling Pairs — unique module pairs that change together across team boundaries within each window. Pairs are not additive (the same pair can appear in both windows).
Contested Modules — modules where a secondary team contributes ≥ 20 % of churn within the window. Computed against all-time ownership so team identity does not shift with the date filter. Unaffected by the noise toggle.
All three numbers are served by a dedicated
POST /v1/analysis/alignment/reorg-comparisonendpoint. The frontend renders pre-computed values only.- Team Architecture KPI Summary endpoint
New
POST /v1/analysis/alignment/kpi-summaryreturns the three all-time headline KPIs (cross_team_pairs,coordination_events,contested_modules) with anexclude_noisetoggle. Replaces frontend aggregation — the UI now has a single API source for the KPI cards.
Improvement
- Team coupling matrix — continuous colour scale
The coordination matrix previously used four fixed amber buckets. It now uses a continuous opacity scale (0.15 → 1.0) so cells with similar but distinct values (e.g. 127, 203, 279) are always visually distinguishable. A gradient legend is shown below the matrix.
- Noise toggle shared across views
The Exclude noise toggle on the Team Architecture page and the Change Coupling page now share state via
sessionStorage. Toggling on one page is immediately reflected on the other. Default is ON after every page reload.- Tooltip and label accuracy
Ownership tooltips now state that module ownership is determined by churn (lines added + deleted), not commit count.
Coordination Events tooltip clarifies that a single commit touching three teams’ modules counts once per affected pair.
Reorg Validator hint explains additivity rules and noise-toggle scope.
Fix
- Stable team ownership across date windows
Team assignment (which team owns a module) is now derived from all-time commit history regardless of the active date filter. Previously the primary team could shift between the before- and after-window, inflating pair counts and breaking the additivity invariant.
before.coordination_events + after.coordination_eventsnow always equals the all-time total.- Deterministic results under alias/churn ties
Five non-deterministic
LIMIT 1queries (user/alias lookups andROW_NUMBERteam-ownership ranking) now have stable secondary sort keys (ORDER BY u.id,ORDER BY al.user_id,ORDER BY team_name). Repeated API calls for the same dataset return identical numbers.- Config import — duplicate user validation
run_config_syncnow validates the YAML before writing to the database:Duplicate
users.name→ aborts with the name and both indices. The database is left unchanged.Team member or
team_memberships.developernot present inusersor aliases → aborts with a full list of unresolvable names.Duplicate alias claimed by two users → warning (permitted — real Git author-name collisions); first match wins.
This prevents the class of inconsistency where a developer appeared in both team-membership tables with different user IDs depending on YAML entry order.
Upgrade
No migration required. Re-run config import if you have duplicate usernames in your
config.yaml — the importer will now report them clearly instead of silently
producing inconsistent team assignments.
./manage.sh pull-latest && ./manage.sh up
v1.22.0 — 2026-11-04
New Feature
- Owner Activity Warnings
Two new warning types surface knowledge-risk situations that the existing
silo_riskwarning does not cover — the activity dimension:``inactive_owner`` — the dominant contributor (≥ 60 % all-time ownership, minimum 5 commits per file) has not committed in the last 90 days. Critical at ≥ 80 % ownership, high at ≥ 60 %. Ownership is computed all-time up to
end_dateso historical authorship is never erased by a narrow analysis window.metric_value= days inactive.``no_active_owner`` — nobody has committed to the file in the last 180 days (minimum 5 all-time commits). Always high severity. Surfaces dormant files that carry accumulated risk with no active watchdog.
metric_value= days dormant.
Both types are purely Git-derived — no CODEOWNERS file required.
confidenceisnull(Git-only signals). Dashboard warnings panel andPOST /v1/analysis/warningsupdated.
Improvement
- Uniform HTTP method — all analysis endpoints now POST
The four remaining
GETendpoints (/v1/config/modules,/v1/config/analysis-since,/v1/context,/v1/repository/info) have been converted toPOSTfor consistency. Infrastructure endpoints (/info,/api/health) remainGET.
Upgrade
No migration required. If you call the config or context endpoints directly
(e.g. from Grafana or a custom integration), update the HTTP method from
GET to POST.
./manage.sh pull-latest && ./manage.sh up
v1.21.0 — 2026-10-27
Improvement
- Change Coupling — relative coupling score and noise filters
The Change Coupling view now uses a relative coupling score instead of a volume-normalised one:
score = shared_commit_count / min(module1_commits, module2_commits) × 100This separates structurally significant coupling (score ≥ 60) from incidental co-changes (score < 20) regardless of module size or activity volume. Pairs are ranked by score rather than raw co-change count.
Two new filters (both on by default):
Hide incidental — removes pairs with score < 20 from the list.
Exclude noise — removes pairs involving known noise modules (
jstests,third_party) that inflate co-change counts without architectural significance.
The Cross-Team Co-Changes KPI now reports
SUM(shared_commit_count)across cross-team pairs — the number of distinct commits where two teams had to synchronise — instead of the file-pair multipliedtotal_co_changesfigure that inflated the count.
Upgrade
No migration required.
./manage.sh pull-latest && ./manage.sh up
v1.20.0 — 2026-10-27
Improvement
- Tracker Confidence Metadata
Tracker-dependent results now include a
confidenceobject reporting what percentage of commits in the database have a resolved ticket type. Consumers can use this to assess signal quality before acting on results.POST /v1/analysis/bug-magnets— top-levelconfidencefieldPOST /v1/files/details—confidenceinsidebug_recencyPOST /v1/analysis/warnings—confidenceondefect_attractorandbug_magnetitems;nullon Git-only warning types
confidence.tracker_coverage_pctranges from 0–100. Computed by a single lightweight SQL query at request time — no additional tracker API call.- Videos page
New page at
/videos/listing the Calyntro feature episode series. Linked from the main navigation.
Upgrade
No migration required.
./manage.sh pull-latest && ./manage.sh up
v1.19.0 — 2026-10-25
New Feature
- Bug Magnet Detection
Files that repeatedly attract bug-fix commits are now surfaced automatically using a decay-weighted Fix-Recency Score. The score sums bug-linked commits over the last 365 days, weighting recent fixes more than older ones (linear decay:
weight = max(0, 1 − days_ago / 365)).New endpoint
POST /v1/analysis/bug-magnetsreturns files ranked by Fix-Recency Score, filterable by module and date range.The file detail response (
POST /v1/files/details) now includes abug_recencysection with score, days since last fix, and 90-day bug count.Dashboard warnings panel surfaces files with score ≥ 1.5 as
bug_magnetwarnings.Requires a ticket provider (e.g. Jira) configured in
config.yaml. Returns empty results gracefully when no tracker data is available.
Upgrade
No migration required. Fix recency is computed live from existing
commit_tickets data — no re-import needed.
./manage.sh pull-latest && ./manage.sh up
v1.18.0 — 2026-10-24
Improvement
- Validation study extended to four codebases
Apache Flink added as the fourth repository in the prospective validation study. T1–T4 finding tests now cover MongoDB, uv, Qt Base, and Apache Flink with reproducible, version-controlled JSON results.
AUC 0.682 (T+3M) / 0.710 (T+6M) on Apache Flink. 50 % of bug-affected files ranked in the top quartile of a churn-based risk ranking.
Upgrade
No migration required.
./manage.sh pull-latest && ./manage.sh up
v1.17.0 — 2026-10-19
Improvement
- API Parameter Naming Standardised
All analysis endpoints now use a consistent parameter vocabulary:
start_date/end_date(previouslyfrom/toon some endpoints),module_name(previouslymoduleorcomponent_name), andlimit(previouslynbr_items). The/v1/analysis/URL prefix is now uniform across all endpoints (previously some routes used/v1/analyze/). The tracker coverage endpoint (/v1/analysis/tracker/coverage) changed fromGETwith query parameters toPOSTwith a JSON request body, consistent with all other filter endpoints.Note: This is a breaking change for clients using the old parameter names.
Upgrade
Clients using the old parameter names (from, to, module,
component_name, nbr_items) or the /v1/analyze/ URL prefix must
update their requests. See the API Reference
for the current parameter names.
./manage.sh pull-latest && ./manage.sh up
v1.16.7 — 2026-10-15
Improvement
- Self-contained release tarball
The release tarball now includes all deployment files alongside the Docker images. Customers no longer need registry access to install or update Calyntro.
Tarball contents:
calyntro_<version>/ images.tar.gz # docker save — backend, ui, nginx manage.sh docker-compose.yml nginx.conf scripts/ generate_config.sh
Upgrade
No migration required.
v1.16.6 — 2026-10-13
Fix
- Config generator: no duplicate user entries
generate_config.shcould produce multipleusersentries with the same name when a developer committed from two different email addresses (e.g. work and personal). The generator now merges entries by canonical name — name variants from all email addresses are collected as aliases.
Upgrade
No migration required. Re-run generate_config.sh to get a clean config
if your existing config.yaml contains duplicate names.
v1.16.5 — 2026-10-12
Fix
- File Detail: misleading trend suppressed for bulk-imported files
The File Detail view showed
±100%trend indicators for files that were added in a single large commit and never modified again (common for third-party or vendored code). The trend is now shown as0%when either half of the comparison window contains no activity, since no directional signal can be derived.Affected metrics: Churn Rate, Silo Risk, Hotspot Score trend.
Upgrade
No migration required.
v1.15.0 — 2026-08-30
New
- AI Pain Point Navigator
A new opt-in endpoint —
POST /api/v1/analysis/brief— aggregates Calyntro’s hotspot, knowledge silo, coupling, and warning signals and asks a configured LLM to prioritise the top 3–5 architectural risks, phrased for engineering leadership.Supported providers: OpenAI, Anthropic, Ollama (local).
Identifiers (file paths, author names, module names) are replaced with opaque tokens before the LLM call and restored in the response — controlled by
ai_brief.anonymize(default:true).Returns HTTP 503 when the feature is disabled, so existing deployments without the
ai_briefconfig block are unaffected.The UI exposes the brief via a new AI Brief view (sidebar, Sparkles icon) with a date-range picker and colour-coded urgency cards.
See api (AI Pain Point Navigator section) and Calyntro Configuration File (
ai_briefsection) for full configuration details.
Upgrade
No migration required. The feature is disabled by default — existing databases and deployments continue to work without any changes.
To enable, add an ai_brief block to your config.yaml and restart:
./manage.sh pull-latest
./manage.sh up
v1.14.2 — 2026-08-21
New
- Go support for complexity analysis
Go files are now fully analysed: cyclomatic complexity, lines of code, and cognitive complexity via Tree-sitter. Repositories with Go components benefit immediately — no configuration required.
- Faster import
Ticket linking during import is up to 7× faster. Large repositories with many commits will see a significant reduction in import time.
Upgrade
No migration required. Existing databases remain compatible.
./manage.sh pull-latest
./manage.sh up