For a founder who hands real delivery work to agents

A ledger of who decided what while your agents did the work

[ILLUSTRATIVE: SCENARIO] A founder hands the actual work to agents — the drafting, the building, the verifying, the shipping — and it goes faster than anything a human team could do alone. Then the board asks one question the founder cannot answer: "Who decided that?" Not who wrote the code. Who decided the pricing posture, the name, the thing that reaches a customer. If the honest answer is "an agent, somewhere, and I’m not sure which one, or when," the founder doesn’t have an operating company — they have a liability wearing a company’s clothes.

See the ledger

No pricing on this page. Engagements are scoped and priced once we understand the work.

What changes

Every human touchpoint gets classified. Every delivery gets registered.

Not reconstructed afterward from memory or a half-remembered chat log — appended in the same session the decision happened in. A small, observed taxonomy of the moves a human actually makes at a decision boundary: steer, approve, constrain, redirect, policy, standard, and more, each one stating what the act actually changed, not just what was said.

The outcome is a record you can hand to a board, an auditor, or a buyer in due diligence — not a claim that governance happened, but the trail that shows it did, with the mistakes left in.

What it is

A taxonomy, a delivery register, and an append discipline.

Touchpoint taxonomy

Not a fixed enum defined in advance — an observed, growing vocabulary of the moves a human actually makes at a decision boundary: identity, disclosure, consent, commitment, taste, and frame, expanded per-touchpoint into tags like direction, constrain, redirect, and approve.

Delivery register

Kept separate from the touchpoint ledger, on its own append discipline: what shipped to production, what stayed a decision artifact, what is still in flight — plus a plain statement of scale, so "agents delivered" never implies more than actually ran.

Append-per-touchpoint discipline

One ledger row per human touchpoint, one register line per delivery, in the same session or any future one. The ledger is never rewritten to smooth over a contradiction — when a decision reversed mid-engagement, both rows stay.

Provenance conventions

Every derived document carries which agent produced it, what it was derived from, and a claim status — proposal, verified, requires approval. Claims inside carry citations rather than being asserted bare.

The ledger itself

All 58 touchpoints. Real, redacted, scrubbable.

This is a real ledger from a real, continuing engagement — our own reinvention of this site — with every removal marked rather than quietly folded away: no names, no repo paths, no PR or ticket identifiers, no pricing. The three documented mistakes stay in, because the honesty of the arc is the proof, not a liability.

All 58 recorded human touchpoints, in order. Deliveries and the three documented mistakes are noted in the last column.
#	Class	The human's move	What it changed	Delivery / note
1	DIRECTION	Set the program mandate: move from a portfolio site to a services and solutions launch, run at multi-agent scale.	Started the program and authorized multi-agent scale.
2	DIRECTION	Called for an incubation lens where early ventures live, and only there.	Created an incubation section, solving a venture quarantine problem.
3	DIRECTION	Required a creative-production line be accommodated on the site.	An unnamed exhibit was added, gated behind a reel.
4	DIRECTION	Asked that a money-movement claim be added.	A grounded claim was added at three scoped placements.
5	DIRECTION	Called for a white-glove, professional-services engagement shape.	A fourth engagement shape was added to the catalog.
6	CONSENT	Approved the plan in full and authorized Phase 1 to begin.	The build was authorized and guardrails were set.
7	CONSENT BOUNDARY	Restricted deployment to a preview environment only.	Production was explicitly withheld, and the ship gate enforced it in code.
8	CONSENT	Approved the preview build and authorized opening it for review — still withholding production.	The change was opened for review and a review gate ran.
9	NAMING	Confirmed a canonical product name after a naming collision surfaced.	The collision was resolved; the name was carried into the next phase.
10	COMMITMENT	Confirmed a set of proposed service-level commitments.	The commitments became renderable, public facts and unblocked the production build.
11	CONTRADICTION-HANDLING	Asked for a re-pass while appearing to contradict an earlier decision, then later confirmed the earlier decision still stood.	The system surfaced the contradiction instead of guessing — the disagreement was recorded, not silently resolved.
12	CONSENT + SCOPE	Ruled on the re-pass: the earlier commitment stands, one legacy item is pulled forward, one is deferred.	Legacy exposure was triaged by the accountable owner, not by default.
13	DELEGATION	Asked for an inventory of publishable specimen candidates plus a redaction recommendation.	A specimen was selected and a wall-safe framing was chosen.
14	PUBLICATION CONSENT	Cleared the specimen for publication and authorized the merge.	Phase 1 shipped to production.	Phase 1 shipped — home, engagement tiers, specimen document, start page
15	DIRECTION	Gave a plain go-ahead to continue.	Phase 2 was authorized.
16	INQUIRY	Asked for the reasoning behind a naming decision before ruling on it.	An informed decision was required before any removal was made.
17	TASTE/SCOPE	Directed that a competitive-detail section be removed from a public page.	A voice-map and landscape comparison were cut from the page.
18	IDENTITY	Ruled that no individual founder is ever named publicly — the company speaks as "we".	Company voice was enforced site-wide; zero personal-name renders were verified.
19	DISCLOSURE REVERSAL	Reversed an earlier openness posture for one venture.	All openness claims for that venture were retired to match observed reality.
20	CONSENT	Approved the naming rationale and authorized the merge.	Phase 2 shipped to production.	Phase 2 shipped — product, design, company pages; voice pass enforced
21	FRAME CORRECTION	Corrected the framing from a list of ventures to a set of capabilities.	Venture-as-product framing was killed and capability research was launched.	Two-doors positioning proposal adopted and executed
22	TASTE VETO	Rejected a hero line as badly written and asked for alternatives.	Six candidate lines were produced, pending a pick.
23	FRAME CORRECTION	Pushed back a second time, reframing the whole offer as a company-operating-system model with three delivery modes.	The taxonomy that anchors the offer catalog was set, and a research effort was dispatched.
24	GOVERNANCE	Required the work be tracked so it could survive interruption.	A task list plus recovery notes and memory layers were added.
25	CONSENT	Adopted a corrected shortlist of offerings and approved two proof builds.	Two manufacture-proof builds were green-lit.
26	TRANSPARENCY	Asked to see every candidate that was considered and cut.	A full candidate map with reasons for each cut was delivered.
27	INQUIRY	Asked to be walked through the open questions rather than rubber-stamping a summary.	Informed decisions replaced a rubber stamp.
28	COMMERCIAL AUTHORITY	Set the pricing posture: engagement-priced, scoped inside a document, never published as a figure.	The document itself became the quoting vehicle.
29	POLICY EXCEPTION	Adopted all five offerings with one scoped exception, recorded rather than left implicit.	One exception was written down beside the rule it excepts.	Offering catalog adopted — ten dossiers, sell-first shortlist
30	CREATIVE DIRECTION	Requested five creative-production packages be drafted.	Five packages were authored, pending a review verdict.
31	TASTE + QUALITY AUTHORITY	Elevated a report page to a craft exhibit and ordered the linked design-system findings executed.	The design mandate was amended mid-build and downstream fixes were dispatched.	Design-system schema and attestation gate shipped
32	META-GOVERNANCE	Asked that this very engagement be tracked with the human-in-the-loop arc visible.	This ledger began.
33	CREATIVE CONSENT	Approved the five creative packages verbatim and authorized generation.	The packages were cleared and a generation pipeline was authorized.
34	TASTE DECISION	Picked the final hero line and authorized shipping it.	A specific hero line was merged to production with the full test suite green.
35	DIRECTION + SCOPE	Directed a single coordinated build across multiple outcome prototypes and business models.	A multi-agent workflow was launched.
36	STRATEGY + DISCLOSURE REVERSAL	Reversed course: keep the real report private, publish only an illustrative marketing version, and make the intake form actually functional.	A prior public-fixture plan was rescinded before it shipped, and two parallel builds were dispatched.	Conformance engine, phases 0 and 1, shipped and dogfooded
37	DIRECTION + TIER EXCEPTION + GATE DEFINITION	Directed one internal venture candidate toward a proof of concept, assigned it a dedicated agent, and set explicit risk gates it must clear before going wider.	A thesis, risk-gate metrics, and a built proof of concept were added to the internal pipeline.	Internal venture candidate — thesis, risk gates, proof of concept
38	PUBLICATION CONSENT	Ordered a review-and-merge pass on everything pending.	Three parallel review gates came back clean and were merged, each with its own record.	Three-lane review-and-merge sweep, all clean
39	DIRECTION	Requested a competitive intelligence capture on a rent-versus-own frame.	A dedicated intelligence pass was dispatched and mapped against the company’s own delivery-mode framing.
40	TASTE VETO + SKILL MANDATE	Judged an internal console’s UI as chaotic despite correct components and ordered a design-sense pass with multiple review lenses.	A written design contract was produced and a redesign wave was scheduled to follow synthesis.
41	CAPABILITY-BUILDING MANDATE	Directed that reusable app-building standards be extracted from this build and raised to a much higher bar.	A versioned skill document was written, each rule citing the failure it prevents.
42	SCOPE + LIFECYCLE	Set the rule that a new standard stays local until proven, and graduates to the shared system only after.	A graduation gate was written: proven builds, one amendment-free build, explicit approval, then promotion.
43	STEER	Asked that the remaining work actively use tracked workflows and close out with merges rather than drift.	The rest of the session was reframed as closure work, with a tracked workflow replacing ad hoc agent runs.
44	CONTINUE	Gave a plain go-ahead to continue.	The next redesign workflow was authorized the moment the prior fix pass cleared.
45	CONSTRAIN + AUTHORIZE	Authorized a merge conditional on a re-score and strictly limited to the files actually touched.	A pinned scope contract was set, with an outside-scope audit re-run at merge time.
46	REQUIRE	Required a private, locally viewable build that also works on mobile.	A private-network build was stood up and a phone-viewport check was added before any merge.
47	STEER	Asked what else on the site still had an open proof gap.	Two site items were unblocked as a new, disclosure-audited branch wave, gated from production.	New offering surfaces landed on a disclosure-audited branch
48	DESIGN DIRECTIVE	Directed that a console use drawers, multi-column layouts, and modal windows where appropriate.	The directive was relayed live to the build and written into the standing design contract.
49	REQUIRE	Required that any separate branch be reachable on its own address for mobile review.	A standing rule was set: each in-flight branch gets its own reachable preview.
50	REQUIRE	Required that a console be reachable only behind a personal login.	Access-gated hosting was recorded as the target, pending a separate go-ahead on billing and domain setup.
51	STEER	Directed that work continue with cost-aware model and effort choices after a usage limit was hit.	Remaining work moved to cheaper model tiers with scoped effort, and interrupted work was checkpointed before resuming.
52	REDIRECT (strategy)	Rejected the pitch’s technical, artifact-first framing and ordered a story-first, business-register rewrite with a sharply higher craft bar across the whole site.	The pitch was reframed around narrative first, technical proof demoted to supporting evidence; a positioning document was commissioned before further build.
53	APPROVE	Approved the new voice on a set of sample rewrites and ordered every page rewritten in it.	The new register became binding and a site-wide copy pass was launched.
54	POLICY	Directed that issues like deployment or document problems be self-healed rather than escalated by default.	Self-healing at the root cause became the default posture for every workstream.
55	STANDARD	Set a standing rule that every surface of every company product must ship an interactive proof element.	A portfolio-wide bar was set and written into the app-building standard as a numbered rule.
56	EXPAND (offer surface)	Directed that the broader monorepo’s capabilities be evaluated for productization, bounded by time and dedicated effort.	Two budget-bounded inventory efforts were launched, producing a scored shortlist for the client to pick from.	Company-OS console — screens, capacity view, approvals inbox (held)A swarm integrator died mid-build on a usage limit. Its work was recovered from disk without loss.An internal capability count was published wrong — stated higher than the real count at the time. It was corrected on the record, not quietly edited away.
57	OPERATING MODEL	Required continuous tracking against context loss, with surgical merges and ongoing documentation.	A single resume-first tracking file was created and kept current before every new workstream began.	A security regression was introduced by an earlier fix, caught by a later background review pass, and fixed forward within the hour — the fix was proven correct by reverting it and watching the regression reappear.
58	PICK	Picked between two synthesized package sets for a new offer extension.	Two package sets were adopted; this ledger itself became a candidate product artifact, pending a redaction pass.	Story-first site rewrite — seven pages, one flaky test self-healed at its root cause

Three ways in

However you want to run it.

Parts — install it yourself

The taxonomy starter set, the delivery-register format, the frontmatter provenance schema, and the continuation convention — as files, for your own vault or repo. Your team keeps the ledger from day one.

Operated — we run it and hand you the receipt

We run the same loop this page demonstrates on your engagement: you direct, agents execute, machines and agents verify, you decide at the boundary, delivery ships, the record updates. You receive the same two artifacts we produced for ourselves — a live touchpoint ledger and a delivery register, both append-only, both yours to check.

Install — your team keeps it

We map your own decision-boundary classes, install the append discipline into your existing tooling, and set a review cadence for someone on your team to audit the ledger against reality. Your operators run it going forward.

What this does not do

The honest limits.

It does not prove the underlying decisions were correct — only that they were made by an identifiable human, at an identifiable point, with an identifiable effect.
It is not a compliance certification or an audit opinion, and it is not a guarantee that nothing was missed — a boundary nobody thought to escalate leaves no row.
It does not prove the taxonomy generalizes cleanly to every buyer's operation — the class names here grew out of one specific engagement, not a universal schema.
It does not, by itself, prevent a bad decision — it makes a decision traceable after the fact, which is a narrower and more honest promise.
No named buyer has adopted this offering yet. What exists is the record of it running on our own build, under real pressure — not a documented market response.

How it starts

One door, whichever mode you want

The first step is the same for every mode, and it is not a call. Send a company URL and one sentence about what is stuck, and what comes back is a reply from a person.

A person reads every one of these, and replies within three business days.

We store what you send here to reply to it. We do not add you to a list, because we do not have one. The full detail is on the privacy page.

Questions

Three fair questions

What does it cost?: Priced per engagement and scoped once we understand the work. There are no bands on this site, because the number depends on what the work turns out to be.
Whose ledger is this, and is it really redacted?: Ours — our own reinvention of this site. Every removal in the specimen behind the timeline is marked, nothing is silently smoothed away.
Does it stop a bad decision from happening?: No. It makes a decision traceable after the fact — a narrower and more honest promise than "governance," and the one this page states plainly rather than oversells.