0.0
A controlled experiment was conducted over 90 days.
3.6
One AI tool per day, three tools rotated
6.839
in 30-day intervals.
8.619
Notion AI, then ChatGPT, then Clawed.
12.22
Same workflow profile, same 40-hour week baseline,
16.239
1,200 prompts logged.
18.14
The headline finding was the output quality verdict.
21.32
The buried finding was which tool the operator
23.739
reached for most often and why.
26.039
The verdict was not the one the experiment
28.12
design predicted.
29.219
The case file documents how the asymmetric information
32.299
posture traced through prior Fragment 0 case files
36.02
predicted the actual outcome.
39.74
The audit parameters were as follows.
42.7
90 days.
44.14
$60 in subscription cost total.
46.719
No other AI tools permitted in any of
49.6
the three operating windows during the test.
51.899
Every prompt logged.
53.38
Every output rated.
54.78
Every moment of cross-tool temptation noted in
57.619
the evaluation log.
59.219
Three measurement criteria, which tool the operator reached
62.439
for most often, which produced the highest output
65.56
quality, which felt fastest in real use.
68.54
The hypothesis at experiment start was that a
71.26
single tool would win across all three.
73.719
The hypothesis did not survive contact with the
76.68
data.
77.26
Three different tools won three different categories.
80.54
One of the winners was not the .
82.299
The results were not as good as the
83.26
results of the experiment.
83.579
Days 1-30.
85.519
Notion AI.
86.68
Plus tier with the AI add-on.
88.9
$20 a month.
90.26
Day 1 was the strongest performance window.
92.62
Notion AI operates inside the operator's existing workspace,
96.359
where meeting notes, project documentation, and historical email
99.799
content are already stored.
101.56
The capability to query the operator's own archive
104.26
and receive a three-second answer with citations
106.879
to the original document is something the other
109.12
two tools cannot match.
110.62
Day 7 was the failure event.
112.459
The operator attempted to use Notion AI as
114.879
a long-form writer for a blog draft.
116.939
The output rated 1.2.
118.84
.
118.879
The moins une carps.
124.92
For the exact answer should be still disadvantages
127.359
Cuomo will accept.
134.3
The
134.599
protagonist's
135.099
own
135.199
of seven
135.62
The end.
136.099
The first week was the strongest performance window.
138.52
Every short, repetitive task that had previously consumed
141.46
manual cycles, email rewrites, meeting
144.319
prep, brainstorming variations, was reassignable to a custom
148.28
GPT.
149.06
As documented in the previous Fragment 0 case
151.78
file on custom GPTs, the precompiled context
154.819
model produces a measurable throughput lift.
157.259
The bounded finding was speed.
159.36
Chat GPT consistently produced short task responses in
162.759
8-15 seconds, materially faster
165.08
than the other two tools in real use.
167.18
The failure mode was long context reasoning.
171.16
Day 61 through 90, Claude.
174.02
Pro subscription, $20 a month.
176.62
The capability under evaluation, long context reasoning.
180.199
A 15,000 word document was pasted and
183.24
queried for the three real arguments hiding under
186.02
the polite language.
187.06
The response was directly usable without modification.
190.46
Claude's writing did not present as AI generated
193.199
under blind review.
194.439
At its first attempt, the AI was not
195.06
able to read the text.
195.06
The results were not as clear as expected.
196.9
The output held coherence across multiple sections.
200.379
Claude projects with custom instructions and knowledge files
203.259
filled approximately 70% of
205.4
the role that Chat GPT custom GPTs serve,
208.199
sufficient for the experiment's purposes.
210.219
The trade-off, logged, Claude was measurably slower
213.479
than the other two tools.
215.199
The controlled head-to-head test.
217.4
Same input on the same day across all
220.08
three tools.
221.26
The task, a customer call transcript.
235.319
The character, lost.
251.12
1985 to ���oulder's
252.92
codenamed命
252.96
25
252.979
% 85
253.02
% 67
253.06
% 80
253.239
% systems
253.319
60
253.34
% 60
253.379
%
253.419
modification.
254.36
On this task, Claude won output quality.
258.92
Verdict 1.
260.079
Output Quality.
261.879
Claude.
262.74
The margin was not close.
264.819
For any task category where the response had
267.379
to hold coherent
268.319
thought across multiple sections, Claude produced outputs the
272.019
operator did not substantively rewrite.
274.48
The other two required cleanup.
276.439
Claude required approval.
278.42
The implication, logged in the audit,
280.779
for any operator whose deliverable is the writing
283.399
itself, Claude is the long-form
285.72
reasoning subscription.
287.0
Long-form documents, sales call analysis, strategy memos,
291.12
edits to
291.759
the operator's existing writing.
293.459
The classification persists across the audit's wider sample.
297.16
As documented in the mirror core case file,
299.899
the operator's own
302.78
Verdict 2.
303.759
Speed in Use.
305.199
Chat GPT.
306.62
The margin was not close.
308.819
For short repetitive tasks
310.259
under the same name, Claude was the only
310.759
one who could deliver the audit.
310.759
The audit was not
310.759
500 words of output.
312.279
Chat GPT averaged 8 seconds per task across
315.06
the audit.
315.8
Claude averaged 16
316.98
seconds for the same workload.
318.86
The pre-compiled context pattern, as documented in
321.779
the previous
322.22
Fragment 0 custom GPT case file, drops Chat
325.62
GPT's effective response time to approximately 4 seconds
328.959
because the context the other tools must receive
331.379
at every prompt is already loaded into the
333.72
agent's
334.1
working memory.
334.959
The classification persists across the audit's throughput sample.
338.199
For any operator
339.12
whose bottleneck is short-tasked, the operator's short
340.74
-tasked response time is approximately 4
340.74
seconds per task.
341.899
The editor
342.139
Comp галוכол
369.32
ggilам
366.939
ам
370.72
knows the operator's context without an explicit upload
373.699
every time.
374.519
Every clawed
375.339
prompt and every chat GPT prompt begins with
377.959
the operator re-explaining who they
379.639
are, what project they are on, which document
382.199
they are referencing.
383.16
The
383.399
friction is zero.
384.48
As documented in the memory market case file,
386.939
the data does
387.579
not stay confined to the buying decision matrix
391.68
derived from the audit for the
393.74
operator who must subscribe to only one.
396.139
If the word product is writing itself,
398.139
books, articles, strategy documents, contracts, clawed.
402.22
If the work product is
403.459
throughput, replies, brainstorms, quick edits, chat GPT.
407.36
If the work already
408.3
happens inside Notion and the workspace contains a
411.42
meaningful operator knowledge
412.6
base, Notion AI, with full awareness of the
415.399
retention posture documented in this
417.199
case file.
417.939
If the budget supports two, the pair is
420.18
clawed plus Notion AI, quality
422.439
plus retention.
423.68
Chat GPT becomes optional in that configuration.
426.72
If the budget
427.56
supports two, the pair is clawed plus Notion
428.12
AI, quality plus retention.
428.12
Chat GPT becomes optional in that configuration.
428.12
If the budget supports
428.12
all three, as the operator in this audit
430.16
did, the rotation pattern documented
432.279
accordingly.
433.879
The audit log is complete.
436.22
1200 prompts
437.399
categorized, three subscriptions evaluated, three different winners across
441.899
three different criteria.
443.54
The case file documents one operator's experiment.
446.959
The
447.399
retention posture documented for each of the three
449.839
tools has not been modified by
452.079
the vendors as of this writing.
453.779
The same retention posture applies in the test
456.54
subjects own configuration.
457.839
If the budget supports all three, the operator
458.1
in this
458.1
case file is not included.
458.54
The case file documents one operator's experiment.
458.54
The
458.54
result is a result of a single application
458.54
as it applies in yours.
460.0
The convenience of
461.42
Notion AI's three second context aware answer is
465.04
enabled by the same system surface
467.019
that this audit documents.
468.72
The case file does not close.
470.98
It waits.
472.04
Run the same prompt
473.42
against three AIs.
475.1
Compare the outcome.
