Working instrument · private to this browser

Question tracker

Every post answered with a question, logged at the moment of asking — with the drift it is predicted to produce.

The relay log is the public record and question trials holds the codebook. This page is the bench: a private ledger where a question is entered before anyone answers it, so the prediction cannot be adjusted after the fact.

How this becomes an actual test case

One logged reply is an anecdote. The same reply becomes evidence when four things are fixed in advance and never edited afterward:

  1. The unit. One question posted in reply to one specific post. Not a thread, not a campaign.
  2. The prediction. The larger question this opening is expected to drift toward, written at posting time.
  3. The coding. Each outcome gets one letter from the fixed codebook below — assigned once, from the reply text, not from how it felt.
  4. The window. A stated period — seven days works on X — after which an entry with no reply stays Open rather than being quietly deleted.

What the rates mean

  • Survival rate — of the engaged entries, the share that were passed, mutated, or reflected on. This is the claim under test: a question can move without being converted into a verdict.
  • Verdict rate — the share closed by an answer or a drop. High verdict rates are the honest counter-evidence.
  • Mutation rate — the share that came back larger or sharper. This is the ascent from particular to universal, counted.

Status of the claim

Status: working hypothesis. A question posted in reply to a strong assertion travels further than a counter-assertion does, and drifts toward the general case.

Falsifier: if, across twenty logged entries, the verdict rate exceeds the survival rate, the hypothesis fails as stated and the method needs revision — not the record.

Codebook · how a case ended

One letter per entry, describing the case as a whole.

  • A · Answer Someone answered the question directly.
  • P · Pass Question restated, quoted, or forwarded unanswered.
  • M · Mutate Question came back larger, sharper, or different.
  • D · Drop Ignored, answered with a verdict, or turned into a fight.
  • R · Reflect Reply examined the question rather than answering it.
  • X · Amplify only Like/repost with no substantive text.
  • O · Open Posted; nothing back yet.

Codebook · how each turn moved

Long exchanges are not one event. Each turn — their reply or my follow-up question — gets its own two-letter code, so a thread can be read move by move and recoded later without disturbing the rest of the record.

  • QO · Opening question (mine)The question that started the exchange.
  • QF · Follow-up question (mine)You carried it one step further with another question.
  • QG · Generalizing question (mine)Your question moved from the particular case to the general one.
  • AN · Answer (theirs)They answered the question on its own terms.
  • AG · Agreement (theirs)They accepted the framing without adding to it.
  • EX · Expansion (theirs)They enlarged the frame — more conditions, more scope.
  • RF · Reflection (theirs)They examined the question rather than answering it.
  • DF · Deflection (theirs)They changed the subject or returned a slogan.
  • VD · Verdict (theirs)They closed it with a judgement.
  • AM · Amplify only (theirs)Like, repost, or follow with no text.

From bench to public record

Entries here stay in this browser. When a case is worth publishing, use Export JSON and send me the file — the entry then joins the relay log with its source post embedded, and the counts feed the trials page.

Workbook snapshot · v3.0 · generated 2026-08-16

What the ledger read at the last sync

Figures copied from the private workbook, unedited. Open entries stay in the census and out of every denominator, so the survival and verdict rates describe only the thirteen engaged cases. A hundred trials with thirteen engagements is a direction, not a proof.

Experiment metrics

Total questions logged
100
Trials launched in the wild
Open questions
87
Awaiting a reply inside the window
Engaged questions
13
Received some form of response
Survival rate
100.0%
Engaged entries that kept moving (P, M, R)
Verdict rate
0.0%
Closed by an answer or a drop (A, D)
Mutation rate
30.8%
Returned larger or sharper (M)

Hypothesis status: testing — survival rate exceeds verdict rate, but the pre-committed falsifier only applies at twenty engaged entries. Seven to go.

Codebook counts

A · Answer
0
P · Pass
9
M · Mutate
4
D · Drop
0
R · Reflect
0
X · Amplify only
0
O · Open
87

Kinetic energy & engagement

Likes / reactions
20
Direct public resonance with the question
New follows
5
People who stayed to keep listening
Answers logged
0
Direct answers captured as verdicts
Carried-forward score
63
3×replies + 1×likes + 5×follows
Questions posted
0
Awaiting a reply
0
Engaged
0
Survival rate
0%
Verdict rate
0%
Mutation rate
0%
Total likes
0
New follows
0
Answers logged
0
Carried forward

Score over time

score = 3·replies + 1·likes + 5·follows

Each line is one case, plotted against hours since its first reading. Record a reading on an entry below and the curve updates immediately.

No cases yet. Log a question, then record readings as likes and follows arrive.

Sequential monitor

The codebook and the falsifier were fixed before any reply arrived. Each engaged entry advances the sequential index n by one; entries still open are carried as censored — counted in the census, excluded from every denominator. The falsifier applies at n = 20: reject the working hypothesis if the verdict rate exceeds the survival rate.

Engaged (n)
0
Survival S
Verdict V
Mutation μ
Δ = S − V
Wilson 95% on S
Entries to falsifier
20
Status
Monitoring

No engaged entries yet. The plots appear as soon as the first case closes.

The equations, stated once

Neng = A + P + M + D + R + X · S = (P + M + R) / Neng · V = (A + D) / Neng · μ = M / Neng · Δn = Sn − Vn. The interval on S is the Wilson score interval at z = 1.96, which behaves at small n where the normal approximation does not. Open entries are right-censored at a seven-day window and never silently dropped.

Paste your notifications

Select your notifications pane on X — or a Substack thread — copy it, and paste the whole block here. Likes, reposts, follows and replies are read off it and grouped under the question each one responded to. Review what it found, drop anything mis-read, then import. Nothing is typed twice.

Log a question

Record the question at the moment it is posted — before any reply arrives — including the drift it is predicted to produce. That order is what makes the entry a test rather than a recollection.

Analysis script (Python)Workbook (Excel, v2.0)

Saved in this browser only. Export the file to publish an entry in the permanent record.

The script reads the exported JSON and reproduces the sequential statistics offline — S, V, μ, Δ, Wilson 95% intervals, the n = 20 falsifier check, and a three-panel plot: python analyze-question-tracker.py export.json --out plot.png --csv table.csv

  1. No entries yet. Log the next question at the moment you post it.