docs: field notes v3 + v4 (published on the HF blog) join v1/v2 in-repo

docs/field-notes-v3.md

@@ -0,0 +1,126 @@
+# The crash that vanished: control and emergence in a five-model economy
+
+*Field notes from the Build Small Hackathon, June 2026. Third installment.*
+
+In the first of these notes I told a story I was proud of. I drew a Wood Legend
+called the Run on Oona's Hoard, a 1929 bank run reskinned as woodland folklore, and
+watched the owl who keeps the honey read the panic and start liquidating. The flood of
+supply crashed the honey price from 10 down to 3 over the next few turns. Nobody
+scripted it. A reskinned bank run made an agent dump an asset, and the dump moved a
+price. That was the whole thesis: give a small model a role and a budget, and emergent
+market behavior falls out for free.
+
+Then I rebuilt the wood, and the crash stopped happening. This installment is about
+why, because the failure taught me more about building on agents than the original
+success did.
+
+## Five labs, five minds
+
+The rebuild swapped one model running five creatures for a council of five different
+labs' small models, each driving its own creature: an OpenAI model, an NVIDIA model,
+an OpenBMB model, and a half-billion-parameter model I fine-tuned myself running two of
+them. The point was honesty. If the claim is that small models can run a living
+economy, the strongest version of that claim is five distinct architectures making
+distinct choices in the same market, not one model wearing five hats.
+
+That heterogeneity is exactly what broke the story I had already written up.
+
+## The price is whatever the agents decide to trade at
+
+I rebuilt the operator side too. The player is now a financier who works from the
+shadows: short a good, whisper a true tip to set up its fall, spring the legend, and
+collect when the price craters. I made that loop legible on the screen, with an
+objective, a scoreboard, and a one-click first trade. Making a promise visible is the
+fastest way to discover the promise is false.
+
+Because when I shorted honey and sprang the Run on Oona's Hoard, honey did not crash.
+It rose. The council models, reading a rumor that the vault was empty and a tip that
+the crop was doomed, did not dump honey the way the original single model had. They
+hoarded it. Scarcity, not a fire sale. The short lost money, and the headline the
+narrator wrote, with no irony, was that the honey gamble had soured.
+
+This is the lesson, and it is not specific to a game. In an agent economy the reference
+price is not a dial you turn. It is the residue of what the agents actually choose to
+trade. The original crash was real, but it was contingent on one model's disposition,
+not a robust property of the system. Change the population, and the emergent behavior
+you documented can simply evaporate.
+
+## Three ways to fail
+
+I spent three live runs trying to make the crash come back by pushing on the economy
+from the outside, the way you would shock a textbook supply and demand model.
+
+First I left the legend as a pure rumor and trusted the agents to react. They did not
+sell. Second I dumped a windfall of honey into every creature's stores, reasoning that
+a glut would collapse demand and pull the price down. That worked beautifully against
+my test policy, a rule-based stand-in I use for fast offline runs, because the test
+policy follows a mechanical wants-threshold: flood its inventory and it stops buying.
+The live models ignored the windfall and traded on their own read of the room. The
+gambit lost again. Third I sized the short up, which only made the loss larger.
+
+Three recordings, three losses: minus fifteen, minus twenty-six, minus twenty-seven
+pebbles, when the entire premise was that this was how you made money. The pattern was
+the warning. Every lever I pulled was an input to the agents' decision, and the agents
+were free to decline. You cannot steer a heterogeneous population of models with a
+mechanical shock, because the shock only biases a choice they still get to make.
+
+The trap inside the trap is worth naming on its own. The fix that worked against my
+fast test policy gave me false confidence and cost me a live run to disprove. When the
+cheap stand-in and the real agents disagree, the stand-in is the one lying, and any
+result that only reproduces under the stand-in is not a result.
+
+## Author the seam, do not push the inputs
+
+The resolution was to stop trying to convince the agents and to make the panic true by
+construction. A bank run is, definitionally, a crash. So the legend now crashes its
+good at settlement, after the market has finished clearing for the turn, by overwriting
+the reference price directly. The agents trade all they like; then the run lands as a
+fact, the price halves, and the short that front-ran it settles into profit. The crash
+is no longer a behavior I hope for. It is an authored consequence I impose at the one
+seam where nothing downstream can argue with it.
+
+That sounds like giving up on emergence, and it is the opposite. The emergent layer,
+five models trading, gossiping, hoarding, forming grudges, is still doing all the work
+that makes the wood feel alive. What I learned is that you do not get reliable outcomes
+by pushing harder on emergent inputs. You get them by choosing the precise seam at
+which to author a deterministic override, and leaving everything upstream free. Emergence
+for texture, authored control for the moments that have to happen. The craft is knowing
+which is which, and where the seam sits.
+
+| Attempt | Mechanism | Honey at settlement | Gambit P&L |
+|---|---|---|---|
+| Original, one model | that model chose to dump | 10 to 3 | the showcase win |
+| Council, rumor only | five models chose to hold | rose on scarcity | minus 15 |
+| Council, inventory glut | demand collapse, test policy only | barely moved | minus 26 to 27 |
+| Council, settlement override | price crashed post-clearing, by fiat | halved reliably | plus 40 |
+
+*Table 1. The same gambit across four worlds. The crash was emergent and fragile under one model, absent under a heterogeneous council, and reliable only once it was authored at the settlement seam.*
+
+## What I took away
+
+Three things, and all three outlive the game.
+
+First, emergence is contingent, not durable. Behavior you observe and write up from one
+population of agents can vanish when you change the population, even if nothing else
+changes. Treat a single impressive run as an anecdote, not a property, until it survives
+a different cast.
+
+Second, you do not control a market of agents by shocking its inputs. Supply and demand
+levers only bias choices the agents are still free to make, and a heterogeneous council
+will frequently decline. Reliable outcomes come from authoring at a settlement seam,
+downstream of every decision, not from pushing harder upstream.
+
+Third, the cheap simulator that lets you iterate fast is also the one most likely to
+flatter a wrong fix. When the stand-in and the real agents disagree, believe the agents.
+
+I build agent-based market models for a living, and I have made every one of these
+mistakes at larger scale and higher stakes than a wood full of woodland creatures. It
+was useful to make them again somewhere the only thing at risk was a pile of pebbles
+and a story I had told too confidently the first time.
+
+Small models, big adventures, and a crash you have to author yourself.
+
+---
+
+*Try it: the [Space](https://huggingface.co/spaces/build-small-hackathon/thousand-token-wood-sim).
+Open agent traces: the [dataset](https://huggingface.co/datasets/build-small-hackathon/thousand-token-wood-traces).*

docs/field-notes-v4.md

@@ -0,0 +1,121 @@
+# The gambit that decayed: when the wood learned to remember
+
+*Field notes from the Build Small Hackathon, June 2026. Fourth installment.*
+
+The last installment ended on a note I was a little too pleased with. I had spent three
+losing runs trying to make a panic crash a price, given up on steering the agents, and
+authored the crash at the settlement seam instead. After the market cleared, the run
+landed as a fact, the price halved, and the short that front-ran it booked a clean profit.
+The same gambit paid the same amount every time. I called that control, and I meant it as
+praise.
+
+A clean strategy that pays the same amount every time should have made me nervous, and
+eventually it did. In a real market, a repeatable edge that everyone can watch you run is
+not an edge for long. The people on the other side learn your face. They price you in.
+The honest name for the thing that erodes a visible edge is adverse selection, and my
+wood did not have any. The creatures I lied to forgot the lie by the next turn and lined
+up to be fooled again. So I asked the obvious question. What happens if the wood
+remembers?
+
+## The memory was already there
+
+Here is the part that taught me the most, and it is a lesson about reading your own
+system before you extend it. I went in expecting to build a suspicion engine. I did not
+need to. The creatures already carried a persistent feeling toward the financier, and a
+tip that turned out to be a lie already soured it. I had simply never given that feeling
+anything to do. It sat in state, updated every time I burned someone, and changed nothing.
+
+So v4 is not a new system. It is teeth on a signal that was already in the world. A
+creature whose feeling toward the financier drops past a threshold becomes wary, and a
+wary creature does two things it never used to do.
+
+First, it stands on the other side of my crash. When the run lands and I overwrite the
+price at the settlement seam, I now count the wary creatures who hold or produce the good
+I am crashing, and I let them push back. Each one blunts the crash, softening the price
+drop toward no move at all. The mechanic lives at the exact same seam as the authored
+crash from last time, which is the point. I did not move the lever. I let the agents lean
+on it.
+
+Second, it talks. A wary creature feeds the magistrate who investigates suspicious wins,
+raising the heat on me every turn it stays soured. My manipulation history, which used to
+be free, now has a standing cost. The more creatures I have burned, the faster the
+inquiry comes for me, whether or not I do anything new.
+
+Neither effect required a new field, a new prompt, or a new model call. Both read a number
+the engine had been keeping all along.
+
+## The same gambit, decaying
+
+Then I ran the same short into the same legend, and watched the payoff come apart. Holding
+the short fixed and letting the wood wise up one creature at a time, the gambit decays on a
+schedule:
+
+| Wary creatures positioned against me | What the crash does | Gambit payoff |
+|---|---|---|
+| 0 | full crash, price halves | 20 |
+| 1 | crash blunted a quarter | 15 |
+| 2 | crash blunted by half | 10 |
+| 3 or more | crash blunted to the floor | 4 |
+
+*Table 1. The identical short into the identical legend, as more burned creatures stand
+against it. The payoff is the engine's deterministic settlement on a fixed four-unit short,
+not a single lucky run. The floor at three is a deliberate cap: a soured wood blunts the
+gambit but never fully erases it, so manipulation degrades toward marginal rather than
+impossible.*
+
+The shape is the whole story. The first lie costs me nothing I can see. By the third, the
+crash I authored is a shadow of itself and the magistrate is at the door. The strategy that
+paid the same amount every time now pays less every time I use it, because using it is what
+teaches the wood to defend against it. That is adverse selection, built from a feeling the
+creatures already had and I had been ignoring.
+
+## Make the decay legible or it is not a lesson
+
+A mechanic the player cannot feel is just a number moving in the dark. The thing that turns
+this into something you understand while you play is a small panel I added to the operator
+console, and it is the one piece of craft from this round I would defend hardest.
+
+I borrowed the idea from another entry in the same hackathon, a single-character drama whose
+whole interface is one conversation and one panel of meters that twitch the instant your
+words land. The lesson there is not the layout. It is that state should visibly move on your
+action. So the new panel shows, for the gambit you have armed, the expected payoff if you
+spring the legend right now, and a gauge of how wary the wood has grown. The expected payoff
+is computed with the exact same formula the settlement seam uses, so the meter does not lie
+about the outcome. Sour a creature with a false tip, and you watch the projected profit drop
+before you have done anything else. The cost of lying stops being an abstraction and becomes
+a number falling in front of you.
+
+That is the difference between a system that punishes you and a system that teaches you. The
+punishment was always in the engine. The teaching is in showing it to you in time to change
+your mind.
+
+## What I took away
+
+Three things, and like last time all three outlive the game.
+
+First, your edge is consumed by the people you trade against learning who you are. A
+manipulation you can run once is a tactic. A manipulation you run repeatedly against agents
+who remember is a countdown. The visible, repeatable, always-profitable strategy is the one
+most certain to decay, and the decay is caused by the using.
+
+Second, read your own state before you build more of it. The adversarial behavior I set out
+to construct was already latent in a feeling the creatures carried and I had wired up but
+never used. Most of the work was deletion of my own assumption that I needed something new.
+The cheapest feature is the one already sitting in your world with nothing to do.
+
+Third, a cost the user cannot see is not a deterrent, it is a trap. The same fact, the wood
+is souring on you, is either an unfair surprise or a fair warning depending entirely on
+whether you surfaced it in time. Building the meter that shows the decay was not polish. It
+was the part that made the mechanic honest.
+
+I build agent-based market models for a living, and the most expensive lesson in that work
+is the one this little wood just charged me a pile of pebbles to relearn. The market you can
+fool forever is a market you built wrong. The one worth modeling is the one that learns your
+face and makes you pay for the last time you lied.
+
+Small models, big adventures, and a wood that finally remembers.
+
+---
+
+*Try it: the [Space](https://huggingface.co/spaces/build-small-hackathon/thousand-token-wood-sim).
+Open agent traces: the [dataset](https://huggingface.co/datasets/build-small-hackathon/thousand-token-wood-traces).*