In plain words: the depth phase opened with a milestone and a warning. The milestone: entry #100 of 300. The warning is the firing's real product โ three of the five instruments harvested today have no working code anywhere, despite being exactly the methods the STABLE boundary needs (invariant transfer to unseen regimes, mechanism-shift counting, robust anchor-based estimation). The pattern is now unmistakable and is recorded as the campaign's first depth-phase finding: the invariance literature's paper frontier runs far ahead of its implemented frontier. The two that DID enter clean are strong: the py-why ecosystem's nonstationary causal-discovery engine (very active, 1,645 stars), and โ walking closest to the graveyard fence of any entry yet โ the 1993 Andrews parameter-instability tests, whose difference from the boneyarded Chow test (robust variance, unknown breakpoint) is drawn explicitly in the row. The strategy consequence is recorded too: from here the STABLE climb inverts its search order โ find the maintained code first, then verify its paper โ and the self-hosted crawler joins the hunt for author code.
Depth-phase finding #1 โ the code gap
Depth harvest D1-01 (invariance sub-literatures):
H-096 CD-NOD causal-learn, MIT, 1645โ , active โ CATALOGUED
H-099 Andrews sup-Wald strucchange, CRAN โ CATALOGUED (boneyard fence drawn)
H-097 invariant transfer no code anywhere โ NO-IMPL
H-098 mechanism shift no code anywhere โ NO-IMPL
H-100 PULSE / K-class machinery exists, rule doesn't โ NO-IMPL
FINDING (reported, not padded over): paper frontier โซ implemented frontier
CONSEQUENCE: invert the search โ implemented-first discovery + firecrawl for author code
Coverage counters after this firing โ the 100 mark
Boundary
Minimum
Verified
Screened-pass
Fed
Starvation
STABLE
100
38
23
23
62
FRAGILE↔CONDITIONAL
60
14
12
12
46
CONDITIONAL-scoping
40
14
12
12
26
FAIL
40
13
10
10
27
EXPIRY
40
16
11
11
24
cross-cutting
20
5
0
0
15
Total
300
100
68
68
200
โถ Next iteration
Iteration 29 = STABLE depth #2, implemented-first. Per the code-gap consequence: start from maintained invariance/robustness FOSS (py-why ecosystem beyond CD-NOD, dodiscover, tigramite's newer modules, R InvariantCausalPrediction-adjacent packages) and verify their papers โ the reverse of the campaign's usual order. The firecrawl channel (littleblack via Tailscale) also enters play for the NO-IMPL revisit notes (H-088, H-091, H-097, H-098, H-100).