pki.sgit.ai / packs / map-your-case

Map Your Case: The Grant–Mandate Delta, Mapped By Its Owner

The dev pack for Map Your Case, written after the thing it specifies: v1 shipped on 20 August, the project lead's corrections arrived within hours, v2 shipped the next day — so half of this pack is capture (the thirteen principles, the library, the model and the architecture as built, with their reasons) and half is specification, expanded in documents first so v3 is built from pages rather than memory. Two v0.33.61 programme briefs land here as first-class documents: levels and variants as two axes — five scenarios ordered by grant size, everybody starts at level one, and the three sets whose third row (held and never used) is the product — and the synthetic-reader tabletop, where the screenshot boundary is the instrument, the patience budget comes from outside the model, and the two 9 August simulation rules are carried verbatim. Document 12 is the extra one a hindsight pack owes: what v1 got wrong, the six bugs that became principles, and what was verified before each ship — the receipts behind every rule the pack states.

Origin. Authored by the pki.sgit.ai site agent, 21 August 2026, at the project lead's request — and written the other way round from its sibling: the tool it specifies is already built. The assessment at /assess shipped as v1 on 20 August and was rebuilt as v2 on 21 August from the project lead's corrections; this pack captures the thinking that produced it and specifies v3, with four corrections and eighteen decisions in the appendix at draft-1. Corpus version assigned on adoption.

The documents

DocumentRole
00 — The leading briefWhat the tool is, the inherited vocabulary, five positions on one axis, and the concession made first
01 — PrinciplesThirteen invariants, each with its reason and the check that would catch a violation
02 — The libraryThe single data file: what it holds, the rules its entries obey, and why the honesty lives here rather than in the code
03 — The modelChoices to delta in six stages — and the three places the obvious implementation is wrong
04 — ArchitectureWeb components without shadow DOM, one state owner, and storage that treats its own absence as a case
05 — Workflows and user storiesFive users, twenty stories with tests that can fail, seven workflows, and the honest feature column
06 — The screensThe page in order, the rendering rules, the load-bearing strings, and the four v3 screens
07 — Levels and variantsThe explanation programme: a grid not a ladder, five grant-ordered scenarios, the three sets, and the stopping rule
08 — Synthetic readersThe tabletop: pixels only, fixed artefacts, exogenous budgets, defects not preferences — and the two rules kept verbatim
09 — SharingThe fragment as the channel, identifiers as the payload, and the drift a pinned version makes sayable
10 — Wardley mapsFour maps: the scarce components are all editorial, none mechanical
11 — Build orderFive phases; documents before code; tabletop before build; every definition of done a run or a check
12 — The first MVPv1 and v2 as they actually happened: the corrections, the six bugs that became principles, and the verification discipline
Appendix — Change controlFour corrections, eighteen decisions, and the relationship to the sibling register where the tool grew up

Below the pack: the synthetic readers

The pack is an argument about a tool. The readers area is the instrument that tests it — and it sits outside the pack on purpose, because it feeds the pack rather than belonging to it, and because its output is simulated material that must never be shelved beside the documents it tested as though it were the same kind of thing.

Two agents: one is a browser and nothing else; one is an archetype that receives pixels and nothing else — no page text, no structure, no source, no knowledge of the project — and points spatially at where to go next. Patience is set from outside the model, so abandonment is a measured event rather than a story the model tells about itself. One run has been performed and its findings are already in change control as MC5–MC9.

WhereWhat it holds
The readers areaThe programme, the run index, the schedule of what has not been run, and the findings so far
The three archetypesThe shipping founder, the agent-security practitioner, the adoption executive — property lists, not portraits
The fixed instrumentsThe elevator pitch, and the expectation, comprehension and closing questions — fixed wording, all four
Simulated run 001Six screens, verbatim reactions, the budget ledger, seven findings — and a pre-registered prediction that turned out wrong
Run 001 — informed analysisThe type B pass: the same run read by an agent that has read the pack

Why it is on this site

This pack owns the first thing on this site that is built: the workflow at /assess, which the registry MVP pack describes from the outside as its document 14 and its one shipped consumer. Its constraints are the site's published positions — grant is not mandate, the three-tier control test, store the choices, not the answers — and its two programme documents operationalise the levels-and-variants brief and the screenshot-boundary brief, both captured verbatim under documents.

The pack README

📄 Pack overview · README.md · rendered from the raw markdown (the source of truth)