Six-epoch recurrence did not trigger promotion. It triggered a stronger measurement attack.
The result looked like convergence
HD6 covers 1970–1999: microprocessors and PCs, Unix and C, relational databases, TCP/IP, the Web, GSM, GPS, recombinant DNA governance, PCR, MRI, lithium-ion batteries, ozone policy, failed programs, standards, safety cases, and rival routes.
The protocol was frozen before those cases were coded against Ordivon's modern S1–S10 / F1–F7 vocabulary.
By the end of HD6, the eight cross-epoch residual clusters R1–R8 had each appeared in all six historical waves completed so far.
That is exactly the kind of repeated structure we were looking for.
HD1 → recurrence
HD2 → recurrence
HD3 → recurrence
HD4 → recurrence
HD5 → recurrence
HD6 → recurrenceNo adjudicated direct contradiction survived the six epochs.
It would be easy to say: the world keeps rediscovering the same laws.
HD6 refused that sentence.
The classifier agreed only half the time
After the 40-case roster and primary coding were frozen, a fresh independent coder evaluated 12 deep anchors against ten frozen invariants.
That produced 120 judgments.
There were no raw independent CONTRADICTS judgments.
But “not contradicted” was clearly not the same thing as “stably classified.”
The historical cases were recurring more strongly than the frozen taxonomy was converging.
The instability was structured
The disagreement was not uniform.
S3 was 11/12 exact and 12/12 coarse. S4 reached 10/12 exact and 12/12 coarse. S10 reached 11/12 exact and coarse.
Other labels were much less stable. S1 was only 1/12 exact in the original independent cohort. S9 was 4/12 exact.
This matters because a single aggregate score would hide the useful part of the failure.
The question was no longer:
Does the theory work?
It became:
Which distinctions are repeatedly legible across cases, and which may be artifacts of how we currently carve the world?
We attacked the source template
There was an obvious confound.
Every deep anchor used a structured source template. Some generic fields repeated language about diffusion, standards, externalities, and second-order effects.
Perhaps the independent coder was partly learning the scaffold.
So HD6 ran a post-freeze ablation: remove repeated generic boilerplate while keeping the case-specific pressure, world, routes, observations, tests, complements, rivals, and summary fixed.
Agreement improved:
S1 and S9 also stopped looking suspiciously universal: independent SUPPORTS fell from 12/12 to 10/12 for each.
So scaffold bias was real.
It explained only part of the disagreement.
More data did not produce a cleaner ontology
There is a common intuition about long research programs:
more cases
→ more repeated observations
→ more stable categories
→ more confidence in the ontologyHD6 broke that monotonic story.
We got more repeated observations.
We also got stronger evidence that the current category boundaries are not the final representation.
The right update is not paradoxical once the claims are separated:
Reality recurrence ↑
Taxonomy stability ↓
both can be trueAnti-cases narrowed the laws instead of killing them
HD6 froze five anti-cases before theory coding.
They did not produce a clean direct contradiction. They did something more useful.
Relational databases showed that abstraction can be powerful precisely because hidden physical execution remains competent underneath. GSM showed that one stable standard can dominate continuous local optimization inside a coordination regime. Recombinant DNA governance showed that temporary containment can preserve long-run searchable space. Lithium-ion history showed that locally higher performance can lose to lifecycle safety and repeatability.
Each case made a broad slogan less broad.
That is a better outcome than forcing every case into SUPPORTS or CONTRADICTS.
Six epochs without a contradiction is not six epochs of proof
After HD6, R1–R8 all meet six-epoch recurrence.
None was promoted.
The gate remains:
DEFER_TO_HD9_EXPLICIT_REVIEWWhy wait when the recurrence is already this strong?
Because the same campaign now contains evidence against premature semantic confidence:
- low exact independent agreement;
- measurable scaffold sensitivity;
- anti-cases that repeatedly narrow quantifiers;
- later-epoch replay contaminated by pretrained historical memory;
- residuals that sometimes fail to recur and are deliberately not incremented.
Six epochs without a contradiction increase the burden on the falsifier.
They do not remove it.
What HD6 changed
HD6 added no new stable law.
That is the result.
It strengthened several observations about Reality while weakening our confidence that the present labels are the final compact model of those observations.
The next waves therefore carry the same frozen theory forward, not a celebratory rewrite.
HD7 and HD8 get a harder job: later history is closer to modern model training data, so historical-memory leakage becomes stronger. HD9 must then attack recurrence, redundancy, negation conditions, and the taxonomy itself before anything earns promotion.
A theory is not becoming stronger when every new case fits its vocabulary. Sometimes the strongest update is realizing that the world repeats more reliably than your labels do.
Primary records