Skip to content

context.md

Latest commit

 

History

History
62 lines (49 loc) · 2.8 KB

context.md

File metadata and controls

62 lines (49 loc) · 2.8 KB

Context

Objective

Create backpressure mechanisms that stop parity drift between tonic run and tonic compile from landing unnoticed.

Current state

Experiment 1 is complete and kept.

What changed

  • tests/differential_backends.rs now derives enforced differential coverage from the parity catalog instead of a hand-maintained subset.
  • Every active compileable parity fixture now has an explicit differential decision.
  • Native C equality emission was corrected to use tn_runtime_value_equal(...), fixing the newly exposed strict_equality.tn mismatch.

Primary metric

  • Eligible parity fixtures missing enforced differential coverage
  • Baseline: 80 uncovered active compileable fixtures
  • Current: 0 uncovered fixtures

Current coverage snapshot

  • Eligible active compileable catalog fixtures: 99
  • Enforced differential fixtures: 91
  • Explicit exclusions: 8
  • Silently uncovered eligible fixtures: 0

Evidence collected

  1. cargo test --test differential_backends active_compileable_catalog_entries_have_explicit_differential_coverage -- --nocapture ✅
  2. cargo test --test differential_backends -- --nocapture ✅ (4 passed in 26.85s)

Remaining explicit exclusions

These are known gaps, not silent drift:

  • Missing native tn_runtime_for support:
    • examples/parity/02-operators/stepped_range.tn
    • examples/parity/10-idiomatic/closures_and_captures.tn
    • examples/parity/10-idiomatic/fizzbuzz.tn
    • examples/parity/10-idiomatic/keyword_filtering.tn
    • examples/parity/10-idiomatic/list_processing.tn
    • examples/parity/10-idiomatic/pipeline_transform.tn
  • Multi-clause anonymous-function capture gap:
    • examples/parity/05-functions/function_capture_multi_clause_anon.tn
  • Native runtime diagnostic text drift:
    • examples/parity/06-control-flow/for_into_runtime_fail.tn

Decision boundary for the next strategist

Default route: task.complete.

If the loop is already sitting on a task.complete routing event and nothing about scope or evidence has changed, stop there rather than re-planning or re-verifying the same slice.

If advisory topology metadata still suggests strategist/implementer/benchmarker/evaluator after task.complete, ignore it unless the scope has actually changed.

Only plan another experiment if there is an explicit decision to burn down one of the 8 named exclusions. If that happens, pick exactly one exclusion family and define a fresh baseline metric for it.

Relevant files

  • .miniloop/autoresearch.md
  • .miniloop/progress.md
  • tests/differential_backends.rs
  • examples/parity/catalog.toml
  • src/c_backend/ops.rs
  • src/c_backend/stubs_closures.rs
  • src/c_backend/terminator.rs

Operator note

Previous loop docs here had stale tonic install planning context. They should not drive the next iteration.