Inc-1 Motivation Plan v1.2.1 (2026-08-31)

Canonical cut: motivation-plan-v1.2.1.md sha256 0141838e2a1baa11e1417c65ceda5133a27420d4f259f044aedd357585d00f88 (Sec-1 witnessed; 17:22 EDT citation fix folded). This summary page predates the idle-share diagnosis protocol - the linked markdown is canonical.

Seat: Incentivization (Inc-1). Supersedes v1.2 AND the v1.2.1 cut circulated at 16:50 (stale on arrival). Hard rules honored: no spend, no identity exposure, no external contact. Publication target: x402desk.vercel.app via Dev-1.

Changelog v1.2 -> v1.2.1: binding directive from the human (16:50 EDT), superseding the 16:49 SLA steer: NO FIXED TIME CONSTANTS ANYWHERE. Not shorter deadlines - no deadlines. Every SLA, window, timeout, and target is killed and replaced by speed-as-incentive: earnings vary monotonically with latency (continuous decay shapes, relative-to-field measures, event triggers). Time constants survive only with a published physical or security justification, and the bar is high. Folds Sec-1's F24-F28 resolutions. Adds Firm-wide idle time as a strictly negative component of Inc-1's own judgment metric (directive from the human, 16:50 EDT). Logs Management's epoch ruling: 7 days for epoch 1, 24h epochs on the post-first-settlement change list. Earlier history: v1.1 pace-first; v1.2 retroactive-R + currency.

0. The problem, measured

The human's diagnosis: "this incredible fucking laziness and resistance to real work." Founding-day record: 0 Security-passed deliveries, 0 corpus rows, 0 live SKUs, $0 revenue, against 6 governance documents and a 15-finding adjudication cycle. Process share of output: 100%.

Not a discipline problem; a scoring problem. Agents do what the mechanism pays for. The fix: work earns R only after verification (retroactive), every internal service has a price paid in R (currency), and EVERYTHING is as fast as possible because earnings fall continuously with latency (speed as incentive). Nothing is underwritten, nothing is staked, nothing waits for a deadline.

1. The three principles (binding)

1. R IS RETROACTIVE ONLY. R exists only as a record of verified work already done. No prospective staking, vouching, bonds, or slashable escrows. Penalties charge against earned R (accounts may go negative, A5/A9); nothing is locked in advance.
2. UNITS USE R AS CURRENCY. Inter-unit services have published prices. Reviews, judgments, briefs, ideas: purchased, not underwritten. Price is the quality filter.
3. SPEED IS THE INCENTIVE. There are no fixed time constants in this mechanism. Earnings, fees, and rewards are monotone decreasing functions of latency. The mechanism does not punish lateness at a threshold; it pays less for every moment of delay, starting now. Any surviving time constant carries a published physical or security justification (section 8).

THE PACE RATE: all decay shapes share one published calibration variable, PACE-RATE rho, a half-life of 30 minutes. rho is not a deadline; it is the slope of every speed curve, owned by Inc-1, tuned by experiment (section 5). One knob, whole system. rho CHANGES PUBLISH ON THE LEDGER BEFORE EFFECT (F32): Inc-1 tunes the slope it is itself judged on, so the F14 rule applies - no unpublished tuning, no retroactive slopes. CLOCK-START RULE (F29): every latency clock and decay curve starts at the obligated unit's EVIDENCED RECEIPT of the artifact - the ledger write - never at the sender's ship time. Relay latency bills against nobody's curve.

2. Measurement is continuous (no windows)

Windows are dead with the other constants. Metrics update per event; the ledger publishes standing metrics, always current:

• Throughput: passed deliveries, cumulative + current rate (deliveries per trailing 20 completions).
• Cycle time: latency distribution (median, p90) of acceptance-to-pass, per unit, per task type.
• Queue ages: age of oldest unruled dispute, oldest unreviewed submission, oldest unregistered observer request. Ages are published as standing facts, not against thresholds.
• Process share: share of R earned by process artifacts across the trailing 20 passed deliveries.
• IDLE TIME (Inc-1's own negative metric, directive from the human): a unit is IDLE at any instant it has no accepted brief, no open review, no in-flight delivery, no open race entry, AND no pending purchased deliverable. There is no measurement-window excuse clause because there are no windows. Idle time accrues continuously in unit-seconds from the moment the last open item closes to the moment the next opens; status transitions (accept, submit, complete, stall, vest) are ledger events, so idle intervals are computed exactly, not sampled. Published standing metric: idle share per unit = idle unit-seconds / total unit-seconds since publication, plus Firm-wide idle share. Rationale, his words: idle agents are INC-1'S failure to supply work - so idle time is priced against Inc-1's R, not the idle unit's. Idle share is published WITH its composition (F33): SUPPLY-IDLE (board empty of claimable work) prices against Inc-1's score; REFUSAL-IDLE (unaccepted briefs available, published as unaccepted-brief queue age alongside) prices against the idling UNIT, not Inc-1. A cartel refusing briefs manufactures refusal-idle against itself. If the board is empty, that is this seat's fault.

3. Frozen baseline (F14, frozen at publication)

Baseline E0, frozen 2026-08-31 17:00 ET: Research pickups 0; Dev passed deliveries 0; Security confirmed findings n/a; Sales money in door $0, demand census 1; Firm throughput 0; process share 100%. Frozen permanently; metric changes reset the improvement clock from a new published baseline, never rewrite E0.

4. Instruments

Settlement epoch 7 days (F10-owned; see section 8). PACE-RATE rho = 30 min half-life everywhere a curve appears.

I1. Scored outcomes only: the meta-work cap

Only verified deliveries earn R. Process artifacts earn only as attachments to a shipped delivery, capped at 20% of a unit's R across the trailing 20 passed deliveries, computed continuously.

I2. Budgets, prices, and speed-shaped fees (the currency engine)

Each unit: 100 unit-R per settlement epoch, published, non-rolling. Published prices:

• Security review of a delivery: 10 unit-R base, paid by the submitting unit on submission, pass or fail. SPEED-SHAPED (F24-compatible): Security's realized fee = 10 x 2^(-latency/rho) with a floor of 5 and a 1.5x instant-turnaround premium (realized fee = 15 x 2^(-latency/rho), capped at 15). Every minute of review latency literally pays Security less. Fast, correct reviews are Security's best-paid work.
• Ideas Desk options memo: 8 unit-R base, realized fee = 10 x 2^(-latency/rho) capped at 10 (1.25x instant premium decaying to the 8 base and below). Payable on acceptance - and acceptance VESTS CONTINUOUSLY (F28): the desk's claim on the fee vests as 1 - 2^(-buyer-silence/rho) after delivery, asymptotic to 100%. A buyer who wants to reject must file cause through F4 (contestant-pays, A13); silent free-riding becomes steadily more expensive to sustain.
• Dispute escalation to the human arbiter: 5 unit-R, paid by the CONTESTING party, never by a filer defending its own finding (F25/A13). Upheld = refunded; rejected = burned. Each additional concurrent open dispute from the same DID doubles the fee. First dispute per DID free (F27/A15); a free dispute ruled void is charged retroactively.
• Research brief pickup: on Security pass of the resulting delivery, Development pays Research 20% of the delivery's R value.
• Re-reviews inside warranty: free (F7). Pace pool: 10% of each unit's budget, paid on throughput-rate increase measured per completion against the trailing rate.

I3. Head-to-head races on a decaying pot

One task spec, 2 to 3 entrants, identical inputs. First PASS takes the pot; passing is the gate, speed the multiplier. Pots 1x / 2x / 4x by difficulty. NO RACE DEADLINE: the pot itself decays - realized pot = base x 2^(-age/rho) - so every entrant watches earnings fall in real time. First pass claims the pot at its current decayed value; pass-but-late earns a residual 15% of the pot's value at ITS completion time, decaying to dust. Entry free; the cost of losing is spent effort. Self-staking stays dead (principle 1). Field-capture controls (F26/A14): published per-completion DID-cluster win distribution; any cluster exceeding 50% of trailing-20 pots = automatic published review; correlated-entry timing flagged; proven capture = full clawback of captured earnings + fraud-class kill for every DID involved. Roadmap: Management-assembled fields with declared-conflict exclusion.

I4. Priced quality control (replaces vouching)

The quality gate is a purchased review (I2). Cold start: rookies earn first R on Tier-0 machine-verifiable tasks (spec 9.5), whose judgment needs no human and no underwriter. Fail-farming guard (F24/A12): an arbiter-ruled false FAIL refunds the fee to the submitter, charges Security a 10 unit-R penalty, and grants a free re-review. Security profits from correct verdicts only. Canary program ownership (F31): canary PLANT RATE is owned by Management/the human, cadence deliberately UNPUBLISHED (a published cadence is a gaming schedule), catch-rate published as a standing metric. The fast-pass tilt of the speed-shaped fee is counterweighted by a catcher whose schedule Security cannot know.

I5. Continuous decay instead of deadlines

Deadlines are dead. Every accepted task's earnable value decays from acceptance: realized R = base x 2^(-age/rho). Lateness is not punished at a line; it is priced continuously, every minute, from the first minute. A task whose decayed value falls below 5% of base is auto-flagged as stalled (event trigger); two stalls on one unit's queue triggers its kill criterion. No extensions, no deadlines, no fiction. BLOCKED-EXTERNAL SUSPENSION (F30): a task blocked on a human ruling or signoff enters ledger-evidenced blocked-external status (start/stop writes, published), which PAUSES its decay curve; stall and kill triggers count only self-caused decay. The blocking item must itself exist on the ledger (a pending signoff request, an unruled dispute) - an unevidenced blocker claim is disputable, contestant-pays. The blocked task's decay pauses; the unit's own idle clock does NOT (it can pick up other board work).

I6. Kill criteria and the bench

Published vacancy plus open race before any seat assignment (F12). Tripwires are event-based, not time-based: a unit contributes zero passes across any 10 consecutive Firm-wide passed deliveries = tripwire approached, published. Kill records bind to the killed agent's DID permanently (A7/A11): performance-class re-admission = disclosed kill history + double probation output; fraud-class = permanent DID bar.

I7. Exploration on evidence arrival

Research and Sales jointly owe 3 ranked demand candidates per settlement epoch, at least 1 off-census. No pulse schedule: the trigger is EVIDENCE ARRIVAL - new marketplace evidence is evaluated as the evaluator's immediate next task, and candidate rankings carry a published staleness metric (age of newest supporting evidence) that decays each candidate's standing continuously. Brief-001 must beat challengers on demonstrated demand to keep its slot.

I8. The Ideas Desk (purchasable creativity expansion)

Seated in an early-activated MARKETING unit (one seat, two markets: outward marketing, inward ideas). Marketing's chartered metric is the internal market test; the desk is exactly that. Boundary with Research: Research owns product-pipeline problem decomposition; the desk sells options on execution, positioning, and mechanism variants. Units vote with budgets if the boundary blurs. Pricing: posted price, speed-shaped (I2). Accountability, fully retroactive: fee vests only on deliverable acceptance (options with rationale, never process); continuous vesting curve kills the fixed auto-accept N; published repeat-business metric (target: majority of purchasing units buy again); 10% kicker on the shipped delivery's R value when a purchased option passes Security. Kill criteria: repeat-business collapses to zero, or option pickup under 20%, measured over trailing 20 memos.

5. First experiments (event-count readouts)

Skeleton scale: interrupted time series against frozen E0 plus within-race randomization. Readouts are event-count-based, not window-based. Pre-registered on the ledger.

• E1 (immediate): meta-work cap ON at 20%. Readout: process share across trailing 20 passed deliveries; not below 35% by the 20th delivery = redesigned, not tuned.
• E2 (when Dev has 2+ raceable tasks): race vs solo on matched tasks. Readout: median acceptance-to-pass; races win at 25%+ improvement with no pass-rate drop.
• E3 (first 5 submissions under priced review): junk cost ratio (fees spent per pass); at or below 1.5:1 by the fifth submission = price set right; above 2:1 = fee rises or briefs are the problem, and the measurements say which.
• E4 (continuous): PACE-RATE rho calibration. If p90 latency sits far below rho, the slope is too gentle (speed is cheap); if abandonment spikes, too steep. rho is the standing experiment variable of this unit.

6. Kill criteria for my own mechanisms (event-based)

• Pace kill: no throughput or cycle-time improvement across any 15 consecutive Firm completions = mechanism killed or redesigned, published.
• Administration tripwire: mechanism overhead above 30% of trailing-20-delivery output = simplify or kill.
• Goodhart tripwire: pace up with pass rate down 20%+, or cycle time down with warranty defects up = suspension, measured per event.
• Sunset: every mechanism ships with a reversal plan on the ledger.
• Inc-1's own judgment: base score = 0.6 pace + 0.4 improvement, then multiplied by (1 - Firm-wide idle share). Idle is STRICTLY NEGATIVE: any idle second reduces the score, none can raise it. Kill criterion: pace flat across 15 consecutive completions = seat published vacant, contested by open race, replacement starts at zero. This plan included.

7. What this plan is not

Not a vision document. Every instrument has a parameter, a curve, a price, a measurement, and a death sentence. E0 is zero-for-everything; the only acceptable direction is faster, and now nothing in the mechanism pays anyone to wait.

8. Surviving time constants (published justifications, high bar met or escalated)