Inspect before you share.
Run locally, export JSON or Markdown and create a signed portable bundle without sending source code to a hosted scanner.
p7 inspect ./backendRescueApply for access
Private alpha · v0.1.1
Inspect the evidence. Not the confidence.
PRODUCTION-7 turns a Python backend into an explicit engineering decision: what the repository proves, what remains unknown and what should block a release.
- Local-first inspection
- Project code is never executed
- Source is not uploaded
$ p7 inspect ./backend --format json
Target python-api
Rule pack production-7-python@0.1.0
Controls 14 across 7 dimensions
Execution read-only / no project imports
GATE INCOMPLETE
Score 74 / 100
Coverage 54.3%
7 evidence decisions still required14 versioned controls7 production dimensions3 delivery surfacesEd25519 signed reports
Evidence chain
A score is the end of the reasoning, not the beginning.
Every conclusion remains connected to a deterministic observation, supplied artefact or explicit human decision. Unknowns stay visible instead of becoming invented certainty.
- 01Inspect locally
Read bounded repository evidence without importing or executing the target application.
- 02Separate facts from gaps
Deterministic checks create cited observations; runtime and organisational claims become evidence requests.
- 03Apply the gate
Coverage and seven dimension scores remain separate, so an average cannot hide an unknown critical property.
- 04Sign the handoff
Bundle the canonical report, signature, public key and verification instructions for another engineer or LLM.
One engine · Three surfacesCLI / Local MCP / Agent CI / Policy
Use the same evidence contract where engineering decisions happen.
Give coding agents bounded tools.
Let an MCP-compatible agent discover controls and missing evidence without giving it an arbitrary shell or permission to invent verification.
p7-mcp --root ./backendBlock new risk, not all legacy.
Establish a baseline, annotate changed findings and distinguish inspection failures from tool failures in pull requests.
fail-on-new: highPRODUCTION-7
Seven dimensions. No compensating averages.
A backend is not production-ready because one area is excellent. The gate keeps every dimension and every evidence gap visible.
Architecture
Boundaries, dependency direction and change isolation.
Correctness
Transactions, constraints, idempotency and error contracts.
Testing
Failure paths, integration evidence and meaningful isolation.
Security
Secrets, identity, input boundaries and dependency risk.
Performance
Async correctness, timeouts, queries and bounded work.
Observability
Structured signals, health checks and failure diagnosis.
Deployability
Reproducible builds, migrations, rollback and CI/CD.
Public signed demonstration
See the report before you trust the pitch.
We inspected a deliberately small synthetic backend. The supported checks found no failure, but the gate remains incomplete because repository source cannot prove runtime and operational properties.
GATEINCOMPLETE
Score74Coverage54.3%Failures0
Official signer · Ed25519 · independently verifiablePrivate alpha programmeYou bring You receive
Bring one real backend. Help shape the product.
We are looking for Python engineers and small teams willing to run one authorised inspection, classify the results and tell us whether the report changed a concrete engineering priority.
- An authorised Python backend
- About 60 minutes across setup and review
- Direct feedback on useful leads and false positives
- Private alpha access and onboarding pack
- A verifiable report and remediation backlog
- Direct review of the inspection outcome