PHANTOM ENGINEERING FEEDBACK: METHOD AND LIMITS
Publisher: AshlrAI, Inc. | Observation date: 7 October 2026

HISTORICAL CI
The historical GitHub run identified in source-metadata.json completed with all
15 jobs successful. Its UTC interval was 04:16:39 to 04:55:43: 39 minutes and
4 seconds. General 3 fixtures accounted for about 90% of Vitest wall time;
isolated fixtures accounted for about 98%.

Fixture, transform, setup and import counters can overlap. They are not additive,
mutually exclusive wall segments. Queue delay and job wall are separate. Fixture
share is not share of the complete release interval. Jobs ran concurrently.
There is no achieved before/after total release improvement claimed here.

CONTROLLED PROJECTION
Node 22.22.3. Four alternating baseline/candidate pairs used the same real private
Git and preparation store, with controlled delivery, run and artifact witnesses.
The original campaign projection ran during mock recovery in both variants.
The baseline also performed an immediate redundant post-recovery projection.
Output equality and unchanged fixture state were checked.

Redundant post-recovery projection reads fell from one to zero. The removed
projection had a median of about 97 ms. Whole helper medians were roughly 1.5 s
and 1.4 s. Exact samples and summary values are in the CSV and source metadata.
Local background load can affect timing. This does not measure full mission
performance, provider latency, billed cost or a population estimate.

BEHAVIOR ACCEPTANCE
Separate genuine private-home and CLI setup/mission modules passed three cases:
two setup cases and one mission covering two scopes, restart and Stop. There
were no provider calls. Deadlines stayed unchanged. The actual mission lasted
about 9.3 min; the combined run lasted about 15 min. This confirms the observed
acceptance behavior, not an end-to-end speed gain.

PREVIEW AND RELEASE STATUS
The phone image is a synthetic controlled component fixture from the 3.24.4
candidate. No provider calls were made. Its illustrative numbers are not
benchmark measurements, account usage or model decoding speed.

Candidate 3.24.4 is undergoing release qualification. Published and installed
3.24.3 remains separate. Source metadata contains exact source references;
CSV retains numeric precision while charts use roughly two significant figures.
The cover is AI-generated conceptual artwork, not a documentary image.

PRIVACY
Only aggregate measurements, public source identifiers and a synthetic preview
are included. No private prompts, account names, credit balances, customer data,
credentials, home paths or runtime ledger contents are published.