Less Than One Minute

This page computes the proof for both halves of the landing-page claim — that manual dependency tracing takes weeks (true = true) [1] and that parseltongue resolves the same checks in under a minute (true = true) [2]. Click any diff to open it in the side panel, then use Focus in Graph for the knowledge graph or Focus in Layers for the computation chain. Taints lists currently-known inconsistencies — mostly documentation that hasn't caught up with the code.

The Integrity Cortex card claims dependency tracing takes <1 min instead of weeks. That's two numerical claims about the same workload — a manual baseline and a parseltongue-resolved one. Both are grounded structurally, not benchmarked: the manual side from the size of the demo's checker, the parseltongue side from the per-edge cost of the taint BFS in the parseltongue codebase itself.

pltg

Load

(import (quote .src.sub_minute sm))
sm.manual-weeks
sm.parseltongue-minutes

Out

0.051000000000000004

The problem scale

The data governance demo models a biopharma data estate with 4 [3] technical catalogs, 25-30 [4] contract documents, and 60-80 [5] business catalog product files — 3 [6] independently authored layers with 15-20% [7] injected corruptions spanning all of them.

The checker discovers entities from raw source documents[8] and fans out compliance checks via axiom pattern matching[9].

The checker's compliance derives

Each derive in the checker represents a compliance check that would require manual cross-layer document comparison. We count 14 [10] compliance derives from the checker — per-department contract compliance[11][12][13][14], aggregated contract compliance[15], contract validity[16], phantom detection[17], product classification[18], omics compliance[19], and cross-layer fact integrity[20][21][22] — culminating in a final policy consistency check[23] with a closing diff[24].

On top of that, 3 [25] cross-layer consistency checks verify source existence[26], path consistency[27], and table consistency[28].

Manual effort: weeks

pltg

Manual side

sm.manual-cross-references
sm.manual-weeks

Out

3060,6.375

Total checks per entity: 17 [29] (compliance derives + cross-layer checks). Each check fans out across entities — conservatively 60-80 [5] products across 3 [6] layers gives 3,060 [30] manual cross-references. At 5 minutes per cross-layer document comparison, the manual effort is 255 [31] hours — 6.38 [32] work-weeks.

That matches the card's instead of weeks claim: true [33][1].

Parseltongue side: sub-minute, by construction

The same N cross-references resolve as edges in the parseltongue fact graph. Taint propagation — the system that surfaces every downstream dependency of a fact — is a textbook BFS over that graph[34]. Each edge is visited at most once; per-edge work is a list.pop + dict.get + set.add, all O(1) on Python's built-in containers[35].

pltg

Parseltongue side

sm.parseltongue-seconds
sm.parseltongue-minutes

Out

3.06,0.051000000000000004

The axiom bounds the rate at (= (src.sub_minute.parseltongue-seconds-per-reference ?cross-refs) (/ ?cross-refs 1000)) [36] — 1000 edges per second, ≈ 1 ms per edge. That's roughly three orders of magnitude slower than realistic CPython throughput on these primitives (each runs in ~50–200 ns on modern hardware), with generous headroom for interpreter overhead, GC pauses, and worst-case branches. So parseltongue-seconds = manual-cross-references / 1000 is a hard upper bound, not an estimate.

For the demo's 3,060 [30] cross-references that gives 3.06 [37] seconds — i.e. 0.05 [38] minutes. Sub-minute holds: true [39][2].

The same bound continues to hold for N in the millions of edges — a 10⁶-edge BFS at this rate completes in under 17 minutes (and in reality, about a second). The claim is structural: as long as the dependency surface is bounded by a graph traversal, the wall-clock cost is linear in edges with a tiny constant.

Two claims, both grounded

Both diffs close: the 6.38- [32]week manual reconciliation is the same workload as the sub-minute parseltongue traversal, differing only in representation.

Checks

References

Lineage