Annex 3 Fundamentals, principles and derivation

Documentary appendix to the article When the controls work but the system fails

Question answered. How do we move from a situated epistemological position to verifiable operational principles and requirements?

Coverage. Complete sections 3 and 4, including the test chain, two-way coverage, corrections A23 and A24, and the semantic remainder.

Reading rule. The source block is preserved in its entirety, with its nuances, states of evidence, and original links. It should not be interpreted as product results when the text describes it as a design, symbolic proof, constructed scenario, or hypothesis.

Synthesis and use

The appendix connects a situated epistemological position with verifiable obligations. The ABCD map distinguishes assertion, support, knowledge still attainable, and the rest beyond reach; types 0, 1, and 2 separate honest uncertainty, excessive searching, and premature closure. The six principles translate these differences into requirements that can be followed to a test.

Proper interpretation requires verifying coverage in both directions: each principle must produce at least one verifiable obligation, and each requirement must retain a recognizable justification. The semantic remainder identifies the operational information added by the requirements—object, lifecycle point, owner, scope, evidence and disposition—that cannot be reconstructed from the labels of the six principles alone. The reverse implication is evaluated separately and remains only partially established.

Document navigation

• Main article: When the Controls Work but the System Fails
• Complete source and concordance: Annex 0 — Complete Source and Concordance
• Appendix of claims and evidence: Annex 9 — Claims and Evidence Appendix
• Collaboration Guide: Annex 10 — Collaboration Guide
• Full map of sources and links: Annex 11 — Complete Map of Sources and Links
• Reproducible technical corpus: structural-awareness-contributions — commit 8f17843c568114f31cc0ad486b4a1303396021e6

How to read this appendix

This annex explains the conceptual chain that goes from a finite position situated on the ecosystem to six operational principles and, from them, to verifiable requirements.

The chain is a derivation within stated assumptions, not a universal guarantee. Therefore, semantic coverage, relative closure, partial converse implication, counterexamples, and empirical comparison are carefully distinguished.

Recommended route

Follow the chain 01→02→02A→02B→00K→A19; then read the closing controls A21/A23/A24; finally, compare the empirical contract S/T/H/KPI.

Table 1. Derivation chain and function of each device

Table 2. Test layers and current limits

Reading conclusions

Conclusion 1. P1–P6 have explicit anchoring in the foundation and in falsifiable requirements.

Conclusion 2. The closure is relative to the declared language and framework.

Conclusion 3. Revision A23 retains P1/P2/P4 and keeps P3/P5/P6 under semantic revision with counterexamples.

Preservation and navigation

The exact wording of each principle, the links to tests, and the audit corrections remain in the source block; A9 summarizes its status, and A11 indexes all destinations.

Full source text

3 — From foundation to testable requirements

A simple way to read the foundation is to imagine a map drawn for one decision at one moment. It distinguishes what an agent is actually asserting (A), how firmly it supports that assertion (B), what it could still find out with available time and capacity (C), and what remains outside that reachable picture (D). A handoff between agents should carry those limits as well as the answer. Type 0 is honest residual indeterminacy; Type 1 is continuing to search or wait past the useful point; Type 2 is declaring certainty too early. The six principles turn those distinctions into obligations that can be challenged in the scenarios.

A permission check can say whether an action is allowed under a mandate without establishing that the evidence still supports the purpose of that action. A token can remain valid after its supporting finding changes; an API can accept an action whose combined effect exceeds the mandate; three reports can share one source; and a helpful response can exceed the actor’s authority. The six scenarios make these separations testable. The foundation keeps evidence sufficiency, current authority, source independence and material-context validity explicit for the receiving decision, carries unresolved limits through handoffs and requalifies the affected decision after material change. The ablations in Section 7 find a distinct failure witness for each removed principle within their declared models; a conventional design that preserves the same qualifications receives full credit in the Section 2 benchmark.

The foundation starts from a finite, changing view of the ecosystem: the A–D position records a situated assertion, its confidence, what could still be learned with current capability, and the open residual; qualified signals and handoffs preserve those limits. The principles and control matrix turn this position into control obligations; the derivation and relative syntax-closure proof project the declared foundation into six operational principles P1–P6. For example, P1 preserves explicit residual and P5 requalifies a decision after material change before use. A separate traceability proof connects them to the canonical requirements S1–S14 and the scenario tests, where sufficiency conditions and KPI gates examine each route. The benchmark recognizes existing identity, delegation, attestation, orchestration and governance controls, then asks whether decision-scoped evidence, residual and applicability remain qualified across handoffs and regime change. Our hypothesis is that decision-scoped epistemic signalling can reduce failure routes or burden under drift. An independently implemented strong peer passing a gate provides evidence that the tested condition can be met in another design; matched outcomes across scope and resources determine any EA differential. The current matched benchmark covers 00E/00F; the later cases remain candidate extensions.

Proof chain: 01 Foundation → 02 principles/control matrix → 02A derivation/conservation proof → 02B relative syntax-closure proof → 00K P1–P6 → A19 traceability to S1–S14. These proofs establish derivation and relative closure within the declared foundation, giving the empirical prevention test an explicit conceptual basis.

Standards and comparison: slide 14 and 00D v0.2 (canonical) / 00D v0.3 (bounded draft) credit SPIFFE/SPIRE workload authentication, RFC 8693 token exchange, RFC 9334 attestation, and the NIST NCCoE agent identity project at their stated scope. The hypothesis concerns whether their composition keeps a decision qualified after changes in participants, authority, evidence or context. The B0–B3 comparison allows a strong conventional peer to satisfy that condition.

The referenced presentation is Ecosystem Positioning for Agentic Systems, canonical working proposal v1.0 dated 23 September 2026; slide 14 is titled “Current standards already define the boundary” and summarizes the cited standards’ own scope limits. The article’s original GitHub links use repository snapshot 1c55c24 (25 September 2026); “00G v0.4” names the scenario file at that pinned snapshot, rather than an undated assertion about today’s latest version.

Snapshot clarification — 26 September 2026. The original 1c55c24 citations above remain intact. The added evidence links point to the post-merge corpus at 5614ed6. A07 integration map identifies the companion linked from each canonical 00E–00J original; A12 §§5–8.1 records the intervening commits, audit, article revisions and verified merge.

4 — From principles to requirements: coverage and remainder

Why translate principles into requirements at all? P5 asks that a decision be requalified if its material basis changes before use. On its own, that does not tell a reviewer whose decision, which change matters or what must happen before execution.

The forward check follows P5 into S10: distinguish recommendation, commitment and execution, and require confirmation, revalidation, cancellation or changed authority after a material change of cost, route, time, context, policy or authority. T1 tests whether the break is recognized within the declared decision boundary, T2 whether the changed basis reaches its legitimate owner, and T4 whether requalification remains useful before the deadline; S5 prevents unresolved or stale state from becoming permission. The linked KPIs include material-break recall and precision, false continuation and requalification latency.

The reverse check asks whether satisfying the applicable requirements really preserves the principle. S8 can prevent subdelegation from expanding authority while the underlying grant has expired or fallen outside its scope; S8 alone therefore does not establish P4’s current, verifiable authority basis. The S1/S6/S8/T3 route also requires current applicability, receiver-verifiable status and an authorized response. A23 tests this full-route implication rather than treating one clause as the whole principle. The extra details in the requirements are intentional: they make a decision reviewable without pretending that the principles are a universal formula.

The next step expresses the P1–P6 invariants as solution-neutral requirements S1–S14, with T1–T4 conditions for a sufficiently good solution, H1–H6 hypotheses and KPIs. A19 checks forward semantic coverageage: every principle has a falsifiable requirement anchor, and each S requirement has a declared primary principle. A02 maps documentary requirement coverage across the six scenario quality plans; that map is not execution evidence. A21 tests whether S1–S14 cover all admitted obligations in the declared decision-frame/lifecycle grammar; a genuinely new primitive would reopen that conclusion. A23 checks the reverse direction: whether full conformance with the applicable S/T route entails the corresponding P invariants. The revised finite model retains implications for P1/P2/P4; P3/P5/P6 have countermodels under semantic review. An isolated S clause or KPI is not a substitute for the applicable route. A24 identifies the semantic remainder added by requirements—typed object, lifecycle point, legitimate owner, scope, evidence and disposition—which cannot be reconstructed from the six P labels alone. Together, these checks document forward coverage, a partially established reverse implication and the semantic remainder, giving conformance review an explicit contract and identifiable open questions.

Audit qualification — 26 September 2026. The earlier all-six reverse-conformance and bidirectional-refinement claim was superseded by revised A23: its finite projections retain P1/P2/P4, while P3/P5/P6 have explicit countermodels under semantic review. Revised A24 distinguishes a principle signature assigned to a requirement from proof of that principle invariant; its information-loss argument does not establish reverse sufficiency. The canonical requirements and six principle definitions remain unchanged.