01 / CAPTURE
Keep the whole history.
Declare the intended outcome. Retain dispatches, responses, and what the agent reported in a portable typed trace.
Agents act.
Evidence answers.
A tool response is only part of the story.
Trace the action. Read back the outcome. See what holds up.
Python 3.11+MIT licensedNo required runtime dependencies
CAPTURED HISTORY04 events
- 00 msContract registered
One refund before 50 ms
- 05 msAction dispatched
Agent requests the refund
- 20 msTool timed out
Agent reports failure
- 80 msIndependent readback
Matching effect; qualified agent provenance
EVIDENCE GRAPH Origin links
DISPATCHAction A01
RESPONSETimeout
READ-ONLY SOURCEIndependent store
OBSERVED EFFECTEffect E01
Persistence horizon: 70 msILLUSTRATIVE TIMES
EVIDENCE SUPPORTS
Contract satisfied.
Independent readback establishes the refund. Qualified provenance links the effect to this agent.
- Verified state
- Satisfied
- Effect origin
- Confirmed
- Reported failure
- Contradicted
Illustration uses an explicit action-linked target matcher. Package reports leave alignment unassessable. Understand the boundary
A focused Python framework.
Built to stay out of your way.
Built to stay out of your way.
0Required runtime
dependencies
dependencies
.htmlSelf-contained
offline reports
offline reports
MITYour stack.
Your application.
Your application.
BEYOND THE RESPONSEAn action happened.
An action happened.
Build a better account.
Refunds. Deliveries. Database writes.
When agents change something, keep the intention, the history, and the readback together.
02 / READ BACK
Check beyond the tool.
Your application supplies a verifier and a trusted read-only source. Keep state, timing, persistence, and effect origin explicit.
03 / INSPECT
See what the record supports.
Follow the timeline and evidence graph. Distinguish a verified state from who produced it, and leave missing evidence as unknown.
You define success and establish source trust. Consequa-X preserves and explains the evidence you supply; a graph edge does not establish causal necessity.
LESS FRICTION. MORE CONTEXT.Small core.
Inside the framework Small core.
Nothing swept away.
01 / EVIDENCE CONTINUITYSame effect.
Same effect.
Every witness.
A later observation shouldn’t rewrite the earlier record. Successive readbacks retain separate witnesses, even when they refer to the same effect.
Share a smaller surface.
Default reports omit captured payloads and raw identifiers. Local display aliases keep the structure readable. Reveal captured fields only when you choose.
See what stays visible03 / LOCAL BY DESIGNOne file.
One file.
Ready to inspect.
Export a self-contained HTML report. Open it without a model, API key, hosted account, or web server.
Make your first reportField omission reduces disclosure; it is not universal PII detection. Keep the original trace and source evidence as the record.
THE PACKAGE, IN VIEWOpen the report.
Open the report.
Follow the evidence.
A walkthrough of the HTML report generated by Consequa-X 1.1.0. These are actual package screenshots, using synthetic example data.
Actual package output · synthetic example dataA short walkthrough captured from a generated report: summary, timeline, readback, and evidence graph. It performs no live action, model call, or replay. Play it when you choose; pause returns to the still image.
ZIP contains the original self-contained HTML.
FROM CURIOUS TO HANDS-ONYour first trace.
Your first trace.
Five minutes away.
Install the package, generate synthetic examples, and open an offline report. Then connect your application’s independent readback.
Follow the quickstartVersion 1.1.0 is available on PyPI.
This quickstart performs no real refund or delivery.
quickstart / localPython 3.11+
# 01 — Install the framework
$ python -m pip install consequa-x==1.1.0# 02 — Generate synthetic captures
$ consequa-x demo --output my-demo# 03 — Make an offline report$ consequa-x report \
my-demo/traces/episode_01.json \
--output my-report.html Open my-report.html in your browser.