AtlatestRepositorysigil-doc

sigil-doc / tree / test / transcripts / sd-p7sabotage-predictions.txt

1SD-P7 sabotage: what each leg should do, written before the run.
2
3Recorded because a prediction that can be WRONG is the only kind worth making,
4and because SD-P1's most useful finding came from committing to per-case
5expectations in advance rather than reading the results and agreeing with them.
6
7leg layer planted defect expected outcome
8--- --------- ---------------------------------------------- ------------------------------------------
9 01 product folio child predicate widened, declaration kept PROFILE refused at construction:
10 "admits a witness the profile declares refused"
11 02 product widened AND declaration replaced by a text node profile CONSTRUCTS; arms REFUSAL and
12 LAUNDER stop passing; other arms still pass
13 03 product anti-narrowing witness set emptied arm NARROW stops passing
14 04 product vocabulary check emptied arm VACUOUS stops passing
15 05 product make-section-with-profile ignores its profile arm E2E stops passing
16 06 product author-normalize back to document-node? arm E2E stops passing
17 07 product author-flatten back to document-node? arm E2E stops passing
18 08 product ambient restore moved inside the thunk arm AMBIENT stops passing
19 09 product make-section admits anything arm DEFAULT stops passing
20 10 assertion an expected task id perturbed arm E2E stops passing
21 12 product default child predicate widened to document-node? PROFILE refused at construction:
22 "admits a witness the profile declares refused"
23 14 product floor on refusal emptied arm FLOOR stops passing
24 15 product admitted-witness check emptied arm VACUOUS stops passing
25 16 product core-block floor on refused neutered arm FLOOR stops passing
26 11 product code-block-language returns #f gate PASSES arm 01, FAILS arm 02
27 13 product a fourth author- name emitted by the parser gate FAILS arm 03
28 17 product document->plain returns "" gate PASSES arms 01-03, FAILS arm 04
30Two predictions I am least sure of, flagged so a surprise is informative:
31 * leg 09 - `make-section` admitting anything may also disturb arm E2E or the
32 construction differential, not only arm DEFAULT. If more than one arm moves
33 that is fine; if arm DEFAULT does NOT move, the arm is not pinning what I
34 think it pins.
35 * leg 11 - the whole point is that arm 01 must stay GREEN. If the phase suite
36 turns out to touch a code block somewhere I have not noticed, the leg proves
37 nothing about arm 02 and the driver will say INCONCLUSIVE.
39------------------------------------------------------------------------------
40AMENDMENTS AFTER REGISTRATION -- recorded so the provenance of each leg is
41readable, rather than leaving later additions looking pre-registered.
43 19 ADDED AFTER REGISTRATION. It did not exist when this table was written; it
44 was added when the leader ruled the describer in, to cover the new
45 default-unreached arm. A correct leg, but it was never predicted, and a
46 prediction table that silently grows is not a prediction table.
47 Planted defect: a describer reaches the default path.
48 Outcome: arm "a profile without a describer emits today's wording,
49 unlocated" stops passing. That is what it did.
51 18 REGISTERED, THEN DELETED. Fable's S-b, the depth shortcut. Its prediction
52 was "arm DEPTH stops passing while arm LAUNDER stays green". BOTH STAYED
53 GREEN, because there is no depth traversal to shorten: @section builds
54 inside-out and each section validates its own children at its own
55 construction. The leg was removed rather than reshaped, because a leg that
56 always passes is worse than no leg -- it reads as coverage. Leader ruling
57 (a); the depth arm is live via legs 2, 6 and 7, which leg 2 now asserts.
59 02 Its prediction was WRONG in its second half and is left uncorrected above
60 on purpose. Predicted "arms REFUSAL and LAUNDER stop passing; other arms
61 still pass". SIX of the arms went red, because five others use
62 folio-section-child? as their positive control. The leg is sound; the
63 lesson is that the arm COUNT overstates how independent the arms are.