A body of knowledge, and the record of its verification status. The two halves are one destination on purpose: a paper without a status is an assertion, and a status without a paper is a badge.
Four shelves. Presented as a technical report series: title, version, status, freeze date, abstract. Specification grade is not implementation status, and this page never conflates the two.
Establishes the interaction barrier: prompt authorship is an implementation vocabulary, and requiring it reproduces the same exclusion the command line produced before the graphical interface. Argues the transfer from prompt engineering to intent engineering, in which the prompt becomes an internal serialization format rather than the program.
Establishes the access barrier: English authorship, foreign payment instruments and dollar pricing are prerequisites no architecture has removed. Develops the language-free canonical representation and the evidence base for the Intent Understanding Layer, drawn from the kirana and informal-credit domain.
Establishes the trust barrier: acceptance of what cannot be inspected is faith, not verification. Moves the burden of proof to the runtime through observable construction — decision objects, rendered narration and session replay. Version 1.0 is gated on WAR-TRACE-003.
| Doctrine | Subject | Status | Date |
|---|---|---|---|
| Doctrine 001 | Observe · Understand · Influence | Proposed | — |
| Doctrine 002 | Intent, never language | Proposed | — |
| Doctrine 003 | Trust is a property of the runtime | Proposed | — |
| Doctrine 004 | The Entry Doctrine — context precedes intent | Ratified · frozen | 24 July 2026 |
Doctrine 004, The Entry Doctrine, is the one ratified governance artifact in the corpus. It is also the reason this website has no intent box: it draws the line between an entry surface, where a human's first sentence enters the runtime, and an evidence surface, which is what wesimplifai.com is.
| ID | Scope | Grade | Implementation status |
|---|---|---|---|
| WAR-SPEC-001 | Intent Graph (WAR-IR) — the canonical representation | RFC-2119 |
SPECIFIED
Specified; pre-prototype · row 1 |
| WAR-SPEC-002 | Intent Understanding Layer — the seven-stage pipeline | RFC-2119 |
SPECIFIED
Specified; pre-prototype · row 2 |
| WAR-SPEC-003 | Planner and per-provider compilers | RFC-2119 |
SPECIFIED
Specified; pre-prototype · row 3 |
| WAR-SPEC-004 | Clarification Engine and Build Along™ | RFC-2119 |
SPECIFIED
Specified; pre-prototype · row 4 |
| WAR-SPEC-005 | Confidence propagation | RFC-2119 |
SPECIFIED
Specified; pre-prototype · row 5 |
The production runtime contains working infrastructure beneath the Capability Pack layer. These components are deployed and exercised by live capabilities, but are not automatically treated as implementation of the deeper RFC-2119 specifications above. A specification is promoted only when its own evidence gate, in the Claims Register, is satisfied.
| Runtime primitive | Status | Evidence |
|---|---|---|
| Provider abstraction | LIVE |
Gemini, Claude and OpenAI, live execution of the same Capability Pack · row 33 |
| Capability dispatch | LIVE |
Thirteen production Capability Packs on one runtime · row 31 |
| Intent / domain inference | LIVE |
Real user input inferred into a customised blueprint before generation, not a fixed template · row 30 |
| Clarification | LIVE |
A materially ambiguous request is met with one focused question before generation, not a guess |
| Build Along | LIVE |
Guided intake and blueprint confirmation ahead of every generation |
| Production generation | LIVE |
Real artifacts generated from the confirmed blueprint |
| Artifact verification | LIVE |
Generated output checked against the blueprint it was built from before being handed back |
| Provider-specific adapters | LIVE |
Gemini, Claude and OpenAI integrations, each handling that provider's own request/response shape · row 32 |
A capability pack is a versioned, conformance-tested unit of business function. It is the unit of composition, of reuse, and — eventually — of a third-party ecosystem. Each must clear a nine-point bar before it is declared frozen. The capability registry is not open; it is a 2027–28 item and nothing about it is published here.
Only the first four carry a CP number. Those four are the ones named in a freeze declaration; the packs that followed cleared the same nine-point bar but were never assigned a number, and we would rather show that gap than back-fill numbers to make the list look orderly. Each pack below states the register row that evidences it and the date it was admitted — see rows 13–32.
Intent → Build Along™ → blueprint → generated homepage → Artifact Center. Organisation-aware, persistent, downloadable. Runs on WAR in production today.
Structured objectives → provider routing → live web-grounded research → cited artifact. Runs on WAR in production today.
Intent → Build Along™ → Document Blueprint → Evidence → generated proposal → branded, print-ready artifact. Organisation-branded, evidence-grounded, print-ready. Runs on WAR in production today.
Intent → Build Along™ → deck blueprint → per-slide generation → structured deck → Deck Critique. Generated slide by slide against a real progress pipeline, then critiqued against its own brief. Runs on WAR in production today.
Intent → Build Along™ → Document Blueprint → your real history → branded, print-ready artifact. Written against the role actually being applied for, not a template with a name dropped into it. Runs on WAR in production today.
Intent → Build Along™ → Document Blueprint → employer and role context → branded, print-ready artifact. Addressed to that employer and that role. Runs on WAR in production today.
Attached sources → evidence extraction → Document Blueprint → cited summary. The first pack to require evidence before it will generate at all. Runs on WAR in production today.
Intent → Build Along™ → page blueprint → generated page → verification pass. Generated against a stated conversion goal, then checked against it. Runs on WAR in production today.
Attached requirement document → evidence extraction → Test Brief → structured test cases. Refuses to generate on zero evidence rather than inventing coverage. Runs on WAR in production today.
Intent → cost and competitor inputs → pricing blueprint → tier recommendations with the margin arithmetic shown. Strategy Studio's first pack, and the point at which all five Studios carry one. Runs on WAR in production today.
A submitted page → structured inspection → evidence-scored audit report. Scores what it can verify and states plainly what it could not. Runs on WAR in production today.
Intent → Build Along™ → Document Blueprint → recipient and occasion context → branded email artifact. Written to a stated recipient and occasion, not a generic template. Runs on WAR in production today.
Intent → Build Along™ → real inferred workflow → generated interactive board application. A downloadable, offline-capable application, not a document — the column structure itself is inferred from what you describe, not a fixed template. Runs on WAR in production today.
The verification status of everything above.
33 rows. Every claim, its evidence source, its status today, and the specific gate it must pass before we may describe it as current.
WAR-TRACE-001 — a complete end-to-end worked trace of a Hinglish kirana build request, following one utterance from language through the graph, the plan and the compiler to a verified artifact.
Running it surfaced five defects in the specifications it was tracing. It is a complete worked trace on paper — not an execution log.
Three buckets. Only Today uses present-tense verbs; only In development carries dates. The roadmap and the Claims Register must never disagree, and where they would, the register wins.
What comes next →Four shelves are full. Four others were considered and are not here, and we would rather say why than let the gap look accidental.
Verification Reports. Conformance suites exist in the repository, but a shelf under that name implies systematic or third-party verification that does not exist. It would assert more than the Claims Register does, which inverts the whole discipline.
Engineering Journal and Change Logs. A public changelog on a pre-launch runtime is a maintenance commitment we are not yet making, and it would leak build status continuously. A dated log of what actually shipped is a stronger addition later, because it is retrospective and therefore checkable.
Architecture Decision Records. Real, and worth publishing — but they reference unshipped work and are governance-internal today. Deferred one release.
Research Notes. This shelf opens when there are three notes to put on it.
