The governing rule
Compile assumptions. Never predict around missing evidence.
The engine traverses only stored-direction paths in a signed canonical release. It carries every original denominator, source-age policy and hypothetical assumption into the result. One-hop, interval-valued, unit-compatible paths can be transformed; every unsupported path remains visible and abstains.
- 3
- registered interval transforms
- 0
- probabilities or point forecasts
- 100%
- scenario assumptions ledgered
Signed deterministic test vector
Inspect one bounded scenario end to end
Loading the hash-bound scenario compilation…
- Release
- Compilation
- Result
From assumed shock subject to covered target
Stored direction only. A connection is not causal attribution.
- Edge interval
- Original denominator
- Freshness
- Universe coverage
Ranges remain ranges
No midpoint is promoted as the answer.
Every endpoint of the declared envelope
These are deterministic arithmetic cells, not likelihoods.
No hypothetical input becomes a fact
Abstentions cannot disappear
What the compiler is designed to answer
| Question | Permitted output | Refused shortcut |
|---|---|---|
| Which India exposure paths are structurally connected? | Exact signed paths, hops, evidence and coverage | A causal transmission claim |
| What follows under a declared shock range? | Registered interval arithmetic in the edge’s original denominator | A point forecast or probability |
| How do substitutions and buffers change the envelope? | Separate hypothetical assumptions and transformed bounds | An optimized recommendation |
| What cannot be calculated? | Typed gaps for unsupported units, multi-hop transforms, stale evidence and missing values | Model-imputed certainty |
| What was knowable? | The exact signed release and knowledge cutoff used | Hindsight from a later release |
The refusal boundary
- A point-only magnitude or duration presented where a bounded interval is required.
- An unsigned, rights-drifted, schema-invalid or temporally inconsistent release.
- A scenario whose knowledge cutoff differs from its signed release timestamp.
- Multiplying across multiple hops or unlike units without a registered transform.
- Hiding a stale edge, unused assumption, truncated search or missing path.
- Any output flag that implies forecasting, probability, causation, advice or a scalar score.
Machine contract
python -m src.shock_compiler RELEASE.json SCENARIO.json --output compilation.json
python -m src.shock_compiler_fixtureThe engine validates the release, scenario, policy registry, schemas, implementation, exposure traversal and source-rights bytes before compilation. Revalidation recompiles the complete output from those signed inputs and requires exact semantic equality.
What must happen before production
IGRM still needs rights-approved real exposure releases, declared India entity universes, independently audited edge intervals, registered multi-hop transforms where defensible, user-specific scenario isolation and deletion, external signing control, and blinded decision-utility tests. Until then this page proves executable architecture and refusal behavior—not real-world coverage, model accuracy, institutional readiness or adoption.
India Exposure DNA · Open evidence standard · Synthetic compilation JSON · Scenario schema · Output schema