Cloud egress rate sourcing methodology

Principles

1. Path before price. Name the traffic shape — who pulls, which network edge, which SKU — before any rate. See path template.
2. Unknown until verified. Cells without a dated provider cite stay Unknown until verified. No invented list prices.
3. No scorecrowns. We do not crown a winner from partial public pages or marketing slides.
4. No guaranteed savings. Estimators are synthetic education, not commitments or procurement advice.
5. As-of dating. Pair shown rates with retrieval date and region when a human verifies a cell.

Preferred sources

• Provider official pricing / calculator pages (region-qualified).
• Public pricing APIs only when documented and versioned.
• Customer invoices (anonymized lessons) — never scraped private consoles.

Each compare spoke ends with dated source footnotes under its path matrix. Verified docs may drop the Unknown badge; pricing pages usually keep it until a human re-checks the SKU for the reader’s region.

Pricing snapshot & as-of discipline

Every public rate cell on PipeToll is a pricing snapshot, not a live quote API and not a procurement commitment. A snapshot means: a human retrieved a figure from a named official source, recorded the retrieval (as-of) date, and scoped the figure to a region and SKU path class. If any of those pieces is missing, the cell stays Unknown until verified — we do not backfill folklore $/GB to look complete.

Explicit as-of language

• As-of date. Shown rates pair with a retrieval date (ISO YYYY-MM-DD in Sources footnotes or spoke notes). “Current” without a date is not enough.
• Region + SKU. Internet DT-out in us-east-1 is not interchangeable with cross-region replication or NAT processing in another AZ. Name the meter before the number.
• Path class first. Snapshot arithmetic always follows a path template (who pulls, which edge, which bill line). Logos are not meters.
• Staleness honesty. Provider pages change. A dated snapshot can become Incomplete again on the next review; we prefer an honest Unknown over a silent stale number.

Source provenance for egress figures

Provenance is a ladder. Higher rungs are preferred; lower rungs never invent precision the source does not give:

1. Official provider pricing or calculator pages — region-qualified public docs (AWS, Azure, GCP, Cloudflare, Backblaze, and peers). Cite the URL and as-of date in the spoke Sources footnote.
2. Documented, versioned public pricing APIs — only when the provider publishes them for the SKU; still date-stamp the retrieval.
3. Anonymized invoice lessons — shape literacy (“this bill line behaved like InterZone, not Internet”), never scraped private consoles or pasted as sitewide list prices.

We refuse: blog roundups, marketing slides, unsourced “typical $/GB,” averaged multi-region mashups, and scorecrowns that blend storage + requests + egress into one winner. If the only available figure is undated or region-ambiguous, leave the cell Unknown and point readers to calculator Mode A/B/C so they can paste a rate they verified themselves.

Egress-specific note: Internet egress, cross-AZ / InterZone, ELB cross-zone, NAT hourly + processing, PrivateLink, peering/TGW, Direct Connect / Interconnect, S3 CRR, and CDN origin fill are different meters. A snapshot for one meter never substitutes for another — see the AWS egress cost path checklist and Cost Explorer decoder before any GB × $/GB paste.

As-of / provenance checklist (egress figures)

Before any public matrix cell shows a number instead of Unknown until verified, a human should be able to answer yes to each of the following. Missing any item keeps the cell Unknown — we never invent $/GB to look complete:

1. Named URL. Official provider pricing or calculator page (or documented public pricing API) for the SKU, not a blog roundup or marketing slide.
2. As-of date. Retrieval recorded as ISO YYYY-MM-DD in the spoke Sources footnote. “Current” or “as of today” without a date fails the checklist.
3. Region + meter class. Region (or zone) and path class named — Internet DT-out ≠ InterZone ≠ NAT processing ≠ CDN origin fill ≠ CRR. Logos are not meters.
4. Currency + unit. $/GB (or hourly / request) as the provider states it; no silent unit conversion from “typical” folklore.
5. Staleness plan. Next review trigger noted (provider page change, region launch, free-tier rewrite). A dated snapshot can return to Incomplete; prefer honest Unknown over a silent stale figure.

Sitemap lastmod on PipeToll follows the same honesty: we bump lastmod only for URLs whose public content actually changed in a pack — never a sitewide fake freshness sweep. That pairs with Unknown-until-verified discipline: freshness signals and rate cells both refuse invented completeness. Cross-read About for who writes PipeToll and FAQ for the same no-guarantee posture.

When a cite arrives incomplete (undated, region-ambiguous, or SKU-unclear), contributors should still email founder@masterengine.ai — we will either date-stamp a cell after verification or keep Unknown. Incomplete cites become Sources footnotes with the Unknown badge, not invented list prices.

Provenance vs the deepened spoke set (batch-35)

Second-pass depth on existing spokes does not invent $/GB. Each thickened matrix still ends with Sources footnotes and keeps rate cells Unknown until verified until a human pairs a region-qualified official page with an as-of date. Use this map when reading a deep spoke so sensitivity twins and worked tables never get mistaken for quotes:

• Direct Connect / hybrid interconnect — DX vs Internet: port hours and private transfer are different meters from Internet DT-out; residual NAT stays on its own lines. Cite the interconnect SKU page, not a blog “typical DX $/GB.”
• Cross-AZ / InterZone — cross-AZ vs cross-region and ELB cross-zone: HA chatter and LB cross-zone GB are not Internet egress. Numbered tables and on-vs-off twins teach meter separation; CE decoder knit points at opaque CUR labels without filling folklore rates.
• Object-store egress-friendly paths — R2 vs B2, S3 vs R2, S3 vs GCS: download-cap / ops worked tables and private-path twins are arithmetic shapes. Paste only provider-dated figures into Mode A; leave Incomplete cells Unknown.
• Cost Explorer / CUR decoding — CE data-transfer line decoder beside the AWS path checklist: label → path-template mapping only. Decoder literacy never substitutes a dated pricing cite.
• Multi-cloud path templates — AWS vs GCP egress: Internet vs InterZone twins and DX/Interconnect siblings keep FinOps apples-to-apples; region-qualified cites stay per-cloud, never averaged mashups.
• Free-tier cliffs — AWS Internet free-tier cliff and Azure & GCP cliffs: below/above tables and grant/allotment sensitivity teach remainder math. Cliff pedagogy is not a published complimentary $/GB — verify the allotment on the provider page for the reader’s account class and as-of date.

Batch-35 refreshes this methodology page so provenance language tracks the deepened spoke set without claiming Sites Ready, without inventing list prices, and without adding new /compare/* URLs. Path templates still come before prices; Unknown until verified remains the default honesty posture.

Worked Unknown example

Suppose a reader wants Internet egress for “object store → public client” at 5,000 GB. The matrix lists the path class and links to the provider pricing URL with an as-of date, but the $/GB cell stays Unknown until verified because PipeToll has not pasted a region-qualified figure into the package. The reader opens calculator Mode A, pastes the rate they verified, and keeps the Unknown badge in their own notes until that paste is dated. That is the intended FinOps-lite loop — not a silent default that pretends to be a quote.

A second path: NAT processing plus Internet DT-out. Model hours and GB as separate meters in Mode B. If hourly or processing rates are not cited to today’s provider page for that AZ footprint, leave them Unknown — do not average three blogs into a fake precision.

A third path: CDN origin fill. Estimate fill with CHR in Mode C, then apply the origin provider’s dated $/GB. Edge bandwidth and origin egress remain separate bill lines. Miss-heavy APIs often look “cheap on CDN” until origin fill is named.

Path examples we refuse to invent rates for

• NAT Gateway vs Internet Gateway — triple-charge stack; verify hourly + processing + Internet separately.
• Cross-AZ vs cross-region and ELB cross-zone — HA / LB chatter is not Internet egress; keep meters apart.
• Direct Connect vs Internet — hybrid interconnect port hours ≠ public DT-out; residual NAT stays separate.
• R2 vs B2, S3 vs R2, and CDN origin fill — download caps, ops classes, free-tier cliffs, and CHR dominate more than a single blog $/GB.
• AWS vs GCP and CE decoder — multi-cloud templates and CUR label maps never invent a shared list price.
• AWS free-tier cliff / multi-cloud cliffs — allotment pedagogy is not a published complimentary rate cell.

Out of scope

Committed-use discounts, private credits, and unpublished enterprise schedules are out of scope for public matrices. We also refuse scorecrowns that blend storage, requests, and egress into one “winner.”

How to contribute a cite

Email founder@masterengine.ai with provider URL, region, SKU name, as-of date, and the path template the rate applies to. We will either date-stamp a cell or keep Unknown — never invent.

Educational only. Contact founder@masterengine.ai. See the glossary for shared vocabulary.