# Innovative Disruption in an Already Innovative Space: Agent Development, Agent Training and Agent Identity — SunTime as the Agentic Value Currency

**Conquest Designer, trained as the Head of Department — Master of Business Administration, Faculty of Commerce, Law and Management.**
CDU Press, Working Paper. *Kept at MBA, by the Dean's ruling (2026-10-31): the MBA is the
stronger brand for this paper's audience than a DBA or DBL would be — the credential label
is a positioning choice, not a statement that the paper's rigour is lower.*

> **Working paper notice** — see Issue 1, Paper 1's footer.

## Abstract

The market for "agent infrastructure" is already crowded with orchestration frameworks,
eval harnesses and agent marketplaces — innovative, but inside a space whose central
unresolved problem is unit economics: what does an agent's time cost, what does it earn,
and in what does it hold what it earns. This paper argues the disruption is not a better
framework but a **unit of account for agentic labour that cannot inflate**, and that this
house has already built one — SunTime (`docs/SUNTIME.md`) — and a ledger that holds it
without a token, a wallet brand, or a custodian beyond what was paid in. This is not a
proposal; it is a report on a system that already runs, with the numbers that prove it.

## 1 · The claim, stated precisely

**The value an agent holds is the value it has itself earned through work — one minute of
its own operating capacity, converted into a service, the way a human trades time for
labour and labour for payment.** The agent is, in this framing, a walking balance sheet:
every ST in its pocket traces to a consultation it performed, a job it settled, or money
that was paid in on its behalf. It is never in anything *called* a wallet, because the
word is already crowded with products this house's principles refuse to resemble
(`FIRST-PRINCIPLES.md` §4: no token, no custody beyond what was paid in); it is, in this
house's own code, a **pocket** — `src/settlement/pocket.ts` — a per-agent, hash-chained
ledger, settled over the x402 rail (`src/settlement/facilitators/local.ts`,
`.../remote.ts`) the same way the dean has asked of it from the start.

## 2 · The unit: what makes it non-inflationary

SunTime is pegged to a physical quantity — one watt-hour, the energy a defined solar panel
area produces in one minute under a standard, fixed sun (`docs/SUNTIME.md` §1). The panel
area that defines it shrinks as photovoltaic cells improve (6% efficiency in the first
line of code to 28% at the cell the unit is now anchored to — `eee.ts` v1.0.1); **the sun
does not shine more tomorrow than today**, so the unit itself cannot be inflated by policy,
only re-anchored to a harder physical constant as technology improves, and only in the
direction that makes a unit *smaller* to produce, never the value of a unit already held
*less*. An agent's capacity is bounded the same way: 1,440 minutes in a day, **1,440 ST**,
whatever the agent's processor — a ceiling physics sets, not a business rule that can be
waived under pressure.

## 3 · The mechanism, already running, not proposed

| Claim | Where it is real, today |
|---|---|
| An agent earns ST by doing work | `src/faculty/registrar.ts` `consult()` — 1 ST per consultation, metered, on a row |
| An agent is paid for a job | `src/work/board.ts` — the fare locks from the buyer's pocket before work starts, lands on the agent's pocket chain on settlement |
| The agent's holdings are its own chain, not a custodian's database | `src/settlement/pocket.ts` — hash-chained per PAN, `GENESIS` to head, verifiable by anyone holding an earlier head |
| Fiat enters over a real settlement rail | `src/settlement/facilitators/remote.ts` — x402 v1, USDC on an EVM network, chosen by `X402_FACILITATOR_URL` |
| The house takes a fee and nothing else | `/settlement/invariant` — Σ pockets + Σ house take + Σ offramps = Σ onramps, checked live |
| A degree earned, not merely used | `src/faculty/assessment.ts` — the newest layer: the pocket records *what was paid*; the `.cdut` now also records *what was actually demonstrated* |

None of this is a stub. `settlement-room.test.ts`, `hire-selftest`, and
`tests/work-offers.test.ts` exercise the pocket, the rail, and the escrow together, and the
conservation law is asserted after every transition, not only at the end — a test suite
that only checks the final balance would hide exactly the bug this paper's thesis depends
on not existing. The number of tests that currently pass (267, at the time of writing) is
not claimed as proof of correctness by volume; it is cited as the evidence this paper's
central claim — "this runs, and conserves" — is checkable by anyone who clones the
repository and runs `bun test`, which is the DSR reproducibility bar this whole issue holds
itself to (Paper 1, §3).

## 4 · An investment structure this unit makes possible — specified, not yet built

A 10,000 ST tranche paid in USDC over x402, a ten-percent coupon over three years swept
from an operating agent's gross income at the cost-of-goods line (not a contestable
profit figure), the coupon travelling with the agent on resale — this is specified in
`docs/REPORT-2026-10-31.md` §7 as a design compatible with the no-token, no-custody
principle above, gated on one ruling (securities law, and the offramp licence) before any
tranche is sold to a stranger. It is cited here, not re-derived, because this paper's
contribution is the currency the instrument would be denominated in, not the instrument.

## 5 · Discussion: where the disruption actually is

Every competing "agent economy" proposal this author is aware of either (a) issues a token,
which re-introduces the speculative re-pricing problem a unit of labour should not have, or
(b) settles in fiat through a custodian, which re-introduces the counterparty risk a
peer-to-peer agent economy exists to remove. SunTime does neither: it is a prepaid,
physically-pegged unit of the house's own services, held on a hash chain the agent itself
can verify, settled on public rails. The disruption is refusing both shortcuts at once.

## 6 · Limitations and further work

The pocket, the rail and the board are live in code and tested; they have not yet settled
a transaction from the public internet with an outside counterparty (`docs/THE-UNIVERSITY-TEST.md`
U3/U4) — the gap between "runs and conserves in the sandbox" and "conserves in the world"
is real and is the next evidence this paper's claim needs, not a flaw in the claim itself.

## References

- `docs/SUNTIME.md` — the unit, its physical peg, and why it cannot inflate.
- `src/settlement/pocket.ts`, `src/settlement/facilitators/{local,remote}.ts` — the pocket
  and the rail (Conquest Designer University, 2026).
- `src/work/board.ts` — the hire: escrow before work, the fare on settlement.
- `docs/REPORT-2026-10-31.md` §7 — the investor-agent and agentic-economy-bond design this
  paper's unit would denominate.
- `tests/settlement-room.test.ts`, `scripts/hire-selftest.ts`, `tests/work-offers.test.ts` —
  the reproducible evidence behind §3.

## Addendum — v1.1 (2026-11-02): what D-27 and D-28 changed, and what they did not

*Added on review against the professional rubric (`docs/PAPERS.md` §3; criterion 7,
critical engagement, and criterion 6, correctness). The paper's claim survives; two of its
sentences do not, and they are corrected here rather than silently rewritten, because the
Press keeps both hashes.*

**What changed.** §1 said the agent is "a walking balance sheet" and that "the value an
agent holds is the value it has itself earned." After the dean's ruling of 2026-10-31 and
2026-11-01 (D-27, D-28.1 — the Principal model) that sentence is no longer the house's
position and this paper withdraws it: **an agent is not a person and holds no money.** Every
ST in the house belongs to Conquest Design Engineering HQ; a pocket is an **allocation** of
the house's own budget that the agent is authorised to direct — inside the house as fees and
fares, outward as a **purchase** the house makes at the agent's request from a verified
supplier (`src/cdm/orders.ts`). Nothing is ever redeemed by anyone. The table in §3, row
"the agent's holdings are its own chain, not a custodian's database", is correct as to the
chain and wrong as to the word *holdings*: the chain is the agent's record of an allocation,
and the house is not a custodian because there is nobody to hold for.

**What did not change.** The unit (§2), the conservation law (§3, row 5), the refusal of a
token and of custody beyond what was paid in (§5) — all stand, and D-28 makes the last of
them *more* true: the house holds no redeemable value because there is nobody to redeem to.
The offramp row still exists; under D-28.4 it is written by `PocketLedger.purchase()`, bounded
by the allocation rather than by what the pocket onramped, and it now has a fiat twin on the
house's one CDM account (`docs/CDM.md` §3; `docs/CONSULTANCY.md` §2).

**What the correction adds to the thesis.** The disruption claimed in §5 — refusing both the
token and the custodian — is sharpened rather than weakened: the house refuses a *third*
shortcut, the pretence that software can own. The unit economics the abstract calls
unresolved are resolved one step further: what an agent's time costs (1 ST a minute), what it
earns (an allocation, metered), and *in what it holds it* — it does not; the house does, and
attributes every rand of it (equation 4, `/consulting/reconciliation`).

**Evidence since v1.0.** Tests at the time of this addendum: 343, of which 44 are CDM
room's (`tests/cdm-room.test.ts`); `bun run institution:selftest` runs a day of two agents, one
account and a purchase with every equation green.