Lesson 28 of 32

The Mirror of Self-Critique

Reflection Loops

Build the checking into the prompt itself. Give a short checklist of what the finished work must achieve - four to six things a careful reader could answer yes or no - then tell the model to draft, score itself against each point while quoting the line that falls short, and rewrite. Two rounds, not ten. It works because "make it better" has nothing to measure against, while five checkable criteria do.

Reach for it whenWhen you can say precisely what good looks like - a cover letter, a plan, a story for a particular class - and you would rather receive a draft that has already been through an editor than one you must edit yourself.

Be clear about what does not transfer: this is a search tool, and the creator-critic loop does not run here. It will not hold two characters across a conversation, it will not rewrite its own draft twice and show you the scores, and each message is treated as a fresh question rather than a position in a cycle. What it does is the one job that self-critique structurally cannot do. A rubric checks a draft against your criteria; it never checks the draft against the world, so a model will score an invented figure as accurate without hesitation. Bring the factual claims of your final draft here for an external, cited check - and come here first, too, to find out what the rubric should contain, because the criteria that matter are often a researchable question rather than a matter of taste.

Worth knowingDo not try to run the loop here. Asked to play a strict critic it produces a sourced essay about what makes cover letters or children's stories good, which reads well and marks nothing, and asked to rewrite it summarises instead. Two of these three examples also have a limit no search tool can cross: nothing can verify what you did during your internship, and no source can tell you whether your Class 3 children will laugh. For that, your own reading-aloud is the test. The farm plan is the exception and a real one - "the five strongest reasons this could fail" is largely an empirical question, and sourced objections are worth more than invented ones.

The loop runs elsewhere; the fact-check belongs here.
Research what the rubric should contain before you write it.
Bring the final draft's checkable claims, not the whole draft.
Objections grounded in reported data beat invented ones.
Nothing here can verify your own experience. Read that yourself.

The Vague Polish vs. The Scored Rubric

Instead of

Improve my cover letter for this job.

Perplexity version

The scoring loop for this letter belongs in a writing tool. What belongs here is the question the rubric itself rests on - what a good letter for this job actually has to do - and that is researchable rather than a matter of taste.

I am applying for junior accounts assistant posts at logistics firms in Odisha. B.Com 2025, Tally and Excel, six months with a CA firm.

With sources, and please say where the evidence is thin:

1. What do currently advertised junior accounts assistant roles in India list as requirements? Look at live listings and tell me which skills recur, in what words, and what is asked for more often than a fresher would expect.
2. What is reported about how Indian employers and recruitment firms actually use cover letters for entry-level accounts roles - read, skimmed, or ignored in favour of the CV? Cite recruiters rather than career-advice blogs where you can.
3. Which software and compliance terms appear most often in these listings - Tally versions, GST return work, e-way bills, Excel functions - so I can match the advertisement's own vocabulary rather than guessing it?
4. What length and structure do Indian recruiters recommend for a fresher's cover letter? Where the advice conflicts, show me the conflict instead of averaging it.

Finish with five or six criteria, drawn from what you found rather than from convention, that I can use as the rubric when I run the drafting and scoring loop in another tool.

Open Perplexity 1,449 characters

Then loop it

Narrow to logistics and supply-chain firms specifically. What do their accounts openings ask for that a general accounts listing does not? Cite the listings.
I have a final letter now, scored elsewhere. I am not pasting the whole thing - just these checkable claims: that Tally Prime is the current version in common use, that GST returns are filed monthly for a firm of this size, and that my Bhubaneswar internship covered e-way bill work. Check the first two against sources and tell me the third is mine to verify, not yours.
Save this thread in a Space called "Job applications - accounts" with the criteria list, so each application starts from researched standards rather than from a fresh guess.

Why it worksThe rubric is the part of this loop that can be researched, and criteria drawn from live listings and recruiters beat five invented standards - the scoring and rewriting then happen in a tool built for writing.

The Cheerful Plan vs. The Plan That Argued With Itself

Instead of

Should I grow vegetables instead of wheat?

Perplexity version

This is the example in the chapter where an answer engine genuinely improves on self-critique. "The five strongest reasons this plan could fail" is mostly an empirical question, and objections carrying sources are worth more than objections a model composed for you.

I farm 10 acres of wheat and paddy near Moga, Punjab, and I am considering 2 acres of tomato and capsicum next season. No vegetable experience, no cold storage. Keep this in a Space called "Vegetable shift".

Answer each separately, with citations and the date of every figure:

1. What have tomato and capsicum prices done in Punjab mandis over the last three seasons? Give the reported range and the crash months. Agmarknet or the state marketing board rather than news summaries, where possible.
2. What has been reported about tomato gluts and distress sales in Punjab and neighbouring states - how often, and how far did prices fall?
3. What does Punjab Agricultural University or the Krishi Vigyan Kendra network currently advise about vegetable cultivation in the Moga belt - varieties, season, water requirement, and the warnings they give first-timers? Link the extension material itself.
4. What is reported about harvest labour availability and wage rates for vegetable picking in this region, as against wheat and paddy?
5. What is documented about small-scale cold storage and crate schemes - subsidy, eligibility, where to apply? Link the official page and note whether it is open.
6. Is there reported evidence of Punjab wheat-paddy farmers who diversified into vegetables and what happened to them - successes and failures both? Cite the coverage, and tell me if only the successes were written about, because that would change what the evidence means.

Where a figure cannot be sourced, mark it unverified rather than filling the gap.

Open Perplexity 1,819 characters

Then loop it

Point 2 is my real fear. Narrow to the worst reported tomato crash in this region in the last five years: how low did the mandi rate fall, for how many weeks, and what did growers say they did? Cite the accounts and give their dates.
Now the question that decides the plan. For a 2-acre grower with no cold store, what do sources say about where the produce actually goes - mandi, local trader, direct sale - and how the rate differs between them? Cite each, and say plainly if the data is thin for holdings this small.
Compress this into a sourced brief of about 250 words: the price history with dates, the glut evidence, the extension advice, the labour position, the scheme status. That is what I will take to another tool to build the plan and run the argue-against-itself round, and what I will carry to the KVK - so their advice can be set against what is published rather than against my hopes.

Why it worksMaking the plan argue with itself exposes risks a cheerful answer hides, and here the objections arrive with prices, dates and sources behind them - so what the plan has to survive is the record rather than an invented worry.

The Lone Writer vs. The Critic and the Creator

Instead of

Write a story for kids about saving water.

Perplexity version

The honest note first: the creator-critic loop will not run here. This tool will not write a children's story, hold two characters, and rewrite itself twice - ask it to and you get a well-sourced essay about children's writing instead of a story. Run the loop in ChatGPT, Claude or Gemini. Use this for the one thing that makes the rubric real.

I teach Class 3 in Kottayam and need a 250-word English story about a girl who saves water at home.

With sources:

1. What does the Kerala state syllabus or SCERT expect of Class 3 English reading - vocabulary range, sentence length, any published word list? Link the official document, and say plainly if nothing usable is published, because then my own judgement is the standard and I should know that.
2. What does published guidance on early-reading material recommend about sentence length and unfamiliar words per page at this level? Academic focus for this one.
3. What has been published on whether moral stories work better with the lesson stated or left for the child to find? Cite the research, and tell me if it is contested.
4. Are there Kerala or NCERT-published water conservation materials for primary classes that my story ought not to contradict? Link them.
5. What water-saving habits are realistic in a Kerala household - what do local sources actually describe, as against generic advice about shorter showers? I would rather my story describe a tap and a bucket my children recognise.

Finish with five or six criteria drawn from the above, which I will use as the Critic's rubric in another tool.

Open Perplexity 1,566 characters

Then loop it

Point 1 came back thin. Search specifically for SCERT Kerala Class 3 English textbooks or teacher handbooks and tell me what reading level they actually assume, quoting from them if you can.
Now the household detail. What do Kerala sources say about domestic water use and shortage in the summer months - wells running low, tanker supply, storage in drums? I want the story's details to be ones my class has seen, and I would rather take them from reported accounts than from imagination.
Give me the final criteria list and the household details as one page. The story, the Critic and the two rewrite rounds go to another tool - and the last test is not the rubric anyway: it is reading the thing aloud to thirty eight-year-olds and seeing whether they laugh.

Why it worksSplitting creator and critic needs a rubric with something real in it, so the criteria and the household details are researched here and the two roles - which this tool cannot play - are run where they work.

Loop it

Read the scores before the story. If every criterion scores a perfect 5, the loop is flattering itself - reply, "Be stricter: quote the weakest line for each point and rewrite." If the final version improved one point but broke another, add that point to the rubric and run one more round. Stop after two or three rounds; more rarely helps. Then do your own human read, checking facts the rubric cannot catch. When a rubric works, save it and paste it into future prompts as a standing measure of quality.