Constitutional-first. Declaration-driven. Schema-grounded. Outputs survive governance, procurement, and review — because the architecture leaves no other option.
- Compiled, not generated
Plans are compiled from declared principles that compose like case law -- not improvised by an agent.
- Immutable lineage
Every intermediate result is stored, timestamped, and linked to what produced it. Delete nothing.
- Grounding gate
Every parameter passes through schema validation. If it doesn't exist in your data, it's rejected — not guessed.
Today
Yesterday
Last 7 days
You have reached the end of your query history.
Features
Provenance produces a reviewer-ready evidence package: pipeline steps, parameters, lineage, intermediate artifacts, and structured warnings. Designed to be forwarded internally without you defending it live.
Re-run the same analysis later
Same inputs, same plan, same output. Deterministic by design.
Explain exactly how a number was produced
Every step, parameter, and data source is recorded and attributable.
Hand to audit / compliance without rewriting work
Export PDF, JSON, pipeline definitions, and plan graph. Designed to be forwarded, not explained.
session evidence package
Query → Plan → Execute → Receipt. Four stages. No improvisation.
Step 1
Intent parsing
- ✓Your question is compiled against a constitutional framework of analytical primitives. New capabilities emerge from existing principles -- no new code required.
- ✓Anything outside the declared primitives is rejected. Not guessed at. Not approximated. Rejected.
- ✓If your question is ambiguous, Provenance asks. It does not assume.
Step 2
Plan compilation
- ✓The system cannot construct an invalid pipeline. This is architectural, not a guardrail.
- ✓No agent decides what tool to use. The plan follows from declared principles -- deterministically.
- ✓Every parameter passes through the grounding gate: your question, a declared default, or your schema. Nothing else enters.
Step 3
Transactional execution
- ✓If step 3 of 5 fails, steps 1 and 2 are preserved. The failure is explained. Replanning starts from step 3.
- ✓Each step produces immutable artifacts with lineage, timestamps, and parent references.
- ✓If an operation has variance, you are told before the result is shown -- not discovered later.
Step 4
Receipt export
- ✓Steps + parameters + artifacts + warnings
- ✓PDF, JSON, pipeline definitions, and plan graph
- ✓Reviewer-ready evidence package -- the Receipt is the product
Fast answers are common. Fast answers you can approve are rare.
What we refuse to do says more than what we claim.
Will not guess.
If your question is ambiguous, it asks. It does not pick the most likely interpretation and hope.
Will not improvise a pipeline.
Every operation sequence is pre-validated. There is no creative problem-solving at execution time.
Will not fabricate columns.
If the field doesn't exist in your schema, the query fails cleanly -- it does not hallucinate a plausible-sounding column name.
Will not silently retry.
If something fails, you see the failure, the reason, and the replan. No hidden loops.
Will not produce an output it cannot explain.
Every number traces back to source data through documented steps. If it can't show the path, it won't show the number.
Will not ship an approximate answer as exact.
If a result involves estimation, rounding, or lossy operations, the warnings are in the Receipt -- not in a footnote you'll never read.
Will not call itself intelligent.
Provenance is deterministic software. It compiles analytical plans from declared constraints. There is no creativity involved. That's the point.
Not better analytics. Different infrastructure.
Dashboards
Trusted outputs. Slow to change. Someone still has to build them.
Chat tools
Fast outputs. No lineage. Good luck getting sign-off.
Provenance
Fast + you can hit forward without a meeting. The Receipt is already in the appendix.
Examples -- the same Receipt model works anywhere analytics needs review.
Energy
Regulatory reporting, internal assurance.
"Outliers vs peer baseline, last 30 days."
Queries, filters, and aggregation logic -- all recorded.
When the regulator asks "how did you get this number?", the Receipt answers before you do.
Retail
Forecast accuracy must be defensible.
"Which regions underperformed vs plan -- and why?"
Data sources, thresholds, and comparison method -- traceable.
The forecast was wrong. The Receipt proves your method wasn't.
Shipping & Logistics
SLA reporting needs traceable derivations.
"Show container dwell time by port, last 90 days."
Date ranges, grouping logic, and source tables -- explicit.
When a client disputes the SLA report, you send the Receipt. Dispute over.
Manufacturing
Yield and quality reports face audit scrutiny.
"Yield rate by production line, week-over-week."
Calculation steps, filters, and intermediate outputs -- saved.
The auditor doesn't need your time. They need the Receipt.
Finance
Board packs, SOX, internal audit trails.
"Break down revenue by segment and region for board review."
Every join, filter, and aggregation -- exportable for audit.
The board won't ask how you got the number. The Receipt is already in the appendix.
Pharma & Life Sciences
GxP-style expectations (traceability, reviewability).
"Batch deviations and contributing factors."
Full lineage from source tables to final output -- reproducible.
GxP traceability without the GxP overhead. The Receipt is the validation record.
Point it at your warehouse. No configuration, no connectors, no onboarding sprint.
Provenance discovers your schema automatically. Tables, columns, types, relationships, value distributions — all inferred.
Marketplace · Q2 2026
If not, Provenance doesn't ship it. Every output includes the evidence needed for sign-off — so review is a formality, not a debate.
These aren't features we promise. They're constraints the architecture enforces.
Structurally incapable of improvising
Structurally incapable of improvising
Every pipeline is compiled from declared principles that compose like constitutional law. New capabilities emerge from existing constraints -- no new code paths. If an operation can't be derived, it cannot execute.
Grounding gate rejects fabrication
Grounding gate rejects fabrication
Every parameter passes through the grounding gate before execution. If a column isn't in your schema, it doesn't exist. If a filter value isn't in your data, it's rejected. Nothing is invented.
Failures explain themselves
Failures explain themselves
If a step fails, Provenance tells you what failed and why -- then replans from the last good state. No silent retries. No hidden fallbacks.
Every intermediate result is kept
Every intermediate result is kept
Not just the final answer. Every step along the way is stored, timestamped, and linked to what produced it. Delete nothing.
Warnings are first-class, not footnotes
Warnings are first-class, not footnotes
If something is approximate, lossy, or non-deterministic, Provenance flags it before you see the result -- not after someone asks.
Zero configuration, no domain setup
Zero configuration, no domain setup
Point it at your warehouse. It discovers your schema automatically -- tables, columns, types, relationships, value distributions. No mappings. No onboarding sprint.
We map your boundary, run 3 sample questions, and deliver the Receipts for review.