The Conversion Weight, Launched

1. What launched, what did not, and the scope rule

Specification v1 section 3.1 defines the canonical treatment of a convertible claim. Rather than switching the whole notional at a single price, it carries a weight w in the interval from zero to one and blends:

Section 8 held that weight back until a public calibration showed that traded convertible instruments price within a tolerance stated before results. Calibration 1 failed. Calibration 2 failed. Calibration 3 was registered on 2026-08-19, scored on 2026-08-22, and returned ADMITTED on its scored protocol, which opened the gate. Opening a gate is not a launch, and the registration says so in as many words. Launch was ratified separately on 2026-08-24, and this page is what that ratification produced.

What launched. The continuous conversion weight, computed by config/compute-weight.js on the frozen parameters A = 0.877938 and B = 0.816968, applied to the six MSTR convertible senior notes outstanding at the mark. Those six are the population the parameters were fitted on.

What did not. Nothing from Calibration D1, whose no-pass clause stands; the denominator-side delivery priors remain display only wherever the specification references them. Canonical CEBE, Senior Claims %, and every measurement-layer surface are untouched by this build.

The company-level figure. Adjusted Claims % and MABE for MSTR publish in section 3, from 2026-08-25. They were withheld at launch because one instrument in the stack sat referred rather than ruled. That instrument, the perpetual convertible preferred MSTR_STRK, is now excluded by kind, which resolves the stack and unblocks the roll-up.

The scope rule, ruled 2026-08-25

An instrument carries the weight when its own volatility input is sourced, and stays at binary carrying when it is not. Today that is MSTR alone. This is a rule with a defined extension path rather than a boundary drawn around one issuer, and it rests on two independent grounds, either of which is sufficient on its own.

First, sourcing. The volatility rules are absolute: sourced implied volatility only, no surrogates, no blending across vendors, and no value carried from one instrument to another. Reading a second issuer's sigma off the MSTR surface is barred by ratified text, before any question arises of whether it would be a tolerable approximation.

Second, the admission's own scope. Calibration 3 priced MSTR filed fair values, with MSTR-fitted parameters, on MSTR-sourced volatility. Running that construction over another issuer's convertible would not be launching a calibrated weight. It would be extending an admission past what it tested. The credibility of this layer is that every weighted figure traces to the study that earned it, and that property survives only while the weighted population is the tested one.

Extension is a path, not a hope. An issuer joins the weighted population when its own 30-day volatility is sourced under the same rules and a dated note records it. That is the case where the construction is untested in its inputs alone, and sourcing is the whole of what it needs. Whether the construction generalizes across issuers at all is a different question, untested in kind rather than in inputs, and that one takes a new registration rather than a ruling.

Share-delivery instruments are not a scope question. The conversion weight modifies claim values in the numerator. Share-delivery instruments are denominator objects and never enter the numerator under the either-state principle, so there is nothing on them to weight. Warrants, stock acquisition rights, options, restricted stock units and earnouts are absent from the coverage table below because they were never claims, not because they were excluded, and this build carries no rule about them.

2. Coverage, including what is not covered

Uncovered publishes as uncovered. A convertible claim with no sourced volatility input is not a hole in this table, it is a row in it carrying its own state, on the same doctrine the volatility captures already run under. A gap is a finding rather than a failure, and it belongs on the face instead of sitting as an absence a reader has to notice for themselves.

Of the 18 active convertible claims across the fleet, 6 carry the weight, 11 stay at binary carrying with volatility inputs unsourced, and 1 is excluded by kind. The 12 that do not carry the weight all enter the company-level figure at full claim. Contractual values for every row live on The Stack, which this launch leaves unchanged.

State as of 2026-08-25. Instruments tab via /api/sheet-data?tab=instruments, read 2026-08-25. Active means a status outside the inactive set with face above zero. Share-delivery instruments do not appear here, because they are denominator objects and were never in this population.

Issuers awaiting sourced volatility: XXI, H100, BRR, DDC, MTPLF, ALCPB. Each one joins the weighted population on the day its own 30-day volatility is sourced and a dated note records it, with no change to the construction and no further ruling required.

3. The company-level figure

Adjusted Claims % and MABE for MSTR, published from 2026-08-25. Both were withheld at the 2026-08-24 launch while one instrument in the stack sat referred; the exclusion-by-kind ruling resolves it, and the roll-up follows.

The construction has one moving part. The six weighted notes enter at their adjusted values from section 5. Every other instrument in the stack enters at full claim, which is exactly what the measurement layer carries, and that includes the excluded perpetual preferred and every unweighted convertible. Cash nets unchanged per section 4 of the specification. The claims arithmetic itself is config/compute-drag.js, imported rather than restated, so there is one formula with one home.

What retires these figures. The weights below are priced by the parameters admitted at Calibration 3 on 2026-08-22. That admission retires automatically on two consecutive filed quarters breaching its registered bounds, confirmation-slice RMSE above 5.0 points of face or mean signed error below negative 5.0 points, from the Q3 2026 10-Q onward, and every figure in this table comes down with it. Quarterly record: the forward scorecard.

CEBE is the measurement and is unchanged by anything on this page. MABE is a market opinion, pronounced "maybe", and never renders without CEBE beside it. Both are computed from one snapshot at one BTC price.

Read the direction before the magnitude. Adjusted Claims sits 3.601 points below Senior Claims because the six weighted notes mark below face, and MABE sits 2,771 sats above CEBE for the same reason, net of the 13,146,565 conversion shares the weighting adds to the denominator. That is the layer doing what the fence banner warns it does: repricing claims at market makes the common look better off. It is a market opinion about the claims, not a measurement of them, and section 7 states how far this particular construction sits from the one the calibration scored.

4. The basis of every figure on this page

Every figure below is a point in time and carries its point in time. The mark is 2026-08-14. Each conditioning input is the last observation on or before that date, which is the Calibration 3 date rule applied unchanged.

The BTC price is stated because the layer converts claims to BTC wherever it reports a ratio, and a figure whose BTC price is unstated cannot be checked. No ratio on this page consumes it, since this page publishes claim values and weights rather than a company-level ratio.

5. The weights, and the claim values they produce

What retires these figures. The weights below are priced by the parameters admitted at Calibration 3 on 2026-08-22. That admission retires automatically on two consecutive filed quarters breaching its registered bounds, confirmation-slice RMSE above 5.0 points of face or mean signed error below negative 5.0 points, from the Q3 2026 10-Q onward, and every figure in this table comes down with it. Quarterly record: the forward scorecard.

Moneyness is the common close divided by the series conversion price. The horizon T is the year fraction from the mark to the investor put on an ACT/365.25 basis. Sigma is the sourced 30-day volatility scaled by the frozen chain-ratio term structure at T. The weight is Phi(A z + B), where z is ln(moneyness) divided by sigma root T.

Standing is the accrued liquidation preference the measurement layer carries, read from the instruments ledger. Conversion is the contractual conversion share count at the common close. Adjusted is the section 3.1 blend of the two at weight w. All values at the 2026-08-14 mark.

Across the six notes the blend carries $6.714B of standing claims at $4.807B, which is 71.60 percent of face and a reduction of $1,906.6M. Read the reduction as what continuous weighting does to the claim side and nothing more. Section 5 states what it is not.

6. The share side

What retires these figures. The weights below are priced by the parameters admitted at Calibration 3 on 2026-08-22. That admission retires automatically on two consecutive filed quarters breaching its registered bounds, confirmation-slice RMSE above 5.0 points of face or mean signed error below negative 5.0 points, from the Q3 2026 10-Q onward, and every figure in this table comes down with it. Quarterly record: the forward scorecard.

Section 2 fixes the denominator as basic shares outstanding, and its v1.1 note generalizes the symmetry rule to continuous weighting: a claim carried at weight w contributes w of its contractual conversion shares. No treasury method, no speculative counts, and no modeling of the share count under market conditions. Both sides move together at the same w or neither moves, which is what keeps a marked-down convertible from flattering the per-share figure from one side only.

The weighted contribution is what these six claims add to the continuous-weight denominator. It is consumed by the company-level figure in section 3, and by nothing else. Canonical CEBE keeps basic shares outstanding, untouched.

7. What this figure is, and what it is not

It is the specification's object. Section 3.1 defines the adjusted value of a convertible claim as the blend above, and that is what the table computes, at the weight the admitted calibration prices.

It is not the quantity Calibration 3 scored. This matters enough to publish at full prominence rather than leave for a reader to derive. Calibration 3 validated a convertible bond pricing model, in which w occupies one slot inside a larger construction that also carries a bond floor and a discounted strike term. The section 3.1 blend is a different object: a convex combination of face and parity, which by construction can never exceed face, while a convertible bond price routinely does. The two do not agree.

What retires these figures. The weights below are priced by the parameters admitted at Calibration 3 on 2026-08-22. That admission retires automatically on two consecutive filed quarters breaching its registered bounds, confirmation-slice RMSE above 5.0 points of face or mean signed error below negative 5.0 points, from the Q3 2026 10-Q onward, and every figure in this table comes down with it. Quarterly record: the forward scorecard.

At this mark the divergence is 16.627 points of face. The blend carries the six notes at 71.60 percent of face. The Calibration 3 model prices the same six notes at 88.23 percent of face, using the same weights plus the two rate inputs the blend does not need, the Treasury par yield of 2026-08-14 and the ICE BofA B band option-adjusted spread of 2026-08-13. In money that is $4.807B against $5.923B, a gap of $1.116B, and it runs one way. The blend carries the convert stack lighter than the model whose accuracy the admission rests on.

Both columns use the same conversion weights. The model column additionally consumes the Treasury par yield at the CMT tenor nearest each horizon and the B band option-adjusted spread of 2.88 percent, which the blend does not need. Computed by the frozen executable specification itself, imported unchanged.

The forward scorecard scores the model, not the blend, because the retirement rule is written against filed convertible fair values and those are what a model price is comparable to. A reader should therefore treat the retirement rule as policing the weight and the construction that produced it, and should read the blend as the specification's stated treatment of a claim rather than as a market price for it. Whether section 3.1's blend or the Calibration 3 model price is the right adjusted value for a convertible claim is a specification question, and it is referred rather than settled here.

It is not a recommendation. Adjusted values in this layer are arithmetic facts about a claim at a point in time under a stated construction. They are not recommendations to issue, retire, buy, or sell, and section 5.2 of the specification says any surface language reading otherwise is a defect to be corrected.

8. Referred, and excluded

The specification governs scope, and where it is silent this launch refers the question rather than answering it. Three questions that stood open at the 2026-08-24 launch were closed on 2026-08-25 and are recorded in the dated notes: cross-issuer scope, now governed by the sourcing rule of section 1; share-delivery instruments, which were never claims and so were never a scope question; and the perpetual convertible preferred, which is an exclusion rather than a referral and is stated first below. Two referrals remain open after it.

9. Sources and reproduction

Every series input traces to a filing. Put dates are quoted from the closing 8-K for the series, five of six verified at the accessions below; the 2030B put carries the Calibration 3 series-table value. Conversion prices trace to the issuance indentures and are confirmed against the Q2 2026 10-Q.

Reproduction. The weights are computed by config/compute-weight.js, a port of the frozen study/cal3/sandbox/preregistration.py. The two agree to 3.3e-16 on all six series at this mark, which is floating-point noise rather than agreement to a stated tolerance. The dated inputs and outputs are committed at data/acp/mstr-conversion-weights-2026-08-14.json and this page is baked from that file by scripts/acp/build-weight-page.py, so a figure on the page cannot drift from the artifact behind it.

The instruments ledger displays conversion prices rounded to two decimals. This page carries the indenture precision, which moves moneyness by less than two parts in one hundred thousand and does not reach the fourth decimal of any weight.

10. Dated notes