Kiln Quest

Tracks

1. 01📉The Rule That Lost

   0/4 lessons

   A defect that survived three generations of rules everybody agreed with

   Every workshop worth building starts at a number, because without one you are just expressing a preference. This track is the number: sixty-seven of seventy-nine documents in a bilingual corpus shipped with the same defect, under three successive generations of explicit rules that every author knew, agreed with, and was trying to follow. That is not a discipline problem — a discipline problem looks different, and this track shows you how to tell them apart. From there it works out what follows: why a rule is just tokens competing for attention against everything else in the room, what it means to move a rule from discipline into structure, and what kind of tool that leaves you with — one that orchestrates and records, and never does the work itself.

   Lesson list (4)

   • 01
     Sixty-Seven of Seventy-Nine~12 min · measurement, process, failure-analysis, bilingual
   • 02
     A Rule Is Just Tokens in a Context~12 min · attention, process, prompting, cognition
   • 03
     The Five Rungs From Advice to Absence~12 min · architecture, guardrails, process, design
   • 04
     Orchestrate and Record, Never Author~11 min · architecture, boundaries, workshop, design

   ✓Quiz · 4 questions→

2. 02🚪The Human Gate

   0/4 lessons

   A workshop that cannot start itself, and why that is a feature you build on purpose

   The second track is about the one capability this workshop deliberately does not have: the ability to begin. Nothing here starts work — not a timer, not a queue consumer, not a retry loop, not a server-side process launcher. A unit of work sits as a record until a person opens a session and claims it, and then the shape of that work is agreed before any of it is produced. This track works through why that gate is structural rather than polite: what a plan gate actually decides, why a click that opens a visible session is categorically different from a schedule that opens one, and the sharpest consequence of taking approval seriously — that an approval attaches to the premises it was given, so a wrong premise voids its own mandate, even when the person approving said yes in good faith.

   Lesson list (4)

   • 01
     The Pipeline That Refuses to Run Itself~12 min · invariants, automation, architecture, design
   • 02
     The Plan Gate Decides the Shape, Not the Schedule~12 min · process, planning, collaboration, design
   • 03
     A Click Is Not a Cron~11 min · automation, security, architecture, dispatch
   • 04
     Approval Attaches to Premises, Not Just Outcomes~12 min · judgment, process, evidence, collaboration

   ✓Quiz · 4 questions→

3. 03📜Contracts, Not Conscience

   0/5 lessons

   How you forbid an action that leaves no trace

   This is the technical heart of the workshop. A stage contract is a table: for each stage of a pipeline, the inputs it is allowed, the inputs it is forbidden, and what closing it means. Written that way, the contract stops being advice and becomes something a harness can hold. The track then walks into the hardest case in the whole design, which is a prohibition on reading. Writing leaves a diff you can check afterward; reading leaves nothing the gate can reach, so no post-hoc check inside the repository can prove a run stayed clean. That single asymmetry forces the guard out of the checks and into the tool call itself, forces the seal to be scoped to stages rather than to whole jobs, and ends at the strongest version available: a stage that receives its material as text and never learns that a path exists.

   Lesson list (5)

   • 01
     A Stage Is Its Allowed Inputs~12 min · contracts, architecture, pipelines, design
   • 02
     The Read That Leaves No Trace~13 min · guardrails, evidence, verification, incident
   • 03
     Put the Guard in Front of the Tool Call~13 min · guardrails, hooks, implementation, design
   • 04
     Prefix Seals and Suffix Seals~12 min · guardrails, pipelines, design, scoping
   • 05
     Ship the Text, Withhold the Path~12 min · guardrails, design, sampling, architecture

   ✓Quiz · 4 questions→

4. 04❄️Continuity and Boundary

   0/4 lessons

   Keep the context where it helps; break it where it contaminates

   The workshop places continuity and boundaries deliberately, and this track is the reasoning behind each placement. Inside a stage, the session that did the work keeps its context, because the minute after producing something is when a question about it is cheapest to answer. Between authoring and repair, the context is broken on purpose, because the thing that helped you write is the thing that will stop you seeing. The track works out why one pass cannot both author and de-contaminate — the sequence is forced, not stylistic — why this is the ordinary practice of publishing rather than a machine ritual invented here, and what the honest failure modes of a cold pass are, including the one that already happened and the one that would pass every check you could build.

   Lesson list (4)

   • 01
     One Dial, Two Opposite Needs~12 min · context, process, design, collaboration
   • 02
     You Cannot Forget on Command~13 min · translation, process, architecture, cognition
   • 03
     This Is Just How Books Are Made~11 min · publishing, process, history, craft
   • 04
     The Two Ways a Cold Pass Fails~12 min · failure-modes, process, verification, craft

   ✓Quiz · 4 questions→

5. 05🔍Review Is the Spine

   0/5 lessons

   Peer review as the only source of decorrelation you have

   Peer review here is not a quality gate bolted onto the end of a pipeline; it is the reason the pipeline is shaped the way it is. This track starts from the finding that made it structural — swapping which system does the writing does not fix generation, because the failure is universal to anything that writes while holding a source, and what single-system dependence actually costs is the absence of a decorrelated reader. From there: why a round with no rejections is treated as a failed proof, why detection is reliable exactly where generation is not, what to do when the reviewer you asked for is simply unavailable and how to measure that without fooling yourself, and the discipline that separates a useful reviewer from an expensive one — verifying the premise a finding rests on rather than only its conclusion.

   Lesson list (5)

   • 01
     Review Is the Spine, Not a Feature on Top~12 min · review, architecture, decorrelation, design
   • 02
     A Review That Rejects Nothing Failed~12 min · review, evidence, process, judgment
   • 03
     Detection Is Reliable Exactly Where Generation Is Not~12 min · review, cognition, modes, evidence
   • 04
     When the Reviewer Is Unavailable~13 min · review, reliability, detection, operations
   • 05
     Verify the Reviewer's Premise, Not Only Its Verdict~12 min · review, judgment, measurement, evidence

   ✓Quiz · 4 questions→

6. 06🧾Records at Work Time

   0/4 lessons

   A log written afterward is a reconstruction, and reconstructions are optimistic

   This track is the unglamorous half that makes everything before it true. A stage contract you cannot show was honored is a claim; a review round nobody recorded did not happen as far as anyone can tell later. So the workshop writes an entry at every stage boundary, at the moment it closes, carrying who ran it and on what — and the reason it does that rather than asking people to summarize afterward is a measurement: a corpus where forty-six of seventy-nine items looked untouched while the truth was three passes in a week. The track covers what makes a log have teeth rather than rot, why a single writer removes an entire class of distributed concurrency bug by construction, and what it leaves to be handled in one place instead, and how to keep an append-only ground truth alongside a queryable mirror without ever confusing which one is the record.

   Lesson list (4)

   • 01
     Unstamped Work Is Invisible Work~12 min · provenance, records, measurement, process
   • 02
     A Log Needs Teeth or It Rots~12 min · records, gates, design, process
   • 03
     One Writer, and One Place the Race Can Happen~12 min · concurrency, architecture, records, design
   • 04
     Ground Truth and Its Mirror~12 min · storage, architecture, records, design

   ✓Quiz · 4 questions→

7. 07🔒Serialize the Landing

   0/5 lessons

   One motion under one lease, on a base you chose deliberately

   The previous track removed the log race by construction. This one handles the other half, which does not have that answer available: several sessions all need to commit into one shared repository, and a commit is not something a single writer can absorb on their behalf. The track works through the shape that survived contact with a real parallel day — a lease that makes commit, push, stamp, and log entry a single indivisible motion; a commit built in a scratch index so a slow pre-commit hook cannot hold the race window open; a base chosen deliberately by fetching first and refusing to guess when history has diverged; and the rule everyone tries to break exactly once, which is that you never hand-stage a file somebody else is editing.

   Lesson list (5)

   • 01
     The Conflict Class You Cannot Remove~11 min · concurrency, version-control, architecture, design
   • 02
     One Motion Under One Lease~12 min · concurrency, leases, records, design
   • 03
     Build the Commit in a Scratch Index~13 min · version-control, plumbing, concurrency, implementation
   • 04
     Choose Your Base Before You Build~12 min · version-control, convergence, operations, design
   • 05
     Never Hand-Stage a File Somebody Else Is Editing~12 min · version-control, collaboration, failure-modes, policy

   ✓Quiz · 4 questions→

8. 08📐Measure, Never Rank

   0/5 lessons

   The instrument that narrows a corpus, and the invariant that keeps it from ranking one

   The workshop holds a hard line: no quality scores, anywhere. Cards, logs, and queues show lifecycle facts; repair is triggered by a reader, never by a number. This final track is what that line looks like when you take measurement seriously anyway. It opens with a full day of failed proxies, recorded so nobody rebuilds them. It works through the one measure that did land, and the three cautions that keep it a threshold rather than a ranking — because within the flagged set it carries no ordering information at all. It covers why a seeded human reader is a different instrument from a scanner and therefore does not violate the invariant. It ends on the mistake that generalizes furthest: two vantages running the same method are one instrument counted twice, which shipped six live links on a worklist of dead ones. Then the quest closes by handing you the design questions rather than the answers.

   Lesson list (5)

   • 01
     Keep the Table of Things That Did Not Work~12 min · measurement, proxies, evidence, process
   • 02
     A Threshold, Not a Ranking~13 min · measurement, statistics, judgment, instruments
   • 03
     A Seeded Reader Is Not a Scanner~12 min · measurement, sampling, invariants, judgment
   • 04
     Two Vantages Running One Method Are One Instrument~13 min · measurement, evidence, independence, verification
   • 05
     Build Your Own Workshop~12 min · synthesis, design, process, closing

   ✓Quiz · 4 questions→