The Six Honest Signals

From the bare timestamps of a real chat — without reading a word of the content — you compute response times, the balance of contributions and the ratio of giving to taking. An intuition about a group becomes a measurement.

What it's about

Put five people in a room and give them a task — after a few minutes something arises that you can almost touch: a mood, a rhythm, a kind of chemistry. For a long time that was considered unmeasurable. This chapter shows that it is measurable, because a group's chemistry lives in its honest signals — in things no one steers consciously.

Six such signals predict surprisingly well whether a group becomes productive: whether there is a strong, connecting leadership and whether it shifts rather than sticking to one person; how quickly members react to each other; how evenly the contributions are shared; how honestly moods may be expressed; and how much the group develops a language of its own. In this activity you pick out three of them that can already be computed from bare timestamps — response time, contribution balance and giving and taking. The leadership question (betweenness centrality) follows in Activity 8.1.

A little background

What we read from timestamps. Every chat line carries three plain facts: when, who, and (the one we ignore here) what. From when and who alone, three measures arise:

• Contribution balance — how evenly the contributions are spread across people. Does everyone talk roughly the same amount, or do a few dominate? A very lopsided chat is a warning sign.
• Response time — the median time between one message and the next from a different person. Short response times mean people listen to each other and react.
• Giving and taking — does someone react to others roughly as often ("giving") as their own messages trigger reactions from others ("taking")? We count both within a short time window. Whoever's messages create a lot of resonance but who reacts little themselves stands out here — as does whoever barely takes part in the lively back-and-forth.

These are rough but honest approximations — they turn "the group feels lively" into a number you can compare.

Exporting the chat

1. Start the export. In WhatsApp open the chat → Menu → "More" → "Export chat" → "Without media". You get a .txt file.
2. Consent & anonymise. Ask the group. Before the analysis, replace the real names with A, B, C … — or let the code do it (see below).
3. Put the file in place. As chat.txt, in the same folder as the script.
4. Analyse. Run the script and read off the three measures.

Analysing with Python

The code splits each line into a time and a sender, throws the content away and computes the three measures. Full, commented code on GitHub.

import re
import pandas as pd

WINDOW = 60   # minutes: what still counts as a "response"

# WhatsApp line:  "15.03.26, 17:44 - Mira: Text..."
pattern = re.compile(r"^(\d{1,2}\.\d{1,2}\.\d{2,4}),?\s+(\d{1,2}:\d{2})\s+-\s+([^:]+):")

rows = []
for line in open("chat.txt", encoding="utf-8"):
    m = pattern.match(line)
    if m:
        date, clock, name = m.groups()
        ts = pd.to_datetime(f"{date} {clock}", dayfirst=True, errors="coerce")
        rows.append((ts, name.strip()))

df = pd.DataFrame(rows, columns=["time", "who"]).dropna().sort_values("time").reset_index(drop=True)

# --- anonymise: real names -> A, B, C ... ---
names = {n: chr(65+i) for i, n in enumerate(df["who"].unique())}
df["who"] = df["who"].map(names)

# 1) contribution balance
share = df["who"].value_counts() / len(df)
p = share.sort_values().values                         # inequality as Gini
gini = 1 - 2*sum((len(p)-i-0.5)/len(p) * pi for i, pi in enumerate(p))

# 2) response time: median to the next message from ANOTHER person
df["dt_min"] = df["time"].diff().dt.total_seconds() / 60
df["other"] = df["who"].ne(df["who"].shift())
response_time = df.loc[df["other"], "dt_min"].median()

# 3) giving & taking within the time window
time, who, n = df["time"].tolist(), df["who"].tolist(), len(df)
gives = {p: 0 for p in set(who)}     # I react to another person
takes = {p: 0 for p in set(who)}     # my message triggers reactions
for i in range(n):
    for j in range(i-1, -1, -1):                       # giving: do I react?
        if who[j] != who[i]:
            if (time[i]-time[j]).total_seconds()/60 <= WINDOW:
                gives[who[i]] += 1
            break
    responders = set()                                 # taking: who reacts to me?
    for k in range(i+1, n):
        if (time[k]-time[i]).total_seconds()/60 > WINDOW or who[k] == who[i]:
            break
        responders.add(who[k])
    takes[who[i]] += len(responders)

print("Contribution share (%):")
print((share*100).round(1).sort_values(ascending=False).to_string())
print(f"\nInequality (Gini, 0=equal .. 1=one person): {gini:.2f}")
print(f"Median response time between people:        {response_time:.0f} min\n")
print("Giving & taking (given / received):")
for pn in sorted(set(who), key=lambda x: -takes[x]):
    g, e = gives[pn], takes[pn]
    print(f"  {pn}  given {g:2d}   received {e:2d}   G/T {g/e:.2f}" if e
          else f"  {pn}  given {g:2d}   received {e:2d}   G/T  –")

Worksheet

When it sticks

Food for thought

• You have measured a group without reading a single word of its content — from the pattern of communication alone. That is exactly what the author once did with his Deloitte mailbox (Chapter 8): not the content of the emails, but their pattern. Structure often reveals more than content.
• These measures are approximations, not verdicts. A high Gini does not mean "bad group" — perhaps someone is giving an important presentation right then. Numbers are a reason to look closer, not the verdict itself.
• Because these signals are so hard to fake, they are powerful — and therefore delicate. Whoever does such analyses carries responsibility: aggregated for the team, personal only for the person. That is the line that turns observation into trust.

Extension

• Over time. Split the chat into weeks and see whether the contribution balance moves — does the dominant person change, or does the leadership stick? The moving is itself one of the six signals.
• Daily rhythm. Plot the messages by time of day. When is the group most active? A first look at the "pulse" of a community.
• Bridge to 8.1. From the "who replies to whom" of this activity you can draw a network directly — and compute the connecting leadership (betweenness) in it. That is exactly what you do next.
• Bridge to Chapter 7. Now you have measured the structure. The symbiont tool of the next chapter also reads the content and assigns each person a role — Bee, Ant, Butterfly, Capybara, Leech.