Stage 1 built · Stage 2 in progress Demonstrator, not a product Meridian Archipelago — engineered fiction

ASSAY.

Assessment Semantics & Scenario Analysis

Intelligence preparation produces a mass of judged questions and answers. ASSAY's premise: those answers can become objective, quantifiable, typed objects — and be exploited by planning machinery honestly, without laundering judgement into false precision.

What ASSAY is

A shared spine, honest by construction

ASSAY explores optimisation techniques for command-and-control planning, anchored in the JIPOE/IPOE doctrine family (JP 2-01.3, ATP 2-01.3, AIntP-17). It is a deliberate re-derivation of a sibling project's principles against a new problem domain: intelligence-fed course-of-action analysis. Where a principle survives, it is validated as general; where it does not, that is a finding.

Knowledge typed · banded Compile stamped world Score banded verdicts Analyse sensitivity · staleness Surfaces four role views every value traces backward to named knowledge, with a named owner
demonstrated in progress planned

See it now

The published demonstrator

ASSAY has no compiled app — its shippable surfaces are static HTML rendered over real Meridian fixtures. These are live, running on the same code the build uses.

Fiction, deliberately. Every surface renders the Meridian Archipelago — an engineered scenario (ASSAY-DEC-8). No service, no plan, and no real operational picture sits behind these surfaces.

The objectives

Eight theses over one environment

ASSAY builds one spine and one small set of role surfaces, then explores a catalogue of theses across them rather than building each separately. A thesis is only ever explored — never claimed proven.

APipelineJIPOE knowledge can flow into an optimiser honestly — typed encoding + compile path.explored
BLeast-worstWhen commitments conflict, minimal relaxation yields a defensible plan-as-argument.planned
CRobustnessScoring plans across the adversary COA set beats optimising against most-likely.planned
DCollectionOptimisation can tell you where to look — COA-hypothesis discrimination.planned
ESensitivity / deceptionThe system can identify which beliefs the decision is leaning on.planned
FStalenessA changed answer identifies exactly what it invalidates.planned
GInterdependencySystems-perspective (PMESII) knowledge made computable — highest false-precision risk.horizon
HReactive redA responding adversary changes which plans survive.horizon

Where it is · plans becoming achievements

Progress the repository can back

The build runs in seven stages — lap first, depth after — behind a spine-complete gate. This tracker records only what the repository actually contains; a stage cannot show "building" before its research note is published.

Stage 0 Foundations — store, trace graph, type-gen, fixtures done
Stage 1 Knowledge & discipline — band pill, provenance chip, encoding firewall done
Stage 2 Compile — knowledge → stamped world channels in progress
·Stage 3 Score, then plan — banded verdicts, oracle cases planned
·Stage 4 Least-worst — relaxation (thesis B) planned
·Stage 5 Scenario robustness — (thesis C) planned
·Stage 6 Analysis loops — sensitivity, discrimination, staleness (D, E, F) planned
🚩  Spine-complete gate — not yet reached. Passing it swaps the Demo from wireframes to the live in-browser demonstrator, and opens Stage 7 (surfaces & narratives).

Reading

How the work is explained

Each stage gate publishes a research note and a dated update; where a milestone warrants it, that update grows into an article that walks through what the slice actually does — with a working embed of the new component or algorithm, never a screenshot (comms plan §6).

Published now

Research notes

The research-first record: each build stage opens with a bounded research note citing primary doctrine, ending in "what we will do differently." Note 00 (foundations) and note 01 (knowledge) are published.

Articles — forthcoming

Blog & embeds

The dated updates feed and its milestone articles — each carrying a live, playable embed of the slice it reports — publish as the stage gates land. Linked here the moment the first article ships; nothing is promised as written before it is.

The stance

Nothing shown as done that isn't

ASSAY's discipline is visible in its interface and in this page. The same rule governs both: no assessed quantity is rendered as fact, and no planned capability is rendered as shipped.

2–5 days · low
Banded honesty

No scalar from an assessed source ever reaches a surface unbanded. Bands are pure closed intervals — no stored midpoint, no distributional claim. Only directly-observed values and commander's thresholds may be scalar.

ASSAY-DEC-n
Register-first

Every decision is recorded in the register before it is cited anywhere else. Peer documents — and this page — are projections of it, kept current by batch propagation, never rival sources of truth.

why? →
Every number can be traced

Any verdict, score, or ranking opens a backward walk that terminates in named knowledge objects with named owners. A chain that dead-ends is a bug surfaced, not hidden.

Go deeper

The canonical set

ASSAY is docs-first: the reasoning is the asset. Start with the vignette and the walkthrough; the register holds every decision.