RATIONALE — v0.1

Version: v0.1 · Authored: 2026-08-05 · Supersedes: v0 · Status: current ·

Immutable hash: aaaed38dc16901be1b18b93f1757533fa631cbac34457ee13e610b64201b2082

Co-publication rule: this document travels with v0.1 at every outlet. At the portfolio, manifesto

and rationale are the same document, so atomicity holds by construction. At any outlet where they

are separable, lineage-integrity refuses to record the publication unless a reasoning_locator

exists alongside the manifesto locator.

1. The one-sentence reason v0.1 exists

v0 could not fail. v0.1 can.

Everything below is a consequence of that change.

2. What forced it — and it was not taste

A research pass across four windows (98 sources: engagement · adoption · 30-year survival ·

300-year endurance) produced two graves that describe v0's defect precisely:

• P3P — W3C Recommendation, 2002. A machine-readable privacy declaration. Sites shipped

technically valid policies containing nothing; browsers dropped support; its own co-architect

published the obituary in 2012.

• Do Not Track — the same contract from the user's side. The W3C Working Group Note records why,

on 17 January 2019: *"Since its last publication as a Candidate Recommendation, there has not been

sufficient deployment of these extensions (as defined) to justify further advancement, nor have

there been indications of planned support among user agents, third parties, and the ecosystem at

large."*

One cause of death, twice: the declaration was optional and ignoring it was free. v0's ten

principles were exactly that shape.

3. The mechanism v0.1 adds

Ten principles → nine machine checks (one check serves a pair). Each check names a failure

condition and returns a non-zero exit code, and each records what it was derived from — a published

result or a dead standard — so a team the check fails can argue with the evidence rather than with

our taste.

Three worth their derivation:

• no-exit-no-pass — portability is a completed round trip judged on preserved capability, not

an export button. Holdable because someone already meets it: Consumer Reports' Data Rights Protocol

runs authorized-agent requests in production across OneTrust, Transcend, Yorba and Permission Slip.

An existence proof turns an aspiration into a procurement requirement.

• atrophy-declared — every capability names the skill its use may erode and the counter-practice

that maintains it, and treats the user's unaided baseline as a measured artifact. Derived from

Parasuraman & Riley (1997) — misuse and disuse are opposite failures, so "add a human" is not

monotonically safer — plus the EU AI Act's Article 26(2) requirement, verbatim: *"Deployers shall

assign human oversight to natural persons who have the necessary competence, training and

authority, as well as the necessary support."* The workflow is removing the competence the statute

assumes.

• human-gate-required — a restatement of Keech v Sandford (1726). Lord King LC ordered a

trustee to disgorge profit from a lease he obtained only because the beneficiary could not. The

court did not ask whether harm occurred; it removed the possibility of conflict. Prophylactic,

because after-the-fact harm assessment is exactly what machine-speed action makes impossible.

4. The corrections v0.1 carries against its own first draft

Three, recorded as disputed cards because a corpus with no corrections was not audited.

1. The audience premise was wrong. Corrected framing: *the gain is real, the cost is real, and

almost nothing on the market gives you the first without the second.* The evidence for the market

was in the data all along — RAND found homework AI use rising 48% → 62% while the share saying it

harmed critical thinking rose 54% → 67%. The same population, more use and more concern. That is

demand, not opposition.

1. The accountability-vacuum framing was out of date. Liability runs upward to the deploying

human or organisation; a California statute forecloses the "the AI did it autonomously" defence;

EU AI Act Annex III became enforceable 2026-08-02. So a missing human gate is not a governance gap

— it is unpriced exposure. That is both more accurate and a better argument.

1. Four load-bearing figures were over-tiered, including a reported AI safety incident sourced to

a weekly news aggregator inside a framework whose whole pitch is provenance.

5. And then the independent citation review found six more

Run after publication, adversarially, against the artifact rather than the search summary.

Full scorecard in AUDIT.md. Two were quote-laundering — a sentence attributed to Bjork that appears

zero times in the source, and a W3C quotation truncated mid-sentence with no ellipsis. Both are

the exact anti-pattern the portfolio's own authoring rules name.

The reason this belongs in the rationale rather than only in the audit: it is the strongest evidence

for why v0.1's central move is right. A careful author, writing a document about provenance, committed

six provenance defects. Care did not catch them; a check did. That is the whole thesis, demonstrated

on its author.

6. What v0.1 deliberately did NOT do

• Did not soften the thesis. v0's surviving claims are strengthened, not hedged.
• Did not add dread to compensate. Anticipated obsolescence causes measurable harm before any

capability arrives (Anders, 1956). Fear is an effective growth channel and using it would commit

the harm the document documents.

• Did not delete the faith commitment. Moved to a declared, separable postscript so a reader can

accept the framework without accepting the faith — and so a reader who shares it sees it named

rather than smuggled.

• Did not proxy the unmeasurable. Four of v0's six closing questions are not machine-measurable.

v0.1 says so and gates the preconditions — exit, disclosure, challenge, human gate, non-engagement

scoring — never the flourishing.

• Did not claim coverage. Verified Sovereignty Coverage is published at 0/14.

7. What would force v0.2

Stated in advance so the next version is triggered by evidence rather than by restlessness:

• Any falsifier in §10 returning a result — in particular, evidence that heavy assistive-AI users

retain or improve unaided higher-order performance over 12+ months.

• A demonstrated full round trip of accumulated memory between two major assistants, which would make

no-exit-no-pass redundant. We would retire the gate and say so.

• A gate proven gameable in practice — a capability passing all nine while plainly violating the

principles. That is a defect in the compilation, not in the principle.

• Closure of either none-yet area (homogenisation, meaning/obsolescence) with a real check.
• An external dispute we cannot answer.

And the honest negative: more sources alone do not warrant a version. 98 became 98 better-verified

sources without changing a principle. Research is maintenance; a version is a change of mechanism.

8. Open decision this version does not settle

The arXiv outlet strategy — whether manifesto versions map to arXiv replacement versions of one

submission (arXiv preserves every prior version permanently, which is itself an audit trail) or to

separate submissions with independent identifiers. Recorded in data/lineage.toml as an empty

publication with status = "PREPARED-NOT-SUBMITTED" so the gap is visible in the ledger rather than

absent from it. See docs/EVOLUTION_PROTOCOL.md §4.

Paul Jialiang Wu · agentic-portfolio-lovat.vercel.app · This page is GENERATED from the hashed source file, never hand-edited, so the text is verbatim by construction. Machine-readable ledger.