The Toll at the Oracle's Gate
When Vanika the merchant wished to consult the oracle of Pratisthana, she paid not in coin but in grain — one measure for every word she asked, and three measures for every word the oracle replied. "The asking is cheap," the oracle's priests explained, "for I read your question in one breath. The answering is dear, for I must shape each word in turn." Vanika, no fool, learned to ask in few words and to forbid the oracle from answering in many. Her grain lasted longer than that of any other merchant.
“You pay more for what is spoken than for what is heard. Constrain the speaking.”
The Long Preamble
Each visit, Vanika began with the same long preamble — her name, her caravan, her trade-goods, her contracts of the season — before her actual question. She paid for this preamble every single time. A young apprentice priest noticed and offered: "Mistress, leave your preamble with us. We shall keep it on a wax tablet. The next time you visit, you will pay only for the new question, not the old story." Vanika nearly wept with relief. The wax tablet cost her one-tenth of what the spoken preamble had cost. Her grain stretched ten times further.
“Pay full price for what is new. Leave what is unchanging at the gate.”
The Express and the Slow Caravan
When Vanika needed a yes or no urgently — which roads were open this morning — she paid for the express oracle. But when she wished to ask a hundred questions about next year's harvests, the priests offered her the slow caravan: drop your hundred questions in a sealed pouch, and the oracle will answer them all at her leisure overnight, for half the grain. Vanika, who did not need next year's harvest by tomorrow, took the slow caravan gladly. Half her grain remained for the urgent days.
“Speed is a luxury. When you do not need it, sell it back to the oracle for half-price.”
The Granary of Common Questions
Vanika noticed that many merchants asked the same questions: "Is the eastern road safe?" "What is the river's level?" "Is the festival market open?" Each merchant paid full price each time. She proposed to the priests a granary at the gate — common questions answered once a day and posted on a board for all to read. The priests, who had been quietly tired of answering "Is the river open?" forty times each morning, agreed. The board went up. Common grain costs vanished. Only the unique questions reached the oracle.
“A common answer asked once and posted feeds many; the oracle should hear only what is uniquely yours.”
The Oracle Who Spoke Forever
Once a young trader, new to Pratisthana, asked the oracle a vague question and forgot to set a limit on the reply. The oracle spoke for an entire afternoon. By dusk the trader's caravan of grain was empty and the answer was still flowing. The priests were merciful and stopped the oracle, but the loss was real. From then on, every contract at the gate carried a written limit: "Reply in no more than fifty words; halt at the first full stop after." The runaway answer never happened twice in the same merchant's life.
“An oracle without a stop-word will speak you into ruin. Cap every reply.”
The Ledger at the End of the Season
At the end of each trading season Vanika sat with her ledger and counted her grain. She did not count only the total. She counted by question type, by oracle, by caravan, by which questions had been answered from the granary board, by which had used the wax-tablet preamble. When she found that one kind of question — pricing in distant kingdoms — was eating a third of her grain, she negotiated a special wax tablet for that question alone, halving its cost. The merchants who did not keep ledgers paid more, and could not even say why. Vanika prospered. Her granary swelled.
🪔 Deepak — the lamp of meaning · what this fable means in code
The grain Vanika pays is the token bill — and the threefold price of replies is the real difference between cheap input tokens (parallel prefill) and expensive output tokens (sequential decode). The wax-tablet preamble is prompt caching: a long stable system prompt billed at a fraction of normal price after the first call, so long as the prefix is unchanged and reused. The slow caravan is the batch API — same model, no SLA, half the cost — perfect for any workload that does not need real-time latency. The granary board of common questions is semantic caching: store frequent queries and their answers; serve repeats from memory; never pay the oracle twice for the same question. The runaway oracle stopped by a written limit is the discipline of `max_tokens` and termination conditions — without them, a single bug or prompt injection can drain a season's grain in an afternoon. And Vanika's ledger broken down by question type is per-feature cost telemetry — the only honest way to know which conversations are eating the budget, and the first step to optimising any of them.

