Issue #794 adoption and timing evidence
Source of truth
This artifact records only measurements actually executed for the Durable operational-assessment adoption guide. Targets from the approved design are listed separately from observations so an unrun journey cannot be mistaken for evidence.
Measurement boundaries
| Journey | Start | Stop | Target |
|---|---|---|---|
| Cold local proof | Invocation of bash examples/durable-postgresql/run-local-proof.sh in a clean prepared checkout |
Final [ok] local operational-assessment proof completed checkpoint after container cleanup is armed |
Five runs; p50 under five minutes; no run above seven minutes |
| Prepared existing-host adoption | Developer opens a restored schema-10 checkout and begins replacing handwritten readiness/admission logic | Exhaustive switch compiles and its focused tests pass | Five runs; median under two minutes; no run above three minutes |
| Migration-inclusive upgrade | Migration-owner credentials and canonical role recipe are available | Schema 9 -> 10, role reconciliation, old/new smoke checks, and admission proof pass |
Under five minutes |
| Findability | Unfamiliar developer opens the repository root | Developer locates the first-run command, direct-admission example, and distinct ASDUR103 permission remedy |
Under two minutes |
Run automated command timings with:
bash Durable/evidence/measure-issue-794-adoption.sh LABEL RUNS -- COMMAND [ARG ...]The harness emits Markdown rows and the command output; it does not mutate this file. Copy only rows from completed
runs and record the exact runner profile. Human adoption and findability journeys must be timed by a participant and
must not be replaced with build duration. Each command run has a 420-second deadline; set
APPSURFACE_DURABLE_MEASUREMENT_TIMEOUT_SECONDS to another whole-second bound from 1 through 86,400 when the
documented journey has a different target. A timed-out command and its process group are terminated and recorded as
timeout.
Runner profile
- Date: 2026-09-10.
- Host: macOS 26.6.2 (25G83), arm64.
- .NET SDK: 10.0.102.
- Docker Engine: 29.7.2.
- PostgreSQL image:
postgres:16.5@sha256:53f3e608f9475ce120ced2d0f430b89458d7faa28530e0b0977a6af64d294877. - Source revision:
0710a18ea08efe631213185cd53ff816cfd1a1a9, the first immutable commit capturing the measured #794 implementation and this evidence; its parent is024e294f49a1b354411051680b51798d80512666. - Cache state: the Docker image and NuGet/source build caches were present. Each local-proof run used a new container and empty database. Each packed-consumer run used a new temporary package feed and consumer package cache.
Observed measurements
These are automated command durations, not human adoption or findability timings:
| Journey | Invocation/run | Started (UTC) | Elapsed seconds | Result |
|---|---|---|---|---|
packed-consumer |
initial/1 | 2026-09-10T09:19:34Z |
19 | fail: two invalid ArgumentOutOfRangeException overloads |
packed-consumer |
fixed/1 | 2026-09-10T09:20:17Z |
18 | pass |
local-proof |
initial/1 | 2026-09-10T09:20:45Z |
23 | fail: -m:1 was forwarded by dotnet run to the CLI |
local-proof |
fixed/1 | 2026-09-10T09:22:22Z |
15 | pass |
packed-consumer |
controlled-build/1 | 2026-09-10T09:24:11Z |
20 | pass |
local-proof |
repeat/1 | 2026-09-10T09:28:27Z |
10 | pass |
local-proof |
repeat/2 | 2026-09-10T09:28:37Z |
11 | pass |
local-proof |
repeat/3 | 2026-09-10T09:28:48Z |
11 | pass |
local-proof |
repeat/4 | 2026-09-10T09:28:59Z |
10 | pass |
packed-consumer-final |
final/1 | 2026-09-10T09:35:59Z |
15 | pass |
The passing packed-consumer run packed all six source packages into a disposable feed, restored fresh consumer projects, and compiled/ran the adopter, Provider, and PostgreSQL consumers. Its PostgreSQL consumer covered all four attempt kinds, asserted zero execution calls for every returned non-completed outcome, and verified one shared custom pump singleton behind both public interfaces. The explicitly labeled contract fakes also propagated pre-call cancellation without entering execution and preserved the same exception instance; these fake checks are API-consumer evidence, not PostgreSQL execution evidence.
The passing local proof used a fresh PostgreSQL 16.5 database. It applied migrations 0001 through 0010, reran the
canonical role recipe, initialized a runtime epoch, accepted Work/Flow/Schedule inputs, resolved the legacy and
admission APIs to one PostgreSQL singleton, returned Completed from the real admission call, observed Healthy,
exercised drain/resume and the hosted worker, and verified startup performed no DDL. It was a prepared-cache
engineering run. Five consecutive fresh-container/fresh-database passes measured 15, 10, 11, 11, and 10 seconds
(p50 11 seconds; maximum 15 seconds). They establish repeatability and are comfortably below the cold-run target, but
they do not replace five clean-checkout cold measurements because source, package, and image caches were present.
Required evidence still unrun
- Five cold local-proof journeys.
- Five prepared existing-host adoption journeys.
- One migration-inclusive
9 -> 10journey including role reconciliation and the actualv0.2.0-preview.8rollback smoke. - One unfamiliar-developer findability journey.