Engineering reference

Reading an Auto-Scatter Workflow Dashboard

A measurement template for comparing prompt-hook settings, responses per prompt, token usage, billing records, and input scores.

Book a 30-minute call Compare services

Methods, definitions, and examples for the topic discussed here. Check applicability against the task and operating conditions.

Mario Alexandre · March 26, 2026 · token-savings auto-scatter sinc-llm

This article presents fields and measurement checks for assessing a prompt hook’s operating costs.

This is a measurement template for a prompt-hook comparison. The fields describe records to collect; no observed result is supplied here.

sinc-LLM — what the hook injects before every prompt

x(t) = Σ x(nT) · sinc((t - nT) / T)

Measurement Template

// Measurement template — choose a comparison window and record hook settings

user_prompts_total

Count user prompts in the comparison window.

assistant_responses_total

Count assistant responses in the same window.

exchange_rate_measured

Divide assistant responses by user prompts for the comparison window.

exchange_rate_baseline

Record the same ratio for a declared baseline window.

output_tokens_7d

Record output tokens for the comparison window.

total_tokens_7d

Record total billed tokens for the comparison window.

actual_cost_7d

Record billed cost for the comparison window.

projected_cost_at_baseline

Estimate baseline cost using a declared workload and pricing method.

savings_7d

Compare baseline estimated cost with comparison-period billed cost.

haiku_scatter_calls

Count scatter-hook calls for the comparison window.

haiku_cost_7d

Record scatter-hook billed cost for the comparison window.

cost_per_scatter_call

Divide scatter-hook billed cost by scatter-hook calls.

savings_per_scatter_call

Compare per-call costs using the same workload and pricing method.

roi_haiku_spend

Compare any cost difference with the scatter-hook overhead.

snr_before_scatter

Record the baseline input score using a declared scoring rule.

snr_after_scatter

Apply the same scoring rule to the comparison inputs.

quality_observations

Record the number of reviewed prompt-response pairs and the quality criteria.

Recording Responses per Prompt

Exchange rate means total assistant responses divided by total user prompts. It is a simple ratio. Record responses and prompts separately for a declared baseline window and comparison window.

For the comparison window, count user prompts and assistant responses, compute their ratio, and record hook settings and template changes. Assess each workload using its own records.

Estimating Baseline Cost

Estimate baseline cost from a declared token volume, comparison method, and applicable API prices. Keep any estimate separate from the billed comparison-period cost.

Billing records need their workload, model settings, token usage, and measurement window before a cost comparison can be assessed.

Scoring Inputs

An input SNR comparison needs a declared scoring rule, prompt versions, workload, and measurement window. Review output quality separately against the task and supplied sources.

The author’s input SNR score describes the chosen prompt-content measure. A high score does not establish first-try correctness or remove the need for source and task checks.

Why I'm Publishing This

A workflow cost comparison needs a defined workload, token records for both versions, model settings, responses per prompt, and a measurement window. Keep the original records with the comparison.

The article describes a logging system in the open-source package. Record the workflow and measurement method with any results. Leave a comment and I will send you the GitHub link.

// Production AI Engineering

Build AI systems that hold up in production.

sinc-LLM designs, audits, and stabilises production AI infrastructure: from vendor evaluation and cost accountability to incident controls and MCP architecture.

See what we do →

Scope and context

sinc-LLM About Blog Services