After the fourteen-case real-transport integration passes, use
prepare_stage3.py --integration <output> --out <fresh-output> to bind a review
package to that evidence. The current concrete scope is
artifacts/implementation/light-lambda/parity-stage3-review-20261022T224001Z/AWS-SCOPE.md.
It contains the exact new names, artifacts, two dedicated configuration layers,
resolved narrow IAM policies and three executable CloudFormation JSON templates.
validate_stage3.py <scope-output> performs local structural/pin checks only.
The old prepare.py flow below is retained as earlier generic preparation.
The new scope requires fresh names and immutable backend :1 output readback;
an unexpected version stops execution. Actual S3 VersionIds are pending until
approved upload and must never be guessed. No deployment/test/cleanup authority
is implied by generating this package. Cleanup remains a separate explicit
decision; no TTL, bucket lifecycle deletion or automatic teardown is installed.
Future stack creation must use --on-failure DO_NOTHING so a failed deployment
does not implicitly delete trial resources before a cleanup decision.
prepare.py only reads ZIPs and writes a fresh review directory. Both templates
are CloudFormation expressed as JSON (valid YAML). They have no public ingress,
resource-based Invoke permission, VPC, denied target or delay target. No command
in the local suite invokes AWS. These are unqualified deployment candidates.
Prepare the concrete artifact scope from repository root:
rtk proxy python3 -B deploy/lambda-parity/prepare.py \
--account 964637446810 --region us-east-2 --profile YOUR_EXPLICIT_PROFILE \
--prefix parity-YOUR-UNIQUE-TRIAL \
--backend-zip /ABSOLUTE/backend.zip \
--java-zip /ABSOLUTE/java-native.zip --rust-zip /ABSOLUTE/rust.zip \
--out /home/steve/workspace/artifacts/implementation/light-lambda/FRESH-aws-reviewUse a lowercase unique prefix. Review review.json, both templates, the local
per-case report, exact ZIP/code digests, IAM policies, selected AWS cases and
estimated regional charges with the owner before any uploads or AWS actions.
deployable:false is deliberate: unresolved versions/configuration and execution
evidence must not be inferred from a template. Example parameters contain
placeholders and cannot be used as a deployment manifest.
After separate authorization, the intended order is:
- Verify STS account, explicit profile and region against the approved manifest. Upload reviewed ZIPs into a versioned trial-owned bucket; record object versions.
- Create the separate backend stack from
backends.template.yaml. Both identities use the same ZIP digest. Resolve and record immutable version outputs and actual CodeSha256, architecture, memory, timeout, environment and role policies. Directly prove modes before treating the backend as an oracle. Preserve failed evidence. - Prepare separate trial-only Java and Rust config layers mapping
/parity/a@postand/parity/b@postto those exact versions. Adapt existing native and Rust packaging tools; do not copy an accepted deployment's private configuration. Start proxy-only with application authentication disabled and mark that scope. JWT profiles require a separate equivalent-claims credential plan. Confirm Java bootstrap passes its config directory property; the Rust directory environment variable alone does not configure Java. Pin config and layer digests/versions. - Review the completed gateway manifest, then create the gateway stack using
template.yaml. Roles can invoke only the two qualified backend versions, plus write their own logs. No runner or deployment privilege belongs on those roles. - Use a separate runner role for exact qualified gateway/backend Invoke ARNs, configuration readback and the four trial log groups. Disable client retries. Recheck all pinned versions before and after the run. Record outer Invoke FunctionError independently from gateway HTTP status. Do not use the local stub runner as an AWS runner: stage 3 execution/collection adapters are pending.
- For each side/case/attempt use new bounded synthetic IDs. Collect entry and completion records for up to 60 seconds. Deduplicate by backend execution request ID, not log line count. Prove delivery with an A/B positive control from the intended version/stream. No completion is expected for FunctionError or timeout. Missing log evidence yields inconclusive, never proof of zero execution.
- Keep selected/total counts and unsupported inventory in the report. Run pet-store checks separately; this suite does not replace them. IAM denial/delay need their own scoped resources and authorization. Do not widen IAM to enable a case.
Manual cleanup requires owner confirmation at action time. Only the two named trial stacks, their generated roles, four functions and trial-owned layers are in scope. Logs and published versions have Retain policies, so stack deletion alone does not clean them up. Inventory those exact retained ARNs before separately approved deletion. Preserve evidence, original pet-store functions, accepted gateways/configuration, shared buckets and shared IAM. Retained resources may incur charges. No cleanup is automated here.
Exact version2 update scope and templates preserve all stage3 version1/layer resources and evidence. Candidate backend ZIP SHA-256 471449cbb17f78547520c1f78b12a5dda94fe6f7c2cc0619aa9346ff38c41130. Local actual-gateway/SDK/backend qualification passes; AWS validation/change sets, actual object versions and published:2 identity verification remain pending separate approval. Only dedicated stage3 resources are proposed for update; no accepted standalone gateways, public ingress, cleanup, deferred capability or JWKS scope. Never treat received-event records as SDK-wire evidence.