Signal Ledger

What this is

This page is the master record behind the thesis series. Six signals are tracked here, each with a definition, a named recurring source, the cadence that source publishes on, two thresholds — one that would confirm it, one that would weaken it — and its latest dated reading. The theses argue; the ledger keeps score.

Signals belong to the ledger, not to any single thesis. A publication updates the signals it brings evidence for and may add new ones; nothing here obliges a publication to appear. The obligation runs the other way and it is dated: the two staked calls in section 04 are graded on their resolution dates, whether they land or miss.

• 6 Signals tracked, all currently active, across six clusters Opened by Thesis 01, 17 August 2026
• 2 Staked calls open, each with a fixed grading date 30 June 2027 and 31 December 2027
• 3 Registries behind the ledger: outcome prices, pricing models, agent acquisitions A fourth, the verdict registry, opens with Thesis 02

How this ledger updates

• Publications update what they touch. A thesis or a note updates the subset of signals it has evidence for. Partial updates are the normal case: a signal a publication did not move is not a signal that was ignored.
• The bi-monthly Scorecard covers the rest. Every two months each live signal is re-read and re-scored, and “no change” is a legitimate result that gets written down. First Scorecard: October 2026.
• Staked calls are graded at their resolution dates. This is the only hard obligation in the system. A call is graded in public on its date, including when it misses.
• Retirement is written down. A signal leaves active status only with a dated line of reasoning. Silent deletion is not permitted, because a ledger you can quietly prune is not evidence of anything.

Lifecycle. Active — read at every Scorecard and by any publication with evidence for it. Dormant — retained, but its source has stopped publishing or the question is parked; no score is issued while dormant. Retired — closed with a dated reason, kept visible in the reading history. Resolved — a threshold was crossed and the question is settled; the crossing is recorded with its date.

Provenance. Every reading carries a date, the publication that logged it and a link to its source. Registry rows carry a disclosure column stating how the number is known: published-primary (the vendor's own rate card or pricing page), company-announced, press-reported, or vendor-undisclosed. Where only secondary reporting exists, the row says the price is unverified instead of repeating it as fact. Scores run −2 to +2 and are mine.

The signals

One table per cluster, one row per signal, each showing its latest dated reading. Net score across the six is +2: infrastructure and pricing move with the thesis, the consumer-transaction signal does not, and bundling moves against the claim that these layers are independently investable.

Commerce

Pricing

Margins

Authority

Structure

Energy

Staked calls

Two dated calls are open. Both were stated in Thesis 01, section 10 and are reproduced here word for word, with the citation renumbered to this page's source list.

CALL 1 — agentic checkout share exceeds 5% of agentic activity by Q2 2027 (HUMAN series). Confidence ~55%. A bet against my own tracked data: the series is flat to down across four readings while total agent volume grew. I make it because the authorization infrastructure announced between June and August 2026 has not yet had a shopping season to run through. If the share is still under 3% by Q2 2027, the consumer-side machine-transaction thesis is behind schedule and section 4's ranking is wrong — authority-to-act would be binding harder and longer than I assume.

Signal R01.1 · resolution date 30 June 2027 · graded in the Scorecard that follows it.

CALL 2 — the share of AI builders offering outcome-based pricing as their primary model stays under 30% through end-2027 (ICONIQ series, currently 23%). Confidence ~70%. Unpredictable exception costs push vendors toward hybrid pricing — platform access plus usage, with safeguards — already the pattern ICONIQ observes. The unit is ICONIQ's, not mine: no public series measures the share of contracts signed on outcomes, so the call is stated in the vendor unit that is actually published and will be scored against it. The uncomfortable part, on the record: that series has run 2% to 18% to 23% in four quarters, so the call needs a steep climb to flatten roughly where it stands ^[2]. It is the least comfortable bet here.

Signal R01.2 · resolution date 31 December 2027 · graded in the Scorecard that follows it.

Both calls are scored publicly in every subsequent publication, including if they are wrong.

Registries

Three registries sit behind the ledger. Each was a byproduct of writing a thesis rather than a separate data project, which is why it can be kept honest: the rows accumulate with time, and none of them can be faked retroactively without breaking a dated source link.

• Outcome price registry10 rows

  What vendors charge for a completed outcome — a resolved conversation, a qualified lead, a handled procedure — with each vendor's own definition of “resolved”, and the per-conversation and per-action meters kept alongside as counter-examples.

  Published rates and press-reported estimates are marked separately; unverified numbers say so.

• Pricing model registry13 rows

  How agent vendors meter, one row per product: access seats, consumption credits, per action, per conversation, per outcome — and which of those a single vendor runs at the same time. The wider view of the same question the outcome price registry asks in one category.

  Every row carries a price-disclosure column: published-primary, press-reported or vendor-undisclosed.

• Agent M&A registry6 rows

  Acquisitions in the agent stack, tagged by layer — authority, execution-outcome, execution-context, rail-routing — with announced date, status and status date, so the bundling signal can be read from rows rather than from memory.

  Prices are labelled company-announced, press-reported or undisclosed; nothing is stated as a fact the buyer never confirmed.

All three are printed below as they currently stand. Each row carries a disclosure label saying how the number is known, and one evidence badge that follows from it: MEASURED where the vendor publishes the price itself, COMPANY CLAIM where a company announced a figure it never published as a rate, and INFERENCE where no vendor rate exists and the row carries a third-party estimate or records that nothing is published. An unpublished price is written as unpublished, not filled in from reporting and left to look like a rate.

Outcome price registry

Two of these ten rows were taken by a larger vendor within five months: Forethought closed into Zendesk on 26 March 2026, and Fin sits under a $3.6B Salesforce agreement announced 15 June 2026 that has not yet closed. The category is being priced and bought at the same time.

Pricing model registry

Eight of the thirteen rows publish a rate; five do not. Seven run two or more meters at once, which is the finding this registry exists to hold: outcome pricing is arriving as an added meter rather than as a replacement for the old ones.

Agent M&A registry

Three of the six are completed; the other three are announced, agreed or pending close. Exactly one carries a price the buyer itself put its name to. Signal R01.5 reads this table, not the other way round.

A fourth registry — public claims adjudicated with their named owners, verdict and reasoning — opens with Thesis 02. It stays unpublished until every claim in it is quoted with an owner, because adjudicating an unattributed claim is arguing with a strawman.

Open core

The current state of this ledger is public and stays public: definitions, thresholds and the latest reading for every signal, free to read and to cite with attribution. That is the core, and the core is never closed.

What may eventually be sold is depth rather than access: the full reading history behind each signal, machine-readable exports of it, a queryable interface over it, alerts when a threshold is crossed, and work applied to one organization's own question. If that layer ever exists, this page keeps working exactly as it does today.

Corrections, better sources and threshold arguments are welcome on LinkedIn.
The Agents & Margins briefing (English) is launching — until then, follow on LinkedIn

Sources

1. 1.HUMAN Security — State of Agentic Traffic, June 2026 edition, published 6 July 2026, with the April and May 2026 editions. No July 2026 edition had been published as of 17 August 2026.
2. 2.ICONIQ — 2026 State of AI Report: The Builder's Economy, July 2026 (~300 executives), and State of AI: Bi-Annual Snapshot, January 2026 (N=297 for pricing, N=269 for margins).
3. 3.Bessemer Venture Partners — The State of AI 2025, 13 August 2025.
4. 4.NIST NCCoE — concept paper, Accelerating the Adoption of Software and AI Agent Identity and Authorization, posted 5 February 2026; comments closed 2 April 2026.
5. 5.Agentic AI Foundation (Linux Foundation) — Model Context Protocol specification 2026-07-28, with authorization hardening; Microsoft Entra Agent ID reached general availability in April 2026 and Agent 365 on 1 May 2026.
6. 6.Zendesk — completed acquisition of Forethought, 26 March 2026, taken as the anchor release for the bundling signal; the other transactions behind the reading are recorded row by row in the agent M&A registry above, each with its own source and disclosure label.
7. 7.Gartner — worldwide AI-optimized IaaS spending forecast, press release 10 August 2026.