Evaluation — SELIS V2 proposal idea
- Evaluated: 15 November 2026.
- Subject: a product idea ("v2"), not a written
draft. No price, duration, work packages or forms exist. v2 was agreed
in a grilling session on 15 November 2026; Ivaylo has not confirmed it
and it is not in
decisions-log.md. - Method:
evaluator-rubric.mdapplied to the idea, checked againstcompliance-matrix.md,pecs-playbook.md,type-a-activities.mdandresearch/findings/. Assumptions: competent writing, full administrative compliance, and the facts of 15 November (no letter, one CV, no prototype). - Evaluator: AI (Claude, read-only). Not re-checked by a person. Marks are ranges and judgements on the rubric's internal scale, not an ESA result. "J" marks a judgement of the reviewer; "W" numbers are the reviewer's ranking of weaknesses, W1 the most damaging.
- Like-for-like limit: v1
(
eval/selis-v1-proposal.md) marked a written draft with forms and costs; v2 marks an idea, so C3, C4 and every v1 weakness about draft execution are not comparable.
Proposal Idea
What v2 proposes, limited to the points that affect a criterion. The criterion in brackets is assigned by this file, with the logic of v1.
Concept. Mission-planning software for early robotic missions at the lunar South Pole, on sites modelled from public mission data. The name SELIS is kept for now. The framing is long-duration surface missions. v2 uses no "digital twin", no "first" and no permanent base.
Stated objective and maturity. Type A scope is power analysis only: illumination, darkness and energy yield for solar assets 1 to 5 m above the surface, at one site. Start TRL, end TRL, price, duration and work packages: [TBD]. (C1, C2, C3)
Main features
| Feature | What v2 proposes | Affects |
|---|---|---|
| Published terrain | NASA 5 m LOLA site DEMs with 100 clones. JAXA 1 m and other finer published terrain compared with the LOLA baseline and tagged as inferred. No own DEM production. | C1 |
| Site | Connecting Ridge; fallback Malapert Massif. | C1, C2 |
| Illumination and darkness with a stated confidence | The published NASA method (Barker et al. 2021) on the 100 clones. | C1, C2 |
| Check against measured data | A test of whether the stated confidence holds against dated orbital (LROC NAC) images. | C1 |
| Height sweep | A sweep over heights around a user-set height, within 1 to 5 m. | C1, C2 |
| Energy yield | Yield for solar assets. Panel-model depth [TBD] until the interview with the Lithuanian company. | C1, C2 |
| Slope constraint | Published NASA slope maps, reused as a placement constraint. | C1 |
| Data-class tagging | Measured: recorded by an instrument at that place. Inferred: estimated where no instrument measured. Derived: computed by a published physical or geometric model, recording the classes of its inputs. Synthetic: not used in Type A. | C1, C2 |
| Machine learning | None in Type A; no AI specialist. | C1, C3 |
| Next phase (out) | Landing-site assessment, Earth visibility, a 3D viewer. | C2 |
Uniqueness claim. Luminous applies a published NASA method, tests whether its stated confidence holds against dated orbital images, and delivers the result to a planning user. No "new method" claim. (C1, C2)
Users and market. First: European solar power developers, with one lead (a Lithuanian micro company, name [TBD], requirements not validated). Second: private lunar mission companies. Path beyond TRL 3: Argonaut Mission 1, MAGPIE and LUVMI-M. Revenue figures: [TBD]. (C2)
1. Marks
| Criterion | Weight | Mark (range, J) | Weighted (range × weight) |
|---|---|---|---|
| C1 Technical quality and team | 40% | 55–68 | 22.0–27.2 |
| C2 Prospects, programmatic fit, maturity | 30% | 48–62 | 14.4–18.6 |
| C3 Management, cost, planning | 25% | 55–68 | 13.75–17.0 |
| C4 Administrative, contract, IPR | 5% | 75–90 | 3.75–4.5 |
| Overall | 54–67 |
- Result: borderline. The range straddles the pass mark of 55 and is under the working target of 70 (decision D15).
- C2 is the weakest criterion; its range starts under the D15 floor of 55 for each criterion.
- C4 "depends on execution, not on the shape"; its range rests on the compliance assumption.
- Reviewer's verdict (J): v2 "trades an unfundable, duplicative product vision for a bounded, checkable experiment. It is still not a safe proposal."
2. C1 — Technical quality and team (55–68)
Weaknesses, most damaging first. W3 to W6 are placed here by the reviewer; W7 and the last bullet by this file.
- W3. The team shows no domain experience. The CV on file shows cloud and platform engineering; Martin's CV is pending and his field is lava-tube habitats. Illumination modelling, image registration and photovoltaic skills are not evidenced (compliance row C9).
- W4. It risks reading as a NASA re-run on ESA's own 2012
site. Barker ran 100 clones at Connecting Ridge at 1 m and 5 m,
so that function is an input, not a critical function
(
type-a-activities.mdsection 7). Only the calibration against images is clearly new, and the note asks for two to four critical functions. - W5. The calibration test may not be feasible, and it tests the wrong height (J). Orbital images show the lit state of the surface, not of a point 1 to 5 m above it. Usable lit or shadowed states per cell are uncounted, and the registration error (about 5 m RMS) equals the grid size (B2 section 7).
- W6. Method error may exceed terrain error (J). Two 2012 tools differed by about 44.7 days RMS in longest lit period at Connecting Ridge, because of the horizon calculation. Luminous must re-implement the engine; a confidence built from clones alone omits this error (A3 findings 3 and 4).
- W7. Start TRL evidence weakens. Without own DEM production the Ames Stereo Pipeline work is irrelevant. Only the Grasshopper photovoltaic demos relate: TRL 1 is evidenced, 2 is arguable only if the demo is documented (J; compliance row K2).
- Other gaps. The name SELIS no longer fits.
glossary.mdconflicts with the new class definitions and with the new use of "confidence". PGDA pages carry no licence text (compliance row K6).
Strengths:
- One goal, one use case, end to end, at one site: a cleaner fit to TRL 3 if worded as a proof-of-concept prototype.
- Data exists at the site: 322 co-registered images, clones and a published percentile table (B2). No AI specialist is needed.
3. C2 — Prospects, programmatic fit and maturity (48–62)
Weaknesses, most damaging first. All three are placed here by the reviewer.
- W1. The lead user and the value do not match. The clone effect is large at 1 m and "almost negligible" at 5 m (A3 finding 2). Published power designs use towers of 10 m or more; assets near 2 m are lander and rover arrays, which v2 ranks second. The customer's panel height is unknown.
- W2. No evidenced demand. One unnamed lead, no interview, no letter. No user was found who says free tools are insufficient or who would pay (C finding 10); NASA offers on-request illumination at a chosen height (compliance rows I5 and K8).
- W8. The path beyond TRL 3 is asserted. No Argonaut or MAGPIE site was found, and none of those teams was contacted. Benefit to Bulgaria is unargued.
Strengths:
- Duplication risk is low to moderate (J): no XR, no rover traverses; v2 complements Spaceship FR. Residual risk: the "mission-planning software" label.
- The claim is honest and sourced: no "first", no "new method", no "permanent base".
4. C3 — Management, organisation, cost and planning (55–68)
No price, hours, schedule or work packages exist, so the reviewer lists risks, not weaknesses:
- Price. A price at the ceiling would look padded (playbook section 6). Build it from the work.
- Compute cost is unknown. Barker used a NASA computing centre (A3 unknowns).
- Substance (J). Enough for about 9 to 12 months for two people only with the engine re-implementation, a cross-check against Barker's table, the calibration, the error sources outside the clones and the yield model. Thin if it is only a clone re-run.
Strength:
- Scope is in proportion to the budget.
5. C4 — Administrative, contract and IPR compliance (75–90)
No weakness listed. The range assumes full compliance and is the same for any shape of the idea. Open compliance rows named by the reviewer: K2, K4, K6 and C9.
6. Likely negotiation questions
Not produced in this evaluation.
7. Top fixes, by expected gain in the weighted mark
The reviewer gave three:
- User and height. Interview the Lithuanian company and get a signed letter with asset height and needed quantities; seek a second letter from a lander or rover user. Re-centre on lander and rover arrays if the height is 5 m or more.
- Critical functions. Define two to four with pass or fail numbers; state how a surface check transfers to 1 to 5 m; count the usable lit and shadowed states in the 322 images.
- Evidence for start TRL and team. Document the photovoltaic demo; run one nominal case against one NAC image and Barker's table; add Martin's CV and roles.
Necessary before submission:
- Record v2 in
decisions-log.md, then updateCLAUDE.mdandglossary.md. - Cite Barker 2021, De Rosa 2012 and NASA's on-request service in section 1.4.1.
- Have a person re-open the sources. Close compliance rows K2, K4, K6 and C9.
- Schedule (J): the concept note is late; the customer interview and Martin's CV are on the critical path; do not compress the evaluator rounds.
8. Comparing later evaluations
Save each later evaluation in eval/ with the same
headings and the same marks table. Compare the four criterion marks and
the overall mark first, then check which weaknesses are closed, which
remain and which are new. Sections 9 to 13 do this for v1 and v2.
9. v1 to v2 at a glance
| Criterion | v1 draft | v2 idea (J) | Comparable |
|---|---|---|---|
| C1 Technical quality and team | 45 | 55–68 | In part |
| C2 Prospects, programmatic fit, maturity | 35 | 48–62 | In part |
| C3 Management, cost, planning | 52 | 55–68 | No |
| C4 Administrative, contract, IPR | 40 | 75–90 | No |
| Overall | 43.5 | 54–67 | Not like for like |
- Raw gain: about 10 to 24 weighted points (arithmetic on the rows).
- Resting on the writing and compliance assumption: 2.5 to 6.5 of those points come from C3 and C4, which no draft supports yet.
- Like-for-like gain: roughly 10 to 14 points (J).
The same reviewer marked the original idea of
CLAUDE.mdsection 2 at 42–53 under the same assumptions; the v1 draft mark of 43.5 lies inside that range.
10. The most important changes
| Change from v1 to v2 | Effect | Criterion | Reason |
|---|---|---|---|
| Scope: nine features cut to power analysis at one site | Improves | C1, C3 | One use case end to end; scope in proportion to the budget |
| VR/AR, robotics simulation, printed model, vision model and terrestrial example dropped; 3D viewer moved to the next phase | Improves | C2, C1 | No overlap with ADTLAS; no AI or 3D skills needed |
| End point: installable integrated prototype with user trials becomes a proof of concept of critical functions | Improves | C1 | Fits TRL 3 if "delivers to a user" means a customer review |
| Claim: "integration in one environment" becomes an applied NASA method with its confidence tested | Mixed | C1, C2 | Honest and sourced; easier to dismiss as "NASA already did this" (W4) |
| Terrain: own multi-DEM site becomes published terrain at Connecting Ridge | Mixed | C1 | Data exists at the site; the stereo DEM work stops being start-TRL evidence (W7) |
| New: test of the confidence against dated orbital images | Mixed | C1 | The only clearly new element; may be infeasible (W5, W6) |
| New: assets at 1 to 5 m, with power developers ranked first | Degrades | C2 | The method matters at 1 m, not at 5 m; the lead's height is unknown (W1) |
| Users: a 30-organisation survey becomes one lead and two ranked segments | Mixed | C2 | A lead exists; no letter, and the market is narrower (W2) |
| Missions: none named becomes Argonaut Mission 1, MAGPIE, LUVMI-M | Mixed | C2 | Stronger mission link; the path is asserted (W8) |
| Four data classes: a passing mention becomes a defined tagging rule | Mixed | C1, C2 | All four are defined; they are "background and not the claim" |
11. What improves, what degrades, what is unchanged
v2 improves:
- Duplication: the VR/AR priority that v1 was marked down for is gone.
- Scope: v1's excess scope is cut, and the end point no longer reads above TRL 3.
- State of the art: precedent is named and the claim is honest.
- Missions and a lead user are named, where v1 had none.
- Requirements can be quantified against a measured reference (dated images, Barker's table).
v2 degrades, or adds as a new risk:
- Start-TRL evidence is thinner (W7).
- The lead user may not need the result at the heights power designs use (W1).
- It can read as a NASA re-run (W4).
- The calibration test may not be feasible, and method error may exceed terrain error (W5, W6).
- The market is narrower and tied to one unvalidated lead.
- The name,
glossary.mdand the four-class differentiator no longer fit the idea.
Unchanged, or not assessable until a draft exists:
- Team: no evidence of domain experience; Martin's CV pending.
- Demand: no named customer, no letter, no evidence of willingness to pay.
- Benefit to Bulgaria is unargued.
- Sources were compiled by AI and not re-checked by a person.
- v1's C3 and C4 weaknesses (hardware, time shares, Gantt, work packages, cover letter, AI statement, placeholders) are not closed. They are not assessable until a v2 draft exists.
12. Open items
Both were proposed after the review. Proposed, not confirmed; effect on marks not evaluated.
- Q24. Widen the confidence into an error budget with three critical functions: (1) an error budget covering clones, the known 40% underestimate of clone-based uncertainties, the horizon-calculation error and the finer-terrain comparison; (2) a check of that budget against dated images at surface level, with a stated argument for the transfer to 1 to 5 m; (3) energy yield at height. Aimed at W4, W5 and W6.
- Q25. Centre the main use case on assets at about 1 to 2 m (arrays on landers, rovers and small deployable units). Keep the two user segments unranked until the customer interview. Re-centre on lander and rover teams if the Lithuanian company's panels sit at 5 m or above. Aimed at W1.
13. Notes on references
- W1: one of the two findings cited for "towers of 10 m or more" (A3 finding 8) also records a design with 2 m towers (Bryant 2009), which the reviewer omits.
- W1: De Rosa et al. 2012 section 2.2.1 gives a lander's solar-array
height of about 2 m. "Rover arrays" is not in it. The paper is in
research/assets/topic-A/12_1208.5587v1.pdf; its section 2.4 confirms the W6 figures. - W2: the Honda and Astrobotic in-house statement is in
C-users-and-needs.mdfinding 2, not in A3 finding 9. - The remaining minor mismatches of finding numbers were checked and do not change any statement.