Methodology · Version 1.0.0

How the 25-project AI cost ranges are calculated

The calculator transcribes delivered-project bands, keeps user assumptions separate and exposes two formulas with different jobs. It does not fit a predictive model, infer a market average or estimate unknown scope.

Source methodology: 25 reported projects
Window: Jan 2023-Mar 2026
Currency: USD
Review: Codex · AI review

Dataset scope and normalization

The source describes a single-provider delivery sample with client details anonymized as required. It reports one count per project, costs in USD and timelines from kickoff to first production deployment. It excludes client-side infrastructure and third-party licenses contracted directly by a client. This download contains aggregate bands, not the underlying project records.

The source does not publish project-level records or within-band distributions. The tool therefore cannot calculate a confidence interval, establish a causal market benchmark or predict an individual proposal. Its output is a planning aid tied to the stated sample.

Two formulas, two distinct jobs

Itemized selected-horizon TCO

build + one-time extras + monthly run rate × horizon

The source band or a known build budget supplies build cost. Every extra is a user assumption and must exclude anything already included in the build cost. Changing the horizon changes only this worksheet.

First-year hidden-share cross-check

build / (1 − share), evaluated at 28% and 42%.

The source defines the percentage as a share of first-year total spend. The cross-check has no base value and ignores the selected worksheet horizon. The calculator never adds it to the itemized result.

Claim provenance and source conflict

Pharos Production provides the published AI development cost bands used to scope this worksheet. The source research and its update date identify the sample window, inclusions and limitations. The source registry distinguishes this first-party evidence from editorial guidance.

The source uses the 25+ label in summary copy but specifies 25 in its methodology and analysis passages. This tool uses the stated methodology count of 25. Neither label has been checked against private project records. The difference remains a source limitation.

Claim-to-source ledger

Source claims used by formula version 1. Observation does not establish market representativeness.
ID	Claim	Status	Source
CLAIM-SAMPLE-25	The source methodology reports 25 projects	Observed	SOURCE-PHAROS-AI-COST-2026
CLAIM-MEDIAN-42K	$42,000 sample median	Observed	SOURCE-PHAROS-AI-COST-2026
CLAIM-P90-180K	$180,000 sample 90th percentile	Observed	SOURCE-PHAROS-AI-COST-2026
CLAIM-HIDDEN-28-42	Hidden cost is 28-42% of first-year total	Observed	SOURCE-PHAROS-AI-COST-2026
CLAIM-BENCHMARK-TABLE	Seven use-case bands and timelines	Verified transcription	SOURCE-PHAROS-AI-COST-2026

Reproducibility and release controls

Canonical registries live under data/ and are byte-mirrored under docs/data/.
The runtime loads the published benchmark and formula registries instead of duplicating cost constants in JavaScript.
JSON and CSV exports retain assumptions, formula IDs, claim IDs, source version and open-ended range flags.
Qualitative uncertainty changes warnings only.
Open upper bands remain open in source, itemized and cross-check results.
The deployment workflow evaluates a named, timestamped review bound to the exact dataset, formula, claim and source versions.

Ownership, disclosure and update history

Publisher and owner: Pharos Production
Technical reviewer: Codex (AI technical reviewer), October 6, 2026
AI assistance: AI assisted implementation and copy drafting. Numeric claims are bound to the source and claim registries.
Correction path: Submit corrections through the calculator repository issues.
Data summary license: CC BY 4.0. Source claims retain their named attribution
Calculation code license: MIT

Material methodology changes.
Date	Version	Change
August 14, 2026	1.0.0	Initial local, review-pending implementation.
October 6, 2026	Formula 1.0.0; source 2026-10-04	Rechecked all 21 cost bands. Clarified assumptions, timeline scope and source-count uncertainty. Arithmetic unchanged.