]> dreyeck.freedombox.rocks Git - topic-maps.git/blob - docs/slice-16b-validation.md
Qualify pinned CXML encodings with reproducible local patches
[topic-maps.git] / docs / slice-16b-validation.md
1 # Slice 16B — Reification validation
2
3 Base: `f24ac27e1e1cc0118411b058d40af3c1c87f3ffb`; local branch
4 `codex/ctm-slice-16b`. Source and artifact hashes, runtime and UTC interval are
5 recorded in `reports/slice-16b/execution.json`.
6
7 ## Reproduction
8
9 From this checkout's root:
10
11 ```sh
12 python3 scripts/validate-slice-16b.py --sbcl /nix/store/hlsiblxpxr1hax3dw9x8znd6zy097ddj-sbcl-2.5.4/bin/sbcl
13 ```
14
15 The complete retained suite executes in fresh SBCL 2.5.4 with user startup and
16 user ASDF registry excluded, under `/usr/bin/sandbox-exec` denying network access.
17 No additional SBCL versions, relocation or cache-management changes are involved.
18 The documented Lisp inspection forms run in a second isolated process.
19
20 ## Results and evidence
21
22 - 9,964 passing Lisp checks: 9,728 retained checks and 236 Reification checks.
23 - Retained Python unit tests, 16,294 independent NFC comparisons and all
24 independent report checkers pass.
25 - Eleven new original CTM/CXTM byte-exact comparisons and two original specific
26 CTM rejections. Exactly these thirteen imported contracts transition to PASS.
27 - Full corpus: **106 PASS, 2 FAIL, 0 UNSUPPORTED, 165 NOT-RUN**.
28 - The FAILs remain `occurrence-datetime2.ctm` and
29 `occurrence-datetime-timezone2.ctm`, with their existing normative discrepancies.
30
31 `reification.json` contains per-check labels, original sources, current model
32 properties, actual Reifier Topic ordinals and EQ/inverse assertions.
33 `reification-actual/` contains actual CXTM bytes; the original baselines remain
34 unmodified. `tests.log` records full execution and independent checks.
35 `describe.log` contains real objects from the documented inspection process;
36 its addresses are historical observations in that process, not durable identity.
37
38 The validator compares all prior 83 successful CTM/CXTM outputs and both known
39 discrepancy outputs, every retained model observation, all prior suite totals and
40 the seven manual XTM2-model comparisons with Slice 16A. Corpus and reference
41 checksums are checked before/after; the pinned upstream material is unchanged.
42
43 ## Model and failure coverage
44
45 All six attachments are checked with actual EQ references in both directions.
46 Tests cover null, idempotent, different-Reifier merging, explicit clearing,
47 all directed cross-kind Reifier conflicts, foreign/invalid arguments and Topic
48 as an invalid reified target. Each structural item kind is merged with separate
49 Reifiers and Item Identifiers; retained handles and original Map membership are
50 checked after further assertions. Role/Association metadata is retained.
51
52 A four-level Name → Occurrence → Association/Role cascade reaches a fixed point.
53 Scope-collapse fixtures reject at each of those four levels through three
54 Topic-merge entry routes and preserve the complete original graph/handles.
55 Additional tests cover setter-induced collapse, identifier-triggered Reifier
56 conflicts, Reifiers also serving as Parent/Type Topics, construction order,
57 serialization purity and explicitly rejecting corrupted Reifier references.
58
59 Tests formerly asserting unsupported Reification now assert successful reading.
60 The old Topic-SI/Occurrence-II restriction tests now assert acceptance without
61 automatic Reification. Their prior assertion counts are retained; these are
62 consumed feature boundaries, not removed coverage. XTM2 models remain manually
63 constructed comparisons; no XTM deserialization contracts are claimed.
64
65 Atomic semantic rejection is the local API policy, implemented by disposable
66 copies of the same TMDM structures using the same merge operations. It does not
67 claim a general transaction engine or recovery from resource exhaustion.
68 See `reading-reification.md` for exact normative clauses, API and exclusions.