What a programme looks like¶

This is real output, produced by mimem build tests/fixtures/docs/synthetic-paper.pdf, from the synthetic two-column paper the test suite generates. Nothing here was written by a model: every sentence is either the paper's own, carrying the span it came from, or one of about a dozen named templates. The elaboration layer that writes glosses, concrete anchors and analogies is M5.

The point of reading it is the shape. Compare it with what a text-to-speech engine does with the same PDF: it starts at the running head, reads the affiliations, says "open square bracket twelve", pronounces 0.0837 as a string of digits, tells you to see Figure 3, and never once asks you anything.

audio.md — what the engine says¶

This is Interphase repair, not particle fracture, limits silicon anode life. It's by A. Researcher and colleagues. Here's the problem they set out from. Silicon-graphite composite anodes lose capacity faster than graphite alone, and the usual explanation blames fracture of the active particles. It'll take about eight minutes.

Before we start, here are three questions to hold on to. You're not meant to know the answers yet.

First. What did the part on the paper in brief say?

(…second, third…)

Part one of three. The paper in brief.

We show instead that the dominant loss channel is repeated repair of the solid electrolyte interphase after volume-driven cracking. Cells cycled five hundred times at zero point five C retained eighty two point one plus or minus one point four percent of their initial capacity, corresponding to a mean loss of zero point zero eight, three seven percent per cycle, p less than zero point zero zero one.

That was the paper in brief.

What did the part on the paper in brief say? Take a few seconds.

(three to six seconds of silence, carried in manifest.json)

Here's the answer about the paper in brief. Cells cycled five hundred times at zero point five C retained eighty two point one plus or minus one point four percent of their initial capacity…

That was one of the questions I asked at the start, the one about the paper in brief.

Part two of three. Introduction.

(…)

There is an equation here. It is in the written notes.

Discharge capacity versus cycle number for silicon-graphite, circles, and graphite, squares, anodes over five hundred cycles at zero point five C.

(…)

That's the paper. Now three questions across all of it, in mixed order.

Five things in that text are rules doing their job:

And one is a rule declining to guess: "There is an equation here. It is in the written notes." A figure or an equation the deterministic verbalizers cannot speak is announced rather than skipped or invented. Those become real descriptions in M5.

study.md — the same material, for reading¶

## Results and Discussion

*where we are:* Part three of three. Results and Discussion.

Capacity declined smoothly over 500 cycles with no abrupt step, which is the signature of
a process that repeats a little every cycle rather than a single mechanical failure.  <sub>p2</sub>

**Q.** What did the part on Results and Discussion say? Take a few seconds.

**A.** Here's the answer about Results and Discussion. Particles recovered from cycled
electrodes were largely intact…  <sub>p2</sub>

The written track keeps the exact numerals the audio track had to spell out (NUM-06), the page each claim came from (GRD-04), and a label on every sentence mimem wrote rather than the paper (VOI-02).

manifest.json — how to argue with it¶

"concepts": {
  "c_a1b2c3d4e5": {
    "canonical": "solid electrolyte interphase",
    "difficulty": 0.62, "importance": 0.88, "budget": 0.546,
    "signals": {"abstractness": 0.71, "interactivity": 0.55, "position": 1.0, "reprise": 0.9},
    "exposures": [
      {"beat_id": "t_4e4756d5592f", "at_seconds": 88.4,  "form": "statement"},
      {"beat_id": "t_9c1e0a77b210", "at_seconds": 271.0, "form": "callback"},
      {"beat_id": "t_1f77b4d0aa93", "at_seconds": 452.6, "form": "prompt"}
    ]
  }
}

Three exposures at increasing intervals, each one a different sentence of the source, because a repeat that is the same sentence twice produces the feeling of knowing without the knowing (REP-01). The scores that decided it earns three rather than two are recorded beside it, so a ranking you disagree with is a number you can change rather than a black box you can only complain about — put it under overrides in registry.json and it survives every future run.

When a beat puzzles you:

mimem explain out/paper --beat t_9c1e0a77b210

which prints the beat's type, the rules that produced it, its cost in seconds, the concept it serves and the source sentence behind it.