# Fluidity Replication: the review, written after the result

Written 2026-10-14, about 20:00 Halifax, after the report of 18:29 Halifax. Colson opened the review with "Okay, let's
review it." He asked for it to be kept with everything else: "Like everything?? Haha... since we first started comparing
the Spiral Staircase, until now, there's been tons of stuff."

Colson, after the review: "I don't want anything forgotten, especially because we have more than just these two passes
to be proud of, although it's just a toy model."

- Everything below was written after the results were known. The two verdicts come from the locked plan. Everything
  else is a description, and readings that go beyond what the plan asked are marked **after the fact**.
- The record is replication_results.txt, replication_results.json and replication_runlog.json, under the lock in
  replication_lock.json and Colson's go in replication_go.json. This note doesn't change or replace them. If anything
  here disagrees with them, they win.
- A toy model: structure inside the model, not evidence about cells.
- Histories, not memory: differences between routes are differences between intervention histories and the states they
  leave.
- The Cause Test FAIL stands as registered.

## What was asked

Colson's two hypotheses, word for word in plan.md:

1. **Coherence distribution.** A temporary coupling-medium intervention concentrates a larger share of its additional
   coherence among the slower, frequency-compatible oscillators than a coherence-matched direct phase push.
2. **Persistence.** The slower oscillator population's contribution to naturally arising coherence persists more
   strongly over the next 100 time units than the faster population's contribution.

PASS needed the gap to be at least 0.10, with the gap minus 3 error bars still above zero.

## The verdicts (locked in advance)

| | Fresh starts 115 to 134 | Pilot | Verdict |
|---|---|---|---|
| Share gap: the slower half's share of the extra, medium against push | 0.999 against 0.688; gap +0.311 ± 0.018 | +0.279 | PASS (lower bound +0.258) |
| Persistence gap: slower half against faster half, next 100 units | +0.318 against +0.074; gap +0.243 ± 0.038 | +0.324 | PASS (lower bound +0.130) |

In plain terms:

- When the medium shifts, practically all of the change in order (99.9%) happens in the slower half. A push to the same
  overall coherence spreads it wider, with about 69% in the slower half.
- In the system's natural rises, order held by the slower half is still there 100 units later, and the faster half's
  barely registers. The slower half's natural order stays clear for 300 to 400 units (+0.088 ± 0.019 over 300 to 400;
  the pilot's stayed clear for 200 to 300).

The run was clean: no failed checks, 198 of 200 moments like for like (the two directions not built are listed with
their reasons in the report), 18 foreground pieces, and the report computed twice with identical numbers.

## The pilot's growth readings, on fresh starts (secondary descriptions, never PASS or FAIL)

- **The medium's effect on later growth** was clear again: S = +0.00238 ± 0.00030 (7.9 bars; pilot +0.00303, 8.4 bars).
  Most of it came from raising the pull, as in the pilot (raised −0.00450, 5.8 bars; lowered +0.00025, not clear).
- **Medium against its matched push, like for like:** the S difference was +0.00117 ± 0.00032 (3.7 bars; pilot
  +0.00181, 3.6 bars). The P difference was the same sign but not clear this time (−0.00076 ± 0.00035, 2.2 bars; pilot
  3.1 bars).
- **The reading was the same as the pilot's:** "differs, same direction", with every secondary number the same sign as
  the pilot's.

So the gold's line, "same coherence, same settings, different future", described starts that played no part in finding
it. The wording kept with it in pilot_review.md still applies.

## Noticed on reading the report

- **The natural share now sits between the two routes.** On its own, the system puts 0.942 ± 0.016 of a natural rise in
  the slower half. The medium goes further (0.999) and the push falls short (0.688). In the pilot, the natural share and
  the medium's couldn't be told apart.
- **After a push, its extra order settles into the slower half (after the fact).** The plan's exploratory section prints
  how much of each half's part of the extra is left, compared with the moment of the push. Much of a push's extra order
  starts out in the faster half, and that part fades fast: averaged over the next 100 units, it's already slightly
  below the untouched runs (−0.142). Meanwhile the slower half's part grows to about 1.5 times its starting size
  (1.485 ± 0.039 over the next 100 units), and is still at its starting size 200 to 300 units later (0.998 ± 0.060).
  This reading is an after-the-fact description, not a test.
- **Early on, the medium's extra coherence outlasts the push's.** Over the first 100 units after the reset, 0.875 of
  the medium's extra is left, against 0.665 of the push's (a difference of +0.210 ± 0.024). From 200 to 500 units the
  difference isn't clear (+0.054, 2.0 bars), as in the pilot (1.6 bars).
- **Some numbers came in smaller than the pilot's,** among them the persistence gap and the growth effects. That's
  typical when a finding was first spotted after the fact. Every one stayed clear.
- **The pushes' own effect on later growth was clearer this time:** against the untouched runs, S = +0.00123 ± 0.00028
  (4.4 bars; pilot 2.6 bars, not clear). The medium's effect was still larger, and the gap between them repeated. This
  design differs from the Cause Test, so that test's FAIL stands as registered.

## What didn't work

The composition question asked whether the slower-half concentration explains why the medium's extra lasts longer. The
re-weighting meant to answer it was unstable (±63), so there is no usable answer. A part with almost no extra at the
moment of the push makes its ratio blow up. The question stays open.

## What a PASS means here

From plan.md, written before the run:

- "A PASS would show the pilot's fluidity observations are robust in this model and consistent with that theory. It
  wouldn't show a new mechanism."
- The theory it names is established work on synchronization. Near the threshold where a group starts to fall into
  step, the shared rhythm is carried mostly by the members whose speeds are close to the group's (Strogatz and Mirollo,
  1991). The faster members' contributions wash out as they drift apart (Strogatz, Mirollo and Matthews, 1992). That
  context was added to the plan for honesty and played no part in the readings.
- Fluidity is Colson's concept. He asked whether coherence has "a twin of polarity, incoherence as its inverse…
  differentiated from the medium", and named it "fluidity" (gathered record, entry 16). The replication tested
  predictions drawn from it.

## Still open

- Whether slowly built coherence is what carries the growth difference. How long coherence was held is the open factor.
- The Δ side of fluidity, the parts' spread of natural speeds. Test 1 turned only the pull K.
- The composition question.
- Test 2 (a medium that remembers) and Test 3 (spread by descent) stay separate and unapproved.
- Nothing new is planned. Any new plan, model or run needs Colson's approval.

## Where the numbers are

- replication_results.txt: "Primary verdicts", "Beside hypothesis 1", "Secondary", "Exploratory" and "Also reported".
- replication_results.json, under "exploratory": each half's share at the moment of the push and how much of it is left
  in each block, for both routes, including the faster half's numbers quoted above.