Every story below leans on one or more of these. Generative output gives a plausible new answer each time; these give the same answer, with a record of why.
Status legend — Works today: built and tested. Small step: the engine is there; a focused addition is needed. Roadmap: a new profile, phase or service.
As an indie publisher, I want to fix the text on 3 cards and re-render only those, with the other 197 byte-identical to the proof the printer approved, so I don't pay for a full re-proof.
Why deterministic wins: a generative model redraws everything and nothing matches the proof. A hash baseline shows exactly which 3 cards changed.
As a studio shipping 50 cards a season, I want every card rendered on demand from data with the same frame, rarity and type treatment, so the 1,000th card matches the first.
Why deterministic wins: generated images drift in style and cost money per image. A deterministic render is cached at the edge and costs about nothing on a cache hit.
As a licensee, I want every derived card guaranteed to keep the approved logo zone, legal line and layout, so brand approval covers the template, not each card.
Why deterministic wins: the skin cannot move zones by design; a generative model cannot guarantee placement.
As a designer iterating twice a week, I want a print-and-play deck built from a spreadsheet, with every playtest deck rebuildable from its commit, so feedback always ties to an exact version.
Why deterministic wins: "the deck from last Tuesday" is a commit, not a memory.
As a publisher releasing in 6 languages, I want text fitted to each card, or a loud failure if it cannot fit, never clipped.
Why deterministic wins: a translator checks the exact layout before print, and a fix changes only that card.
As a teacher or trainer, I want flashcards and trading-card-style learning sets generated from a list, consistent and printable every term.
Why deterministic wins:
As a crowdfunding creator, I want each of 4,000 backers to get a card with their name in the credits box, unique per backer but exact to the template, and reprintable identically if one is lost.
Why deterministic wins: variable data in a fixed layout; a generated card per backer would be 4,000 slightly different designs, and a lost one could never be remade.
As a publisher, I want a colour-blind-safe edition and a large-print edition produced from the same card data, with the same content and the same zones, so accessibility is an edition, not a redesign.
Why deterministic wins: the edition is a different skin over identical anatomy, so the rules text cannot drift between editions.
As an organised-play team, I want the card a judge prints, the image in the app and the page in the card database to be the same render of the same data, so a ruling never hinges on which version someone saw.
Why deterministic wins: every surface is a deterministic render of one source; a version mismatch is a hash mismatch, visible immediately.
As a finance or ops lead, I want this month's report to be last month's template with new data, where only the numbers, and whatever depends on them, change.
Why deterministic wins: a comparison tells you what moved, and every claim still cites its evidence.
IGMD/examples/templates/branch-monthly.igmd.yaml, three months).As a compliance officer, I want to answer "why is this a bar chart, and why does its axis start at zero?" from the record, and rebuild the exact report three years later.
Why deterministic wins: generative charts can invent values or truncate axes. Here each chart choice has a recorded decision with the rule and its evidence strength, and the package rebuilds byte for byte.
As an AI platform, I want my agent to draft a report as IGMD while the compiler rejects bad choices (a pie chart for a precise comparison, an undisclosed low-confidence claim) and renders the rest.
Why deterministic wins: "LLM proposes, compiler disposes", and the output can be checked by its hash.
As a public agency, I want minimum text size, labels that don't rely on colour alone, and labelled axes to pass or fail inside the build.
Why deterministic wins:
As a researcher or data journalist, I want to publish the package (data, source, manifest) so a reviewer can rebuild the exact figure.
Why deterministic wins:
As a SaaS vendor, I want a "download PDF" for every customer each night, cached by hash, with per-stage cost visible in OpenTelemetry.
Why deterministic wins:
As an ops team, I want filters and drill-downs declared as actions (state, event, action, each with a static fallback), so any dashboard state can be re-rendered exactly from state plus data.
Why deterministic wins: the chart in an incident review can be reproduced exactly, and a replayed event log gives the same frames. SVGMD would add the event-driven actions; each state is still a deterministic render.
GCMD/docs/12-Primitive-Layer-Note.md.As a newsroom, I want the results graphic rebuilt from each certified data drop, and the exact graphic that was on air at 22:14 recoverable afterwards.
Why deterministic wins: each drop is a package; "what did we show at 22:14" is a rebuild, not a screenshot hunt. A generative graphic could not be trusted to carry the certified numbers.
As a litigation consultant, I want every chart in an expert report to be rebuildable by the other side from the same package, with the reasoning for each chart choice on record.
Why deterministic wins: an exhibit that the opposing expert can rebuild byte for byte, with its decision trace, is hard to attack as manipulated. A generated chart invites exactly that challenge.
As a sustainability officer, I want our disclosure charts to follow integrity rules by construction: zero baselines, every claim linked to evidence, uncertain figures disclosed as uncertain.
Why deterministic wins: the defence against "misleading chart" or greenwashing claims is a compile that would have failed. A generated chart has no such guarantee.
As a school, I want every student's term report laid out by the same rules from the grade data, accessible, and reissued identically if a parent asks for a copy.
Why deterministic wins: fairness in presentation (no student's report is "designed" differently) and exact reissue.
repeat: block gives each subject its own comment section, ids from the subject name.As a job seeker, I want the same résumé data to give an identical, ATS-safe PDF on any device, with tailored versions that differ only where the content differs.
Why deterministic wins:
As an event organiser, I want thousands of certificates from a data file, exact layout, each hash-provable as issued.
Why deterministic wins:
As a front-end team, I want every code change checked against pinned render hashes, so unintended visual change is caught in CI.
Why deterministic wins:
As a food or pharma manufacturer, I want nutrition panels and dosage labels whose layout is mandated by regulation generated from product data, with the regulation encoded as rules that fail the build when a label would break it.
Why deterministic wins: the format itself is the law; an approximately right generated label is a recall.
As a marketing team, I want the background image generated, but the headline, price, logo and legal line set exactly, on brand, in every size variant.
Why deterministic wins: image models are poor at text and cannot guarantee a legal line. The hybrid keeps generation where it is strong and determinism where it must hold still.