Back to Language as Infrastructure
working paper2026Deployment manuscript

Beyond Controlled Comparison: Deployment Properties of Script-Aware ASR for N'Ko

This deployment paper moves beyond controlled comparison and looks at the properties that matter once script-aware ASR is expected to run on actual devices: latency, memory, failure gating, and the gap between an offline model and a live microphone.

Paper workspace

Live draft structure

working-draft

Artifacts

Draft PDF

Deployment draft. Useful, but live calibration status has changed since this render.

Open artifact

Final split-paper render

Final split-paper artifact for the AGP/deployment branch.

Open artifact

Editable source

Draft PDF exists. It should be updated with the current iPhone live-gate evidence before release.

Source anchors

nko-brain-scanner/paper/current/paper5_deployment.tex

nko-brain-scanner/paper/final/04-agp-deployment/paper.tex

Method tags

deploymenton-device ASRquality gates

Ingest intersections

deploymentiphonecoremlasrcalibration

Status

Drafted; venue-split set prepared (four-paper release plan).

Key claims

01

A good offline result does not automatically become a usable live recognizer.

02

Deployment exposes timing, memory, and gating problems that benchmarks hide.

03

The iPhone harness should be treated as a measurement surface, not a demo claim.

Public reading note

Drafted; live calibration proof still pending.

Standard skeleton

What this paper must keep proving

Schema

problem

An offline model can look good while failing as a live user-facing recognizer.

method

Track latency, frontend parity, overfire gates, packet review, and promotion boundaries separately.

implementation

iPhone harness, CoreML encoder/head, live mic capture, calibration packet export.

data

Replayable live packets and reviewed labels, not machine output treated as truth.

evaluation

Latency, no-garbage gating, packet replay, reviewed transcript correctness, and training admissibility.

references

Mobile ASR deployment, CoreML, speech calibration, dataset governance.

openQuestions

How many reviewed packets are needed before live recognition becomes a credible product claim.

Checkpoints and references

Proof chain

paperpending

Claim checkpoint

central-claim slot

Every central claim must point to a proof anchor or remain labeled as speculative.

implementationpending

Implementation checkpoint

implementation-map slot

Every method should identify the code path, harness, schema, or protocol that embodies it.

experimentpending

Evidence checkpoint

evidence-manifest slot

Every reported result should point to run IDs, packet IDs, data snapshots, commits, or review artifacts.

external-referencepending

Reference checkpoint

references slot

Every external claim should resolve to a cited paper, benchmark, standard, or documented prior system.

paperpending

Release checkpoint

release-gate slot

Every PDF needs a named condition before it can move from draft to citation-ready.