HYPOTHESIS — THREE STATES, ALL DERIVED

There is no status control on this page. The state under the claim is computed from the criteria every time they change, and the rule that computed it is printed underneath so it can be checked rather than trusted. Falsifying criteria are not styled as a variant of the others — they sit in their own band above them, because one of them settles the page on its own.

⌘K

⌂

▤

▲

⌖

◻

HYPOTHESIS {{ ageChip }} {{ revChip }} + criterion + experiment history

THE CLAIM

▲{{ claimMeta }}

STATE · DERIVED recomputed when any criterion changes

all criteria met → supported · any falsifying criterion met → falsified · otherwise → inconclusive {{ overrideHint }}

THIS IS A RESULT

The reanalysis did not shrink the pooled effect. That answers the question this hypothesis came from — the small-study-bias explanation is not sufficient on its own — and it is worth as much as the outcome I was hoping for.

CAPTURE WHAT IT RAISES

? If it is not small-study bias, what is producing the between-lab dispersion? ⏎ capture

pre-linked to this result · will appear in the Inbox with the falsifying run attached

CRITERIA {{ criteriaMeta }}

FALSIFYING · {{ c.id }} if met, this decides the page alone {{ c.outcome }}

! {{ c.editedText }}

No experiment attached yet. Stated before any run — which is the point of writing it down first.

CONFIRMING · {{ c.id }} {{ c.outcome }}

! {{ c.editedText }}

No experiment attached yet.

DIAGNOSTIC · {{ c.id }} informative · does not decide {{ c.outcome }}

No experiment attached yet.

DESIGN NOTES

CAME FROM

Is the overnight retention benefit attributable to consolidation, or to encoding strength?

research question · promoted to hypothesis {{ promotedWhen }}

POSITION HISTORY

EXPERIMENTS

Runs live in the analysis repo. This page records what each one was taken to show, not how it was produced.

OVERRIDING INCONCLUSIVE — A WRITTEN ACT

The override is not adjacent to the state and is not a control on the page; it is reached from the small line under the derivation rule, and it cannot be completed without prose. What you write is not a field on the hypothesis — it becomes an entry in the position history, permanently, next to the derived state it contradicts.

OVERRIDE THE DERIVED STATE esc

INCONCLUSIVE derived → SUPPORTED asserted by you

WHAT YOU ARE OVERRULING

F1unresolved — no run has tested the preregistered-exclusion pooled effect

C2not met — the effect did not shrink monotonically with sample size

WHY YOU ARE OVERRIDING required · goes into the history and stays there

C2 is not met because the monotonicity test is the wrong test for this sample — I wrote it before realising the size distribution is bimodal. F1 is unresolved and will stay unresolved because the preregistered subset is too small to pool. I am calling this supported on C1 and C3 and recording that the criteria were poorly chosen

override and record cancel the page will read SUPPORTED · asserted, not SUPPORTED

AFTERWARDS, ON THE PAGE

STATE · ASSERTED 12 Sep

SUPPORTED overrides INCONCLUSIVE

“C2 is the wrong test for this sample — the size distribution is bimodal…”

read the full justification in the history →

The asserted state never loses its qualifier. The derived state stays printed beside it, so anyone reading the page later — including you — sees both what the criteria said and what you decided instead.

Editing the criteria to make the derivation come out right is the alternative the design deliberately makes worse: that path is the one that raises a permanent flag on the criterion, while this one only asks you to write a paragraph.

A CRITERION EDITED AFTER EVIDENCE ARRIVED

Criteria are not locked — a badly worded criterion should be fixable. But an edit made after a run is attached is the exact shape of post-hoc storytelling, so it is marked at the moment it happens, the mark never expires, and the previous wording stays readable on the page rather than only in the log.

AT THE MOMENT OF EDITING

Two experiments are already attached to this criterion.

Changing the wording now will be recorded as an edit-after-evidence and shown on the criterion permanently. If the criterion was simply wrong, that is a fine reason — write it below and it will read as one.

The monotonicity test was the wrong operationalisation — the sample-size distribution is bimodal, which I only saw after run 14

save the edit add a new criterion instead

The second button is the escape hatch that keeps the record honest: a new criterion beside the old one costs nothing and leaves both visible.

FOREVER AFTERWARDS

CONFIRMING · C2 met

The pooled effect is lower in the large-sample stratum than in the small-sample stratum.

!EDITED 6 SEP, AFTER 2 EXPERIMENTS

was — “The effect shrinks monotonically as study sample size increases.”

“wrong operationalisation — the size distribution is bimodal, which I only saw after run 14”

outcome flipped: not met → met · see history

The flag is the same weight as the criterion text and carries the old wording with it, so the page cannot be read as though the criterion had always said this.

POSITION HISTORY — CLAIM, CRITERIA, AND EVERY OVERRULE

One timeline for the whole argument: revisions of the claim, edits to criteria, override justifications, and the runs that moved things. Explained entries lead; silent ones stay visible as dated rules. Edits made after evidence and overrides of the derived state are never quiet, whatever else is.

Reanalysing the open nap datasets with a preregistered pipeline will shrink the pooled effect below d = 0.20.

newest first

Two entries here can never be quiet: the criterion edited after run 14, and the override of the derived state. Both are rendered as blocks with their own borders whether or not a reason was written, because the absence of a reason is the thing a reader most needs to see.

Everything else follows the Research Question's rule: recording your reasoning is optional and rewarded. A claim that narrowed three times with two explanations still reads as a train of thought.