Shared core for the veridical-simulation program: lexical Agda parser, generic octad importer, octad veridicality prober, and a cross-repo bridge probe. Originally extracted from `email-octad-experiment’s phase-1 plumbing; first ingested formal artefacts in Phase 2 (echo-types) and Phase 3 (absolute-zero CNOs).
The framework treats raw artefacts as the territory and any structured
representation as a map. Each artefact becomes one octad-entity in
verisimdb across eight shapes (Graph / Vector / Tensor / Semantic /
Document / Temporal / Provenance / Spatial). A prober scores each shape
in [0, 1]; the geometric mean is the map fidelity. An empty shape
is a finding, not a failure — the framework refuses to fake content
the territory does not have.
| Crate | What |
|---|---|
|
Lexical Agda parser. Extracts top-level definitions (data / record / function / postulate / nested module) plus their leading doc comments and lexical references. Conservative — unclassified lines are surfaced honestly rather than silently dropped. |
|
Two-pass importer: pass 1 creates one octad per definition with
Document / Semantic / Vector / Tensor / Provenance / Temporal shapes;
pass 2 wires |
|
Veridicality probes per shape, plus parser-coverage and reference- integrity meta-probes. Emits A2ML report + Nickel schema. Tensor is intentionally capped at 0.75 (per-definition statistics ≠ corpus-level proof-tree-shape Tensor — a standing finding); Spatial scores 1.0 iff empty (faithful to a non-spatial territory). |
|
Given a bridge |
Each phase ships a small per-target theorem bundle alongside the empirical
probe report. All proofs typecheck under %default total with no banned
patterns (believe_me, assert_total, Admitted, sorry, unsafeCoerce).
| Bundle | Theorems | Buys |
|---|---|---|
|
3 |
Importer-side invariants the Graph and pass-2 logic depend on: qualified-name determinism, suffix-match agreement, lexer reference-set monotonicity. |
|
3 |
Phase 1 fixes are identity functions on their respective fields:
temporal RFC 3339 round-trip, |
|
3 |
Conservation of definitions ( |
|
3 |
|
|
6 |
|
|
6 |
Per-capability |
|
3 |
Per-snapshot |
Verify any one with:
cd src/abi && idris2 --check <Bundle>.idr
Outstanding follow-up theorems (LIFO undo cascade for januskey, the
includes? ↔ ∈ expand biconditional for chimichanga, retention
idempotence for arbitrary lists, etc.) are queued for echidnabot
under echidnabot-priorities/<phase>-proofs.a2ml with claim
statement, strategy, and what each buys.
agda-octad-importer --source /var/mnt/eclipse/repos/echo-types/proofs/agda \
--territory echo-types \
--report docs/import-report-echo-types.json
agda-octad-prober --source /var/mnt/eclipse/repos/echo-types/proofs/agda \
--import-report docs/import-report-echo-types.json \
--out docs/octad-veridicality-echo-types.a2ml \
--schema nickel/veridicality-echo-types.ncl
Result: 391 definitions parsed across 55 files, 388 octads created. Per-shape scores all 1.0 except Tensor (0.75 cap, by design) and Spatial (1.0, scored as faithful-empty for a formal corpus). Overall geometric mean = 0.9647.
agda-octad-importer --source /var/mnt/eclipse/repos/verification-ecosystem/absolute-zero/proofs/agda \
--territory absolute-zero \
--report docs/import-report-absolute-zero.json
agda-octad-prober --source /var/mnt/eclipse/repos/verification-ecosystem/absolute-zero/proofs/agda \
--import-report docs/import-report-absolute-zero.json \
--out docs/octad-veridicality-absolute-zero.a2ml \
--schema nickel/veridicality-absolute-zero.ncl
Result: 42 definitions parsed from CNO.agda, 42 octads created.
Overall geometric mean = 0.9647.
cross-repo-bridge-probe \ --bridge /var/mnt/eclipse/repos/echo-types/proofs/agda/EchoCNOBridge.agda \ --report-a docs/import-report-echo-types.json \ --report-b docs/import-report-absolute-zero.json \ --out docs/cross-repo-echo-cno.a2ml
Finding: EchoCNOBridge.agda resolves 20 references into the
echo-types corpus, 0 into the absolute-zero corpus, 12 unresolved
(stdlib). Bridge fidelity 0.00. The file is a name-bridge — uses CNO
terminology in name — but defines its own local CNOOp data type
rather than importing absolute-zero’s IsCNO predicate. This is a
candidate informal-formal disagreement signal, not a failure.
MPL-2.0 (preferred); MPL-2.0 is the automatic legal
fallback until PMPL is formally recognised. Crate metadata declares
MPL-2.0 explicitly because crates.io requires an OSI-approved
licence string.